Smart Precision Weeding in Agriculture Using 5IR Technologies

Abstract

The rise of smart precision weeding driven by Fifth Industrial Revolution (5IR) technologies symbolizes a quantum leap in sustainable agriculture. The modern weeding systems are becoming promisingly efficient, intelligently autonomous, and environmentally responsible by introducing artificial intelligence (AI), robotics, Internet of Things (IoT), 5G connectivity, and edge computing technologies. This review discusses a comprehensive analysis of the traditional and contemporary weeding techniques, thereby focusing on the technological innovations paving way for the smart systems. Primarily, this work investigates the application of 5IR technologies in weed detection and decision-making with particular emphasis on the role of the aspects such as AI-driven models, drone-robot integration, GPS-guided practices, and intelligent sensor networks. Additionally, the work outlines key commercial solutions, sustainability metrics, data-driven decision support systems, and blockchain traceable practices. The prominent challenges in the context of global agricultural equity pertaining to cost, scalability, policy alignment, and adoption barriers in accordance to the low-resource environments are discussed in this study. The paper concludes with strategic recommendations and future research directions, highlighting the potential of 5IR technologies on the smart precision weeding.

1. Introduction

1.1. Background

The global food system faces unprecedented challenges driven by population growth, climate change, geopolitical conflicts, and shifting consumer behaviors, demands, and expectations of the products they consume. By 2050, the world population is expected to reach 9.7 billion people, which requires a 70% increase in food production [1]. Even with the current global population of 8.2 billion, approximately 25–30% people face food insecurity and do not have access to sufficient nutrition as shown in Figure 1 [2,3]. However, to meet the increasing demand, agricultural expansion alone is unsustainable, as it leads to deforestation, loss of biodiversity, and resource depletion.

One of the key obstacles to achieving higher agricultural productivity is weed interference. Weeds compete with crops for essential resources such as water, nutrients, and sunlight, leading to significant production loss [5,6]. Estimates suggest that weed infestations can reduce global crop yields by up to 34% without any effective weed management. Numerically, yield losses may range from 50% to 70% in the United States and Canada’s soybean and corn production [7], 50% in Bhutan and 80% in India rice production annually depending on the severity of weed infestation [8].

The severity of weed–crop competition depends on weed density, emergence timing, and resource consumption. Weeds can also release allelochemicals, which are phytotoxic compounds that hinder or promote crop germination and growth [6,9]. Some notorious allelopathic weeds include nut sedges, crabgrass, Canada thistle, and spotted knapweed [6]. However, this process can go both ways, as shown in Figure 2, i.e., the growth of some weeds may be affected by certain types of cultivated crops such as winter rye, mustard, and radishes [6].

Weeds influence soil microbial communities by harboring specific bacteria and fungi that can either benefit or harm crops. Plus, in pastures, weeds reduce available forage, lower stocking rates, and diminish annual income [6]. They also affect livestock health through poisoning or physical injuries, and may contaminate farm products, like milk, by tainting them [6].

1.2. Problem Statement

To minimize those losses, farmers across the world invest billions of dollars in herbicides, mechanical weeding, and manual labor. Efficient weed control is, therefore, crucial in ensuring optimal crop production and economic viability for farmers. While traditional methods have been effective to some extent, they come with severe limitations. The most common form of weed control, manual weeding [8], is labor-intensive and time-consuming, making it impractical for large-scale farming. This trend incentivizes farmers to adopt more labor-efficient and cost-effective weed management strategies, including the increased use of highly potent herbicides. Notably, the herbicide market in India has been expanding at a remarkable annual growth rate of 15%, as reported in 2021 statistics. However, this reliance on chemical weed control poses significant challenges, including the emergence of herbicide-resistant weed populations, soil degradation, water contamination, and potential health risks to consumers due to the accumulation of chemical residues in agricultural produce. These challenges highlight the need for more precise, efficient, and sustainable weed control solutions.

Precision agriculture is a data-driven farming approach that uses advanced technologies to optimize agricultural practices in real time. By leveraging sensors, GPS, drones, AI, and automation, precision agriculture enables farmers to apply targeted interventions based on real-time data, minimizing resource waste and maximizing productivity as described in Figure 3. For example, in chemical weed control, integration with precision agriculture eliminates the need for blanket herbicide application [12], ensuring that only problematic areas receive treatment, thus reducing environmental impact while improving yield efficiency.

Unlike the Fourth Industrial Revolution (4IR) [14], 5IR prioritizes human–AI collaboration to enhance creativity, productivity, and social good [15]. At the core of 5IR is Human-Centered AI (HCAI), which are AI systems designed to align with human values, prioritize user needs, and ensure ethical decision-making. HCAI integrates AI into human workflows without replacing human judgment, making AI more trustworthy, explainable, and adaptable [14,15].

1.3. Purpose of Study

The following are this study’s primary contributions:

• Provides a cohesive understanding of the functions and synergies of important 5IR technologies, including artificial intelligence (AI), robots, the Internet of Things (IoT), 5G/6G, and edge computing, in the context of smart precision weeding.
• Demonstrates the technological advancements that are driving this change in agricultural operations by methodically charting the evolution from conventional weeding methods to contemporary intelligent systems.
• Identifies important obstacles to adoption in low-resource environments and makes tactical suggestions to promote the global use of smart weeding technologies that are inclusive, scalable, and in line with policies.

1.4. Structural Approach

To provide a detailed discussion of various technologies, the selection of sources is conducted through Google Scholar by entering the relevant keywords related to precision weeding technology. From the search results, articles are selected based on relevance, clarity, and perceived quality, resulting in a working set of approximately 160 papers. In addition to the academic literature, reports from credible organizations such as UNFAO and WHO have been referenced to provide more detailed infographics. This paper is structured as follows. The evolution of traditional weeding into smart weeding is covered in Section 2. The fundamental 5IR methods for weed removal are covered in Section 3. Robot innovation and sustainability initiatives are covered in Section 4. The integration of smart weeding with blockchain and decision support systems for improved farm-to-table services is covered in Section 5. Limitations and difficulties are covered in Section 6. Section 7 and Section 8 address initiatives for the future as well as suggestions for overcoming present difficulties.

2. Traditional Weeding and the Transition to Smart Technologies

2.1. Overview of Current Weeding Techniques

Mechanical weed control involves using tools and machinery to remove weeds [16]. Manual weeding involves hand-pulling or using simple tools to remove weeds. This method is particularly effective in small farms, gardens, and organic farming systems. Grazing animals are sometimes used to manage weeds in pastures, orchards, and rice fields [17]. Herbicides are widely used in modern agriculture to control weeds efficiently, especially in large-scale monoculture farming. Altering crop sequences disrupts the life cycles of the weeds and reduces the dependency on herbicides. For example, inter-cropping peas (Pisum sativum L.) alongside barley (Hordeum vulgare L.) cuts weed biomass by three times while enhancing soil nitrogen levels, compared to cultivating peas alone [7]. Cover crops, planted between growing seasons, improve soil health, reduce erosion, and suppress weeds by competing for resources and forming a mulch layer that inhibits weed germination. Some cover crops even release allelochemicals for added suppression. A meta-analysis on corn–soybean rotations found that cover crops significantly reduce weed biomass. Rye (Secale cereale L.) is a popular winter cover crop in corn and soybean systems for offering effective weed control while maintaining yield [7].

Water management is another method in the series of weed control assisting mechanisms. Some weeds have a much higher water demand than crops; for example, Cynodon dactylon has a transpiration coefficient of 813, more than double that of maize, 352. Weeds compete aggressively with crops for water, especially in non-irrigated or dryland farming systems, where they can extract up to 1250 tonnes of water per hectare. Terrestrial weeds thrive in aerobic soils, while semi-aquatic and aquatic weeds establish in saturated or flooded environments. For instance, Cyperus difformis grows best in 1.0 cm flooded soil, whereas Echinochloa crus-galli prefers 75–90% soil water content [18]. These differences suggest that strategic water management techniques, such as controlled flooding or soil drying, can be effective in suppressing specific weed species [7]. These strategies, when combined with traditional methods, enhance sustainable and cost-effective weed control. A summary of alternative weed control strategies can be seen in Figure 4. However, farmers must understand

• Crop–weed interactions to implement crop rotation and diversification effectively.
• Soil conditions and fertility management to optimize organic fertilizers and allelopathic plants.
• Weed biology and growth cycles to time cover crop planting and mulching correctly.
• Water management techniques to suppress weeds without harming crops.

Figure 4. Weed control methods. (a) Alternative strategies. (b) Impact of cover crop species on weed suppression [19] (smaller response ratio = greater weed control; blue = broadleaf crops, green = grass cover crops).

Figure 4. Weed control methods. (a) Alternative strategies. (b) Impact of cover crop species on weed suppression [19] (smaller response ratio = greater weed control; blue = broadleaf crops, green = grass cover crops).

Table 1 summarizes the pros and cons of all traditional weed control strategies.

2.2. The Evolution Towards Smart Weeding Technologies

The transition from traditional weeding methods to intelligent automation has been gradual, beginning with basic mechanized solutions and progressing toward advanced, data-driven technologies to accommodate with the mutating weed species which are undergoing a rapid transformation alongside as seen in Figure 5. Initially, mechanical weed control techniques, such as tractor-mounted cultivators, served as alternatives to manual labor. Over time, these systems have evolved to incorporate artificial intelligence (AI), computer vision, and automation, allowing for more precise and efficient weed management [5,12,20,21,22]. There even exist autonomous robotic weeders equipped with mechanical or laser-based removal systems and GPS-integrated IoT solutions [23,24,25,26] for optimized field operations [12,26,27,28]. This evolution reflects a significant shift from generalized mechanical weeding to precision-based approaches that minimize labor dependency while enhancing overall agricultural sustainability [29].

Table 1. A brief comparison of traditional weed control methods.

Category	Methodology	Advantages	Disadvantages
Mechanical	Weeds are physically removed using hand tools, cultivators, or machinery by uprooting, cutting, or tilling [30].	Effective against annual weeds, which are easily uprooted [31]. Improves soil aeration, enhancing root growth and microbial activity. More environmentally friendly and sustainable than herbicide use [32].	Less effective on perennial weeds, which may regrow from root fragments. Risk of soil erosion due to repeated tillage, especially in dry conditions. High labor and fuel costs.
Manual	Weeds are pulled by hand or with simple tools, ensuring root removal.	Highly selective, removing only weeds while preserving crops. No environmental pollution and no risk of herbicide resistance.	Extremely labor-intensive and time-consuming. Not feasible for large farms due to high labor costs. Weeds can regrow if not removed properly, particularly deep-rooted species.
Animal-assisted	Grazing animals, such as goats or sheep, feed on weeds to naturally control their growth.	Provides weed control while producing meat, milk, or eggs. No chemical inputs, making it an environmentally friendly method. Improves soil fertility through natural manure.	Requires proper management to prevent overgrazing and soil degradation. Some weeds may be toxic to livestock, such as ragwort poisoning in cattle [33]. Not effective for all cropping systems, especially row crops.
Thermal	Weeds are exposed to direct heat (flaming, steam, or hot water) to disrupt cell structures and cause desiccation [34].	Chemical free and effective on young weeds. No soil disturbance like herbicides and can fertilize soil by turning weeds into ashes.	Multiple applications usually necessary for perennial weeds. Not effective on deeply rooted weeds. Can cause heat injury to nearby crops.
Herbicides	Chemical solutions are applied to selectively or non-selectively kill weeds.	Highly effective and time-saving, allowing farmers to cover large areas quickly [31]. Reduces dependency on manual labor, lowering operational costs. Can be used in combination with no-till farming, reducing soil erosion.	Overuse has led to resistant weeds, such as glyphosate-resistant Palmer amaranth in the U.S [35]. Prolonged exposure to herbicides may affect farmers’ health, and residues in food may pose risks to consumers (e.g., Molinate has been banned in the US since 2009 due to reproductive toxicity).
Cover crops	Cover crops suppress weeds through smothering, competition, and allelopathy.	They physically block weed growth by forming a dense cover (smothering), compete for nutrients, water, and light (competition), and release allelochemicals that inhibit weed germination (allelopathy) [19].	Need knowledge about plant compatibility or it might have negative effects rather than positives [19].

Table 1. A brief comparison of traditional weed control methods.

Category	Methodology	Advantages	Disadvantages
Mechanical	Weeds are physically removed using hand tools, cultivators, or machinery by uprooting, cutting, or tilling [30].	Effective against annual weeds, which are easily uprooted [31]. Improves soil aeration, enhancing root growth and microbial activity. More environmentally friendly and sustainable than herbicide use [32].	Less effective on perennial weeds, which may regrow from root fragments. Risk of soil erosion due to repeated tillage, especially in dry conditions. High labor and fuel costs.
Manual	Weeds are pulled by hand or with simple tools, ensuring root removal.	Highly selective, removing only weeds while preserving crops. No environmental pollution and no risk of herbicide resistance.	Extremely labor-intensive and time-consuming. Not feasible for large farms due to high labor costs. Weeds can regrow if not removed properly, particularly deep-rooted species.
Animal-assisted	Grazing animals, such as goats or sheep, feed on weeds to naturally control their growth.	Provides weed control while producing meat, milk, or eggs. No chemical inputs, making it an environmentally friendly method. Improves soil fertility through natural manure.	Requires proper management to prevent overgrazing and soil degradation. Some weeds may be toxic to livestock, such as ragwort poisoning in cattle [33]. Not effective for all cropping systems, especially row crops.
Thermal	Weeds are exposed to direct heat (flaming, steam, or hot water) to disrupt cell structures and cause desiccation [34].	Chemical free and effective on young weeds. No soil disturbance like herbicides and can fertilize soil by turning weeds into ashes.	Multiple applications usually necessary for perennial weeds. Not effective on deeply rooted weeds. Can cause heat injury to nearby crops.
Herbicides	Chemical solutions are applied to selectively or non-selectively kill weeds.	Highly effective and time-saving, allowing farmers to cover large areas quickly [31]. Reduces dependency on manual labor, lowering operational costs. Can be used in combination with no-till farming, reducing soil erosion.	Overuse has led to resistant weeds, such as glyphosate-resistant Palmer amaranth in the U.S [35]. Prolonged exposure to herbicides may affect farmers’ health, and residues in food may pose risks to consumers (e.g., Molinate has been banned in the US since 2009 due to reproductive toxicity).
Cover crops	Cover crops suppress weeds through smothering, competition, and allelopathy.	They physically block weed growth by forming a dense cover (smothering), compete for nutrients, water, and light (competition), and release allelochemicals that inhibit weed germination (allelopathy) [19].	Need knowledge about plant compatibility or it might have negative effects rather than positives [19].

3. Core 5IR Technologies Driving Smart Precision Weeding

3.1. 5IR Technologies in Smart Precision Weeding

The increasing focus on food safety, sustainability, ethical sourcing, and evolving economic conditions is driving the need for advanced agricultural technologies [1]. The 5IR emphasizes human-centered and sustainable approaches by integrating AI, robotics, and IoT with ethical and environmental considerations [38]. 5IR-driven innovations directly contribute to United Nations’ Sustainable Development Goals (SDGs) through resource-efficient production as shown in Figure 6. Enhanced food safety and traceability technologies support SDG 3 (Good Health & Well-being) [36], ensuring consumers have access to safer and more nutritious food. Technologies such as online grocery platforms, meal kits, and QR-code-based traceability systems [36,38] allow consumers to access detailed product information instantly and greater transparency in food labeling. As consumer preferences drive the adoption of these advancements, 5IR is not only transforming agriculture but also playing a crucial role in achieving a more sustainable and equitable global food system [39]. The key upgrades in each industrial revolution, as well as real–time integration of these technologies into field robots, can be seen in Figure 7.

Figure 6. (a) Changing agricultural consumer trends. (b) Their contribution to the 2030 SDGs [40].

Figure 6. (a) Changing agricultural consumer trends. (b) Their contribution to the 2030 SDGs [40].

3.1.1. AI-Driven Decision Intelligence for Weed Management

Smart precision weeding is an example of how 5IR technologies, specifically artificial intelligence and sensor-based systems, are transforming weed management by enabling more focused, effective, and ecologically friendly methods. Traditional weed management relies on broadcast application of selective herbicides, leading to significant environmental contamination through run-off and drift. Hydraulic sprayers, including boom and air-blast sprayers, cause inefficient herbicide distribution, leading to only 30–40% target coverage [41]. Variable Rate Spraying (VRS) or Variable Rate Application (VRA) is gaining popularity for its ability to reduce agrochemical use by up to 97% while enhancing weed control efficiency and minimizing costs using various sensors such as the following:

• Near-infrared(NIR) sensors: To detect plant canopy structures [42].
• Ultrasonic sensors: To estimate plant height and total volume [43]
• Laser scanners: To measure canopy volume in real-time [44].

However, the sensors exhibit struggle in certain situations as NIR lacks detailed feature recognition, ultrasonic sensors cannot filter out background noises, and laser scanners do not function properly in foggy or snowy situations. Thus, more machine vision techniques were introduced to further improve VRS by allowing real-time weed detection. Computer vision enables real-time weed detection based on size, shape, color, and texture differentiation. Transformer-based models, such as SSRT [45], demonstrate the potential of deep learning in interpreting fine-grained interactions, which can be adapted for complex weed recognition in precision weeding systems. Lamm et al. [46] developed a vision-based sprayer that accurately differentiated 88.8% of weeds in cotton crops. Advanced imaging methods, such as high-speed imaging and remote sensing, further enhance weed detection and targeted spraying efficiency. Also, deep bidirectional LSTM models, known for auto-evaluation tasks, offer a basis for intelligent decision-making engines in AI-powered weed assessment platforms [47].

Figure 7. (a) Key upgrades in each industrial revolution [48]. (b) Robotics application in various aspects of agriculture [49].

Figure 7. (a) Key upgrades in each industrial revolution [48]. (b) Robotics application in various aspects of agriculture [49].

Micro-dosing systems take precision spraying a step further by delivering minimal doses of non-selective herbicides directly to weeds, applying the combined concept of computer vision, agriculture, and robotics [50]. This technology uses 99.3% less herbicides than normal brute application [12], offering several advantages, such as (1) more precise weed targeting, reducing chemical exposure to crops; (2) significantly lower environmental contamination through reduced drift and run-off; and (3) improved cost efficiency, as less herbicide is required for effective weed control [12,51]. However, for micro-dosing to be effective, accurate plant identification and position tracking are essential [12]. Weed detection in this genre relies on Color Vegetation Indices (CVIs) and Image Segmentation for crop type differentiation and plant–soil discrimination.

CVIs combine visible and NIR reflectance to generate a single grayscale image that highlights plant regions. Different color indices help differentiate plants from soil with varying accuracy [12,51]. Some methods convert images into different color spaces before applying indices: RGB normalization adjusts for lighting variations by normalizing pixel values as shown in Equation (1). Several indices are used for plant–soil discrimination. Excess Green ($ExG$) enhances green vegetation while suppressing soil and non-green objects as shown in Equation (2). Similarly, Excess Red ($ExR$) highlights red components Equation (3), sometimes useful for distinguishing certain crop types [51].

$$r+g+b=1$$

(1)

$$ExG=2g-r-b=3g-1$$

(2)

$$ExR=1.4r-g$$

(3)

where r = red, g = green, b = blue, $ExG$ = Excess Green, $ExR$ = Excess Red, and $ExGR$ = Excess Green minus Excess Red.

The Normalized Difference Index (NDI) or Normalized Difference Vegetation Index (NDVI) compares differences between two color channels to enhance vegetation features. NDI uses NIR and red reflectance to measure vegetation health. These indices work differently depending on lighting conditions, crop types, and background soil composition [51]. After applying color indices, image segmentation is performed to separate plant parts or plants from the background (soil) on the resulting grayscale images. The most common method is Otsu’s thresholding [52], which automatically selects an optimal threshold value to separate plant pixels from non-plant pixels [12]. This method shows the optimal result when the weeds are visually distinct from the soil. While color indices help identify plant regions, they do not differentiate crops from weeds. To achieve this, additional shape and textural features are analyzed. Shape features measure characteristics like size, roundness, or edge sharpness to distinguish different plant types. Texture Features analyze surface patterns to differentiate crop and weed species [12,41,51].

On the other hand, machine learning techniques enhance weed classification and spraying accuracy, with SVM and fuzzy decision-making achieving up to 92.9% accuracy [41]. On the deep learning side, CNNs eliminate the need for manual feature selection, but they require high computational power. A study evaluating AlexNet, VGG-16, and GoogleNet for weed classification and precision spraying in a strawberry field (12,443 images) found that VGG-16 achieved the highest accuracy (precision, recall, F1-score) with a 93% complete spraying (CS) rate [41] compared to the mentioned counterparts. Nevertheless, sprayer effectiveness declined at speeds above 3 km/h, highlighting the need for speed optimization [41]. Table 2 compares the explained technologies with their corresponding performance metrics.

Table 2. Examples of AI-based technologies in precision weeding.

Technology	Application	Method or Mechanism	Outcomes	Performance Metrics
VRS [53]	Control the amount of pesticide application	Electromechanical flow control	Increases yield while reducing pesticide expenditure	–
Microdosing [54]	Weed control in tomato	Hyperspectral imaging (Bayesian classifier) + thermal microdosing with food-grade oil heated to 160 °C	95.8% of S.nigrum and 93.8% of A.retroflexus eliminated with 2.4% damage to tomato crops	Bayesian classifier: 95.9% accuracy
ML [55]	ML-based weed identification using SVM and WCTATP method	Image processing pipeline incorporating Curvelet Transforms and optimized feature selection for accurate weed identification	Spot spraying of herbicides becomes more efficient	SVM: 97.3%, WCTATP: 98.3% accuracy
DL [56]	Weed detection and classification	Applies several DL models to detect and classific corn and surrounding weed species	YOLOv7 attained the highest accuracy and speed among all models	mAP of YOLOv7: 89.93%, YOLOv8x: 89.39%, and Faster-RCNN: 81.29%
DL [57]	CNN for site-specific weed management	Targeted herbicide application gets easier via weed mapping based on weed’s responsiveness	Weeds prone in bermudagrass turf categorized by sensitivity to ACCase, ALS, and synthetic auxin herbicides.	DenseNet: 99.85%, GoogLeNet: 99.53%, and ResNet: 99.80% accuracy
HTP [58]	Soybean phenotyping to optimize crop breeding	Soybean Phenotype Measure-instance Segmentation (SPM-IS) algorithm [feature pyramid network + PCA + instance segmentation]	Reduces time and labor invested in manual phenotyping	mAP: 95.7%

Table 2. Examples of AI-based technologies in precision weeding.

Technology	Application	Method or Mechanism	Outcomes	Performance Metrics
VRS [53]	Control the amount of pesticide application	Electromechanical flow control	Increases yield while reducing pesticide expenditure	–
Microdosing [54]	Weed control in tomato	Hyperspectral imaging (Bayesian classifier) + thermal microdosing with food-grade oil heated to 160 °C	95.8% of S.nigrum and 93.8% of A.retroflexus eliminated with 2.4% damage to tomato crops	Bayesian classifier: 95.9% accuracy
ML [55]	ML-based weed identification using SVM and WCTATP method	Image processing pipeline incorporating Curvelet Transforms and optimized feature selection for accurate weed identification	Spot spraying of herbicides becomes more efficient	SVM: 97.3%, WCTATP: 98.3% accuracy
DL [56]	Weed detection and classification	Applies several DL models to detect and classific corn and surrounding weed species	YOLOv7 attained the highest accuracy and speed among all models	mAP of YOLOv7: 89.93%, YOLOv8x: 89.39%, and Faster-RCNN: 81.29%
DL [57]	CNN for site-specific weed management	Targeted herbicide application gets easier via weed mapping based on weed’s responsiveness	Weeds prone in bermudagrass turf categorized by sensitivity to ACCase, ALS, and synthetic auxin herbicides.	DenseNet: 99.85%, GoogLeNet: 99.53%, and ResNet: 99.80% accuracy
HTP [58]	Soybean phenotyping to optimize crop breeding	Soybean Phenotype Measure-instance Segmentation (SPM-IS) algorithm [feature pyramid network + PCA + instance segmentation]	Reduces time and labor invested in manual phenotyping	mAP: 95.7%

3.1.2. Computer Vision and Sensing for Weed Detection

Practical weed detection presents significantly greater challenges compared to laboratory testing, as it must operate in real-time on moving platforms such as tractors and drones. In controlled laboratory settings, factors such as background, lighting conditions, and viewing angles remain constant. However, in field applications, these variables fluctuate unpredictably [59]. Additional challenges include the presence of unidentified plant species or insects that may resemble plants. Plus, when detection is integrated with active intervention methods such as spraying or lasers, the system must not only identify weeds but also execute the elimination process immediately. Therefore, achieving high accuracy within a constrained time frame is essential for effective weed management. These challenges are common in many fields that require computer vision and image processing algorithms to overcome those hurdles [60,61,62]. Blue River Technology’s See & Spray system is an AI-driven precision weeding technology that optimizes herbicide application through CV, ML and robotics, and it was acquired by John Deere in 2017 for $305M. Unlike traditional broadcast spraying, which applies herbicides across entire fields, See & Spray uses high-resolution cameras to scan crops and weeds in real time [63,64]. DL algorithms analyze these obtained images to differentiate weeds from crops, enabling targeted herbicide application only where necessary [65]. This model achieves a 98% hit rate [63] while using 77% less herbicide% [64] than traditional broadcasting. Additionally, it has been proven to reduce the likelihood of herbicide resistance by 83% [63]. It leverages a database of over 1 million weed images. Once a weed is detected, the system activates a precise nozzle on the sprayer boom, releasing a small, targeted amount of herbicide to inhibit weed growth while minimizing impact on surrounding crops. This is made possible using 25 on-board Nvidia Jetson AGX Xavier graphic processors [65,66]. Also, the usage of vision sensors with different field of views and multitasking algorithms help in mitigating various scenarios efficiently [67,68]. Plus, this precision can be further enhanced by integrating IoT and edge intelligence into weed management systems.

3.1.3. IoT Architecture and Edge Intelligence in AgriTech

The integration of IoT and edge computing in agriculture is transforming weed management through real-time monitoring, sensor networks, and remote sensing. These technologies enable precise identification and control of weeds, improving efficiency while reducing environmental impact. IoT sensors collect data from fields, and edge computing processes the data locally [69], reducing latency compared to cloud computing. This enables faster decision-making for weed detection and treatment, often through AI-based image processing and robotic weeders. Kalyani & Collier (2021) [39] present a systematic survey on integrating wireless sensor networks (WSNs) with cloud and edge computing for smart agriculture. Cloud computing struggles with high latency, bandwidth limitations, and security risks, making it inefficient for real-time weed detection and control. Fog and Edge computing solve these issues by processing data closer to the source, enabling faster, localized decision-making [38]. By analyzing weed detection data on local fog nodes (such as farm-based servers), fog computing serves as a link between IoT devices and the cloud. This reduces latency and allows real-time weed classification using AI models. Contrarily, edge computing [70] allows for immediate decision-making without the need for external servers by processing data directly on IoT devices like robotic weeders and drones. [38,39]. A hybrid Cloud-Fog-Edge approach is also common, as seen in Figure 8; cloud computing stores long-term weed trends for AI-driven insights, fog nodes compile farm-wide data, and edge computing allows real-time weeding [39,71].

For the weed classification, traditional machine learning techniques like Support Vector Machines (SVM) and Random Forest have been widely used. These methods frequently rely on manually crafted feature extraction methods like color indices, textures, and shape analysis. However, these techniques show inconsistent classification performance because they are extremely sensitive to environmental changes, such as occlusions and changes in illumination [12,71,74]. Image-based agricultural classification has greatly benefited from recent developments in deep learning (DL), especially with Convolutional Neural Networks (CNNs), which can automatically extract hierarchical features from input images without the need for manual preprocessing as described in Figure 9. However, unfavorable weather conditions, like haze, may diminish the effectiveness of these models by introducing reduced contrast, color distortion, and noise artifacts, which ultimately affect model performance [71]. In order to overcome these obstacles, scientists have used image dehazing methods(e.g. the Dark Channel Prior-based dehazing algorithm) to improve visual clarity prior to classification [71,75,76].

The integration of CNN-based models and IoT-driven robotic systems for precision agriculture has been the subject of numerous studies. For instance, a recent approach deployed a Raspberry Pi-based CNN model for on-field weed detection [77] to improve region-based classification accuracy. This has led to the increasing adoption of deep learning-driven haze removal techniques, which enhance input image quality before classification, thereby improving the overall accuracy of automated weed detection systems (see Figure 10). The following describes one of the algorithms used for synthesizing hazy images to produce training sets.

$$I\left(x\right)=J\left(x\right)t\left(x\right)+A\left(1-t\left(x\right)\right)$$

(4)

where $I\left(x\right)$ = hazy image, $J\left(x\right)$ = clear image, A = global atmospheric light, and $t\left(x\right)$ = portion of non-scattered light that enter the camera sensor [78].

$$t\left(x\right)=e^{-\beta d\left(x\right)}$$

(5)

where $\beta$ = medium extinction coefficient and $d\left(x\right)$ = scene depth [78].

In parallel with these visual enhancements, GPS-based navigation technologies provide accurate spatial targeting in agriculture.

Figure 9. Deployment of robotics in precision farming. (a) Schematic diagram for weed detection using a CNN framework on a Raspberry Pi machine for real-time weed detection. (b) A robot supporting modular combinations for specific agricultural tasks [79].

Figure 9. Deployment of robotics in precision farming. (a) Schematic diagram for weed detection using a CNN framework on a Raspberry Pi machine for real-time weed detection. (b) A robot supporting modular combinations for specific agricultural tasks [79].

Figure 10. (a,c) Images of broadleaf, grass, soil, and soyabean without any processing (b,d) Same images after applying hazing filters [71].

Figure 10. (a,c) Images of broadleaf, grass, soil, and soyabean without any processing (b,d) Same images after applying hazing filters [71].

3.1.4. Satellite Navigation and GPS-Enabled Precision Targeting

While future 5G and 6G networks seek to improve connectivity for smart farming solutions, GPS and remote sensing allow for accurate soil assessments and real-time crop health monitoring. By using ground-based reference stations to rectify signal errors typically present in standard GPS, Differential GPS (DGPS) achieves sub-meter precision and increases standard GPS accuracy [23]. Furthermore, there are new, inexpensive DGPSs that are more accurate than conventional GPS [80]. In agricultural applications, GPS enables accurate field mapping, facilitating precise boundary delineation and optimized land planning. It also simplifies Variable Rate Technology (VRT), ensuring the site-specific application of fertilizers and pesticides to minimize waste. Additionally, GPS helps with data collection for well-informed agronomic decision-making, which supports yield monitoring. Precision planting, spraying, and harvesting are made possible by autonomous machinery operation, which is another important application. For example, tractors can precisely follow pre-programmed routes thanks to GPS-guided systems like John Deere’s AutoTrac technology, which minimizes field operations overlap and uses less fuel [81]. In fact, GPS technology has been shown to increase operational efficiency by up to 30%, making fieldwork more cost-effective as in the case of Figure 11b.

3.1.5. Aerial and Proximal Remote Sensing in Smart Farming

Remote sensing techniques play a crucial role in site-specific weed management by enabling real-time monitoring and automated analysis. High-resolution multispectral, thermal, and radar imagery from satellites like Sentinel-1, Sentinel-2 [25,83], Landsat and Planet Scope, along with UAV-based imaging, support hybrid interpolation for weed detection [83]. These technologies facilitate fully automated weed mapping, reducing the need for manual field inspections [26]. Machine learning-based remote sensing also aids in assessing soil parameters essential for fertilization. The integration of Synthetic Aperture Radar (SAR) and optical satellite missions enhances daily, high-resolution monitoring, making remote sensing an indispensable tool in precision agriculture as seen in Figure 12 [84,85].

3.1.6. Ultra-Fast Connectivity: Role of 5G and 6G in Smart Weeding

5G technology offers high bandwidth, low latency, and the ability to connect one million devices per km² [84], a significant advancement compared to former networks, as shown in Figure 13. Robotic weeders have evolved significantly, with the core challenge being the accurate classification of weeds and crops [27] where 5G enhances detection through high-speed data transmission, enabling fine-grained identification using DL and image processing techniques [28]. Unmanned Aerial Vehicles (UAVs) controlled via 5G networks are increasingly used for weed management [28]. These UAVs can accurately spray herbicides through various nozzle systems, thereby reducing chemical waste [84]. Plus, a 5G network can assist disease identification and pest control through the integrated implementation of thermal cameras and AI-powered image processing, ensuring higher crop yields [88].

A 5G network also facilitates cloud-based virtual and augmented reality applications, improving training programs for farm operators. It enables real-time monitoring of environmental conditions, optimizing decision-making in smart farming. As an example, hill agriculture remains one of the least mechanized sectors despite having resources that support higher production. The advancement of 5G and future 6G networks will bridge this gap by enabling seamless connectivity in remote locations, enhancing automation, precision farming, and smart tech integration [90]. A Chinese study indicates a growing interest of 31.47% among farmers in adopting 5G solutions, emphasizing its potential in transforming weed management and sustainable agricultural practices [90].

3.1.7. High-Throughput Phenotyping in Precision Weeding

High-throughput phenotyping (HTP) has revolutionized agricultural practices by enabling the collection of vast amounts of data on plant traits [91]. Precision weeding, an essential component of modern agricultural systems, efficiently detects, classifies, and manages weeds by combining machine learning and HTP technologies. HTP platforms, comprising both aerial and terrestrial sensors, have been proofed to successfully detect and classify weeds in a range of crop species [91]. UAVs and autonomous ground robots equipped with RGB, hyperspectral, LiDAR, and thermal imaging sensors take high-resolution images of crop fields almost instantly [91,92]. For example, UAV-based platforms have been used to detect weeds in wheat (Triticum aestivumL.) [93]), maize (Zea maysL.) [94], and sunflower (Helianthus annuusL.) fields. These systems employ machine learning classifiers such as SVMs and k-nearest neighbors (kNN) to differentiate weeds from crops, allowing for targeted herbicide application and mechanical weeding [95,96]. This method commonly employs an organized ICQP (Identification, Classification, Quantification, and Prediction) workflow as seen in Figure 14 [91]:

• Identification: ML models analyze RGB, hyperspectral, and thermal imaging data to detect weeds at early growth stages. Techniques like Gaussian mixture models (GMMs) and quadratic discriminant analysis (QDA) [97] can be used to identify species-specific weed infestations.
• Classification: ML algorithms categorize weeds based on species, growth stage, and spatial distribution. DL models have shown high accuracy in classifying weeds across different environments [91,95,96].
• Quantification: Once classified, AI models assess weed biomass and density within a given area [91]. Regression-based models and DL approaches provide estimates of weed severity, aiding in decision-making for targeted treatments.
• Prediction: Predictive models use historical and real-time phenotypic data to forecast weed outbreaks, optimizing herbicide application schedules and mechanical weeding strategies.

Figure 14. HTP-driven precision weeding. (a) Role of UAVs in HTP [91] (b) Sample workflow for HTP [91].

Figure 14. HTP-driven precision weeding. (a) Role of UAVs in HTP [91] (b) Sample workflow for HTP [91].

HTP and AI-powered precision weeding are transforming sustainable agriculture by improving weed management through automated, data-driven approaches [91]. Future advancements in sensor fusion, edge computing, and deep learning will further refine these technologies, increasing efficiency in diverse cropping systems [5,91]. Sensor-based autonomous platforms, such as ground robots, enable site-specific weed removal with minimal crop disturbance. However, challenges remain in data processing, model generalization across environments, and real-time decision-making [27,94]. Plus, despite its potential, adoption remains limited due to high costs, technical complexity, and connectivity challenges.

3.1.8. GIS-Based Weed Mapping and Spatial Analysis

Geographic Information Systems (GIS) are essential for storing, analyzing, and processing agricultural data. GIS is used in land resource surveys, crop suitability evaluations, agricultural management, and soil erosion protection. By enhancing decision-making, GIS improves resource efficiency and boosts crop yields [98]. The integration of “3S” technology (remote sensing, GPS, and GIS) is transforming agricultural management. RS captures image data, GPS enables navigation and positioning, and GIS provides robust data storage, processing, and analysis. Together, these technologies form a comprehensive information system that supports decision-making in agriculture as described in Figure 15. The system facilitates tasks like data mining, information processing, analysis, and 3D dynamic simulation, which are crucial for solving complex agricultural problems [98].

GIS is incredibly useful to yield mapping and prediction in real-world applications. In order to help with marketing and storage planning, farmers can utilize GIS-based tools to generate yield maps that forecast the amount of crops that will be harvested [98,100]. Drones are used to count oil palm trees in Indonesia because they can collect data quickly and with a 95% accuracy rate [99]. As per C. Palaniswami [101], the sugar industry also uses GIS for selective harvesting as one precision agriculture application. This approach can optimize environmental protection and manage the agricultural production system in fragile ecosystems, such as coastal floodplains. Soil heterogeneity, which significantly impacts crop yield variability, is another challenge in agriculture that GIS helps address [102]. A study by Oshunsanya S.O. et al. at the University of Ibadan used the Inverse Distance Weighting (IDW) interpolation technique to map the spatial distribution of soil properties. The results were reclassified into different soil management categories, and these maps were overlaid to identify specific site management units (SSMUs) [103]. This approach helps manage the varying soil properties and mitigates yield losses caused by improper use of agrochemicals.

3.1.9. Sensor Networks for Real-Time Agro-Environmental Monitoring

Field sensors play a crucial role in monitoring key environmental parameters such as soil moisture, temperature, humidity, and nutrient levels [21,104]. These sensors are deployed in agricultural fields and continuously collect data, which is then transmitted to farmers via a wireless network [21,104]. Plus, these data are integrated with the remote sensing data obtained using UAVs to access real-time information that allows for more precise irrigation management, optimizing water usage, and weed control [21,38,39,59,105]. For weed detection and management, several types of sensors are commonly used in precision agriculture. These sensors help identify, classify, and control weeds more efficiently while minimizing herbicide use. The usage of several state of the art algorithms alongside various sensor configurations can aid in better weed detection and management as seen in Figure 16 [106]. Table 3 describes the commonly used sensors.

Table 3. Commonly used sensors in precision weeding.

Sensor Name	Functionality	References
RGB Cameras	Capture high-resolution images	[63,64]
Multispectral Sensors	Detect weeds based on differences in light reflectance compared to crops	[107]
Hyperspectral Sensors	Provide detailed spectral data	[108]
NIR Sensors	Differentiate weeds from crops based on light absorption	[42,51]
LiDAR Sensors	Create 3D models of fields and identifies weeds based on height and structure	[109]
NDVI Sensors	Measure plant health using vegetation indices	[51,52]
Thermal Infrared Sensors	Identify temperature variations	[110]
GPS and GIS Sensors	Enable weed mapping and precision herbicide application	[98,100,101,103]
Electrochemical and Dielectric Soil Moisture Sensors	Detect soil moisture and nutrient levels	[111]
Mechanical Sensors	Detect the force required to penetrate soil	[111]

Table 3. Commonly used sensors in precision weeding.

Sensor Name	Functionality	References
RGB Cameras	Capture high-resolution images	[63,64]
Multispectral Sensors	Detect weeds based on differences in light reflectance compared to crops	[107]
Hyperspectral Sensors	Provide detailed spectral data	[108]
NIR Sensors	Differentiate weeds from crops based on light absorption	[42,51]
LiDAR Sensors	Create 3D models of fields and identifies weeds based on height and structure	[109]
NDVI Sensors	Measure plant health using vegetation indices	[51,52]
Thermal Infrared Sensors	Identify temperature variations	[110]
GPS and GIS Sensors	Enable weed mapping and precision herbicide application	[98,100,101,103]
Electrochemical and Dielectric Soil Moisture Sensors	Detect soil moisture and nutrient levels	[111]
Mechanical Sensors	Detect the force required to penetrate soil	[111]

3.2. Real-World Applications of AgroTech

AGRO (AGRiculture mOnitoring), a real-time monitoring system that partitions a field into smaller sections known as grids, is an example of how sensors are used in precision agriculture. Numerous sensors that monitor soil conditions are installed in each grid. Each grid is given a grid leader which collects sensor data and sends it to a central sink node. After that, the sink sends the data to a mobile app, providing the farmer information regarding the status of the field. The farmer can improve his water conservation measures by using a digital valve to water only the affected area, rather than the entire field, if a particular grid needs irrigation [21]. Despite its advantages, real-time monitoring in smart agriculture presents challenges related to data volume and sensor energy consumption. The continuous collection of sensor data generates a massive amount of information [21,37], which can lead to (1) increased network congestion, (2) higher energy consumption, and (3) complexity in decision-making due to the vast amount of data, making it difficult for farmers to extract meaningful insights in a timely manner [37]. To address these challenges, different data collection models are employed in sensor-based networks [114]:

• On-Demand Sensing: Data is collected only when requested by the farmer.
• Periodic-Based Sensing: Data is collected and sent at fixed time intervals, which is the most used approach in smart agriculture. However, periodic sensing can contribute to big data problems, causing unnecessary data transmissions that consume energy and increase network congestion [115].
• Event-Driven Sensing: Sensors transmit data only when a significant change in field conditions is detected (e.g., soil moisture drops below a threshold). This can also avoid data overload usually happening in periodic-based sensing.

In addition to above models, modern smart agriculture systems are implementing data reduction techniques, such as hierarchical data aggregation where the grid leader summarizes sensor readings and only sends relevant updates instead of all the raw new data [21] and machine learning-based decision support where AI algorithms can analyze large datasets to identify patterns and assist farmers in making optimized decisions without requiring all raw data [21]. The commonly used weed datasets to develop the AI systems are summarized in

Table 4. Comparison of various public weed datasets (the climate patterns are inferred using Köppen–Geiger climate classifications [116]).

Dataset	Description	Properties	Location and Climate Zone	Source
Plant seedlings dataset [117]	960 plants (weed, crops, and others) of 12 species	Annotated RGB, 10 pixels/mm resolution	Denmark, cool temperate, oceanic	https://vision.eng.au.dk/plant-seedlings-dataset/(Access date—11 June 2025)
DeepWeeds [118]	17,509 images of 8 weed species in northern Australia	Annotated, Inception-v3: 95.1%, ResNet-50: 95.7% accuracy	Northern Australia, tropical savanna	https://github.com/AlexOlsen/DeepWeeds (Access date—11 June 2025)
Crop/Weed Field Image Dataset [119]	60 top-down field images of 162 crops and 332 weed plants	Multispectral images, annotated with segmentation mask and weed–crop labels	Bonn, Germany, temperate oceanic	https://github.com/cwfid/dataset?tab=readme-ov-file (Access date—11 June 2025)
Weed detection dataset [120]	39 images of carrot-weed	RGB images taken under variable light conditions	North Macedonia, continental, hot summer	https://github.com/lameski/rgbweeddetection (Access date—11 June 2025)
Weed Detection in Soybean Crops [121]	15,336 images, composed of 3249 soil, 7376 soybean, 3520 grass, and 1191 broadleaf weeds	Annotated, RGB satellite images	Brazil, tropic or subtropical, wet summer	https://data.mendeley.com/datasets/3fmjm7ncc6/2 (Access date—11 June 2025)
weedNet Dataset [122]	465 images are taken from 40 × 40 m field using micro aerial vehicle (MAV)	Multispectral images labelled as weed, crop or backgroud. RGB images are also avaible but not annotated.	Sydney, Australia, warm temperate, dry winter	https://github.com/inkyusa/weedNet?tab=readme-ov-file (Access date—11 June 2025)
Sugarbeet 2016 [123]	5TB data worth of images, containing sugar beet crops and weeds, are taken from a sugar beet farm in Bonn, Germany, over three months	Multispectral, RGB, 3D point cloud data, laser data, odometry measurements, and GPS positions are included.	Germany, temperate oceanic	https://www.ipb.uni-bonn.de/data/sugarbeets2016/index.html (Access date—11 June 2025)
Leaf counting dataset [124]	9372 images of weeds are collected in fields across Denmark using both cell phones and industrial cameras	RGB images of weeds with the number of leaves counted	Denmark, cool temperate	https://vision.eng.au.dk/leaf-counting-dataset/ (Access date—11 June 2025)
Remote Sensing 2018 Weed Map Dataset [125]	18,746 images taken from Eschikon, Switzerland, and Rheinbach, Germany, using drones.	Annotated multispectral or orthomosaic or tiled images, 8.2 cm/pixel and 13 cm/pixel resolutions.	Switzerland and Germany, humid temperate	https://projects.asl.ethz.ch/datasets/doku.php?id=weedmap:remotesensing2018weedmap&s[]=weedmap (Access date—11 June 2025)
Crop vs weed discrimination dataset [126]	40 images of carrot, onion and weeds taken using Teledyne DALSA Genie Nano camera	Annotated multispectral images at 2.5 px/mm resolution	UK, cool temperate, maritime	https://lcas.lincoln.ac.uk/wp/research/data-sets-software/crop-vs-weed-discrimination-dataset/ (Access date—11 June 2025)

. From the collected datasets, it is evident that most publicly available and widely used weed datasets originate predominantly from Europe, North America, and Australia. This geographical bias introduces limitations in the generalizability of weed detection systems to other agricultural regions, particularly underrepresented areas such as Southeast Asia. Weed species vary significantly by region, and datasets collected in temperate zones may not reflect the diversity or growth patterns of weeds in tropical or subtropical climates. Moreover, imaging conditions, such as light intensity, soil color, and canopy structure, differ across geographies, directly impacting the performance of computer vision models. Projects that integrate the aforementioned technologies into practical applications will be covered in the next section.

4. Robotic Innovations and Sustainable Product Solutions

4.1. Commercial Products and Research Solutions

The Carbon Robotics’ Laser Weeder eliminates 99% of weeds by identifying them with cameras and using high-powered lasers to heat water inside their cells, causing rupture and death within 0.1 seconds. The system employs 42 high resolution cameras and deep-learning crop models which were trained on more than 40 million plants. AI determines weed locations, guiding lasers to their meristem, the growth zone of a plant, for effective elimination. Scanners linked to rear-mounted cameras enable sub-millimeter accuracy, preventing damage to crops even in challenging conditions such as occlusion by muck soil [127].

The ARA Field Sprayer by Ecorobotix uses AI-driven technology to reduce plant protection products and fertilizers by up to 95%. Its ultra-targeted spraying lowers costs, offering a payback period of 2–4 years [128]. With an operating speed of 7.2 km/h (4.5 mph), ARA ensures efficient, targeted application, reducing chemical use by 50% to 95% depending on crop stage and weed density. This minimizes environmental impact, prevents crop stress, and promotes earlier harvests by reducing phytotoxicity. Minimal drift design ensures precise application even in windy conditions, making ARA a sustainable, high-precision alternative to conventional broadcast spraying [127]. As of March 2025, the ecorobotix algorithms work successfully on 17 crops species [128].

As part of this review, research is carried out to detect stem rot in palm trees due to basal stem rot disease (BSR), caused by the Ganoderma boninense fungi. The disease causes the palm oil industry an approximate loss of USD 500 millions annually, as the discoloration affected by the pathogen is extremely challenging to detect with the naked eye. So, the critical period to give treatment is usually missed. Raspberry Pi 3-based robots equipped with SLAM technologies and pragmatic intelligent monitoring systems that automatically go around the farm to detect the disease symptoms are low-cost options that can efficiently mitigate the disease. Figure 17 indicates the image of a palm tree affected by the fungi and the vision system implementation. More information on robotic applications specific to precision weeding is described in Table 5.

4.2. Integration of Drones and Ground Robots

The integration of drones and ground robots offers a synergistic approach to precision weeding by combining aerial monitoring with targeted ground-level action. Drones capture real-time data on weed distribution and crop health, which is then shared with ground robots to precisely target and eliminate weeds. This combination enhances the efficiency and scalability of weed control efforts by leveraging drones’ broad coverage and ground robots’ accuracy [93,94,99,129]. Additionally, ground robots can serve as mobile docking stations, providing recharging opportunities for drones and enabling continuous operation, further increasing the effectiveness of the system [130].

4.3. Control Algorithms and Navigation Systems

SLAM (Simultaneous Localization and Mapping) plays a pivotal role in the navigation of autonomous weeders, field monitoring, and 3D mapping [131]. SLAM is a technique that allows a robot to simultaneously create a map of its surroundings and track its own position within that environment, all in real time without prior geographical knowledge of that region [131,132] as described in Figure 18. Path planning algorithms combined with SLAM ensure that autonomous weeders can continuously update their location and adapt to changes in the field, providing consistent and accurate weeding with minimal human oversight and crop disruption. SLAM allows the robot to map its environment even in fields that may lack distinct landmarks or where visual cues are limited. This capability is critical for autonomous navigation, as it enables the weeder to avoid obstacles, including crops, trees, and irregular terrain, while maintaining precise control over its movement [131].

4.4. Sustainability and Green Technologies

As sustainability becomes a top priority, energy-efficient solutions are critical to lowering the environmental impact of this sector. Statistically speaking, 53% of methane (CH₄), 78% of nitrous oxide (N₂O), and 21% of carbon dioxide (CO₂) are released through various stages of farming processes [133,134]. Solar-powered weeding robots harness renewable energy to operate, reducing reliance on non-renewable energy sources and minimizing the carbon footprint. Incorporating climate-smart practices, such as drought-resistant crop varieties, can enable farmers to improve their resilience to climate variability while contributing to the overall goal of sustainable food production systems [95]. Nevertheless, the effectiveness of solar panels is directly related to geographical location and weather, making it challenging to operate consistently in places with little sunlight [135].

Table 5. Overview of agricultural weeding robots: sensors, applications, and performance.

Robot	Sensors/Imaging	Speed/Coverage	Application Type	Power Source	Performance/Accuracy
Asterix [136]	Nvidia Jetson TK1, 4682 4MP sensor	0.8 m/s	Chemical	Hybrid (48 V DC, 4-stroke generator)	Drop-On-Demand (DOD) system has been proven to kill off 100% of weed, reducing herbicide usage 10 times
AgBotII’s multimodal weed control [137]	RGB camera, Novatel SPAN-IGM-A1 (RTK)	1.3 m/s	Machanical or chemical	Battery	Classification Accuracy: 92.3%
Laserweeder [37]	Nvidia GPUs, RGB cameras, GPS, Lidar	0.4 m/s	Laser	PTO-driven generator	Kills up to 99% of weeds
Robotti	Laser scanner, camera, RTK-GPS	2.2 m/s	Mechanical	Diesel-hydraulic	Navigational accuracy <±2 cm
Model B Sprayer	4K machine vision camera	NA	Chemical	PTO	Weed identification accuracy 99%; 95% chemical reduction
Lee’s precision spraying robot for tomato [138]	Sharp GPB-2 board, AuxLUT card, RGB camera, CIO-DAS 1600 board	1.2 km/hr	Chemical	NA	Accuracy 73.1% for tomato, 68.8% for weeds
Astrand weed removal robot [139]	vision sensors and two cameras: front camera for row recognition and back for offset calculation	0.2 m/s	Mechanical	battery for indoor testing, combustion engine for application	Accuracy 77% for sugar beet, 87% for weeds
Bonirob weed removal bot [140]	four-channel JAI camera	NA	Mechanical	NA	93.86% weed control rate
Kolberg and Wiles’ steam weeding [141,142]	four-channel JAI camera	NA	0.8 m/s	Thermal	90% control rate of lambsquarters and redroot pigweed

Table 5. Overview of agricultural weeding robots: sensors, applications, and performance.

Robot	Sensors/Imaging	Speed/Coverage	Application Type	Power Source	Performance/Accuracy
Asterix [136]	Nvidia Jetson TK1, 4682 4MP sensor	0.8 m/s	Chemical	Hybrid (48 V DC, 4-stroke generator)	Drop-On-Demand (DOD) system has been proven to kill off 100% of weed, reducing herbicide usage 10 times
AgBotII’s multimodal weed control [137]	RGB camera, Novatel SPAN-IGM-A1 (RTK)	1.3 m/s	Machanical or chemical	Battery	Classification Accuracy: 92.3%
Laserweeder [37]	Nvidia GPUs, RGB cameras, GPS, Lidar	0.4 m/s	Laser	PTO-driven generator	Kills up to 99% of weeds
Robotti	Laser scanner, camera, RTK-GPS	2.2 m/s	Mechanical	Diesel-hydraulic	Navigational accuracy <±2 cm
Model B Sprayer	4K machine vision camera	NA	Chemical	PTO	Weed identification accuracy 99%; 95% chemical reduction
Lee’s precision spraying robot for tomato [138]	Sharp GPB-2 board, AuxLUT card, RGB camera, CIO-DAS 1600 board	1.2 km/hr	Chemical	NA	Accuracy 73.1% for tomato, 68.8% for weeds
Astrand weed removal robot [139]	vision sensors and two cameras: front camera for row recognition and back for offset calculation	0.2 m/s	Mechanical	battery for indoor testing, combustion engine for application	Accuracy 77% for sugar beet, 87% for weeds
Bonirob weed removal bot [140]	four-channel JAI camera	NA	Mechanical	NA	93.86% weed control rate
Kolberg and Wiles’ steam weeding [141,142]	four-channel JAI camera	NA	0.8 m/s	Thermal	90% control rate of lambsquarters and redroot pigweed

5. Intelligent Systems and Ecosystem Integration

5.1. Decision Support Systems (DSS)

Beyond energy-efficient practices, data-driven strategies are reshaping how weeds are monitored and managed across ecosystems. Data-driven decision making through big data is revolutionizing weed management [143]. Data processing, data interpretation, data sharing, and data acquisition are important procedures to detect, track, and manage invasive weeds in diverse landscapes [144]. Vulnerable regions can be accurately mapped by integrating environmental factors, species distribution data, and statistical modeling [145]. These predictions are further improved by machine learning algorithms, allowing for more focused weed control tactics [144,145,146]. For example, Early et al. (2016) suggested a global spatial forecast of weed invasions in the 21st century through his analysis of large-scale spatial data on factors influencing the introduction and spread of invasive alien species(IAS) [147]. Such predictive analytics, if powered by big data, will allow for more proactive intervention before weeds become widespread. A significant example is Ambrosia artemisiifolia L. (common ragweed), a major weed which attacks crops like sunflower, maize, sugar beet, and soybean, not only impacts on yield but also its extensive production of pollen causes serious allergic reactions in humans [148]. Big data-driven research can improve the understanding of such parasitic weeds leading to the discovery of more effective management approaches [149,150].

5.2. Blockchain for Traceability and Transparency in Sustainable Practices

One of those management approaches is blockchain technology, which reduces risk and costs in business networks by tracking both tangible and intangible assets in a shared, unchangeable ledger. It enables faster, more accurate information exchange by providing immediate, shared, and observable data stored securely [151]. Integrating blockchain with smart weeding technologies allows farmers to create an immutable record of crop cultivation processes. By eliminating the need for intermediaries, blockchain enables peer-to-peer transactions that are transparent, secure, and trust-based. This shift removes the reliance on central authorities, replacing trust in intermediaries with cryptography, and granting farmers access to premium markets [152]. As a result, blockchain can restore trust between producers and consumers, reducing transaction costs in the agri-food market and improving food quality and safety. Fraud and malfunctions can be quickly detected, and issues can be addressed with smart contracts [153,154]. Several smart farming models combine IoT and blockchain, such as the lightweight blockchain-based architecture for smart greenhouse farms proposed by Patil et al. (2017) [155]. In this model, IoT sensors manage a private local blockchain that is centrally controlled by the owner. Taiwan’s farmland irrigation associations (Lin et al., 2017) also use blockchain to better manage data and public interaction [156]. Blockchain technology securely records product information from its origin to retail, allowing stakeholders to verify data on product quality, such as DNA of livestock and pesticide residues in produce. Major companies like Walmart, Alibaba, and JD.com from China are already using blockchain for food traceability in their supply chains [152]. In August 2017, global food and FMCG suppliers like Walmart, Nestle, Dole, and Golden Food partnered with IBM to integrate blockchain into their supply chains, detecting misconduct more quickly [152].

Table 6. Examples of 5IR-based integrated technologies in precision agriculture.

Core IoT Technology	Methodology	Requirements	Reference
Hyperspectral imagery using UAVs	Classification algorithms are applied on oat field images to perform weed mapping.	Most drones are equipped with digital cameras. Hyperspectral cameras are considerably more expensive and precise camera calibration is necessary to obtain high-quality images.	[157]
UAV-based crop monitoring	A low-cost UAV is used to acquire multispectral images. The obtained images are later processed to extract NDVI orthomosaic images.	A drone, a Raspberry Pi microcontroller, and two digital cameras for the hardware. The commercial multispectral cameras can cost up to $5000, while the proposed setup only costs $250.	[158]
Improving path tracking precision of autonomous weeders using GNSS	Integrates visual navigation with satellite technology to reduce path deviation. Compared to the visual system alone, the new system reduces the side-to-side drifting by an additional 30.1%.	Navigation hardware devices, including GNSS antennas, receivers, and Ionospheric Pierce Point (IPC) systems, along with industrial-grade cameras for capturing high-resolution images required for visual navigation and processing units to run path-tracking algorithms and process sensor data have to be installed on weeder.	[159]
Blockchain integrated with IoT devices for weed detection	IOT devices retrieve data from sites and send them to aggregation and normalization nodes. Multiple blockchain nodes serve as distributed, tamper-resistant data storage sites.	Distributed blockchain nodes to form a decentralized network, a blockchain ledger to store hash data, and off-chain storage for large files connected to hash references on-chain.	[160]
Sustainable Farm Assistant Bot	Solor powered robot equipped with advanced sensors and an intelligent control system is used to both monitor the farm and perform labor-intensive tasks such as plowing, sowing, watering, and pesticide spraying.	An Arduino board, DC motor, rain sensor, water pump, IR sensor, soil moisture sensor, soil fertility sensor, and NodeMCU are currently installed on the prototype model.	[161]
ROW SLAM, autonomous cornweed-removing robot	Semantic segmentation results are integrated with tracking modules to build a map that contains location, orientation, and semantic information of corn stalks	Structure-from-motion (SfM) for segmentation, SLAM for more accurate tracking results than GPS, and a robot equipped with multiview cameras to obtain front, back, and side views.	[162]

showcased the integration of the aforementioned technologies in real life. However, while such technologies excel at traceability and data capture, the question remains: how effectively is this data being translated into tangible improvements on the farm?

6. Current Limitations, Market Dynamics, and Economic Considerations

The body of research on Smart Farming Technology (SFT) has expanded significantly; however, between 2014 and 2020, studies predominantly focused on data recording, with limited emphasis on translating these measurements into practical agricultural applications. This indicates a critical gap in leveraging recorded data for on-farm decision-making and efficiency improvements [163]. From an industrial perspective, the distribution of SFT categories is more balanced, with recording technologies and Farm Management Information Systems (FMIS) being the most prevalent in commercial products. In contrast, guidance technologies, having reached maturity, attract fewer new entrants. Furthermore, many companies conduct internal research and development (R&D) on advanced SFTs but refrain from publishing findings in academic journals to safeguard intellectual property [163].

6.1. Market Analysis for Smart Weeding Technologies

The available market analysis on smart agriculture is relatively scarce, particularly when it comes to reliable sources and comprehensive global coverage. McKinsey’s 2022 report [164] examines key agricultural economies, including China and India in Asia; Germany, France, Spain, and the Netherlands in Europe; Canada and the United States in North America; and Argentina and Brazil in South America. However, it is important to note that this discussion does not encompass other significant agricultural producers such as Japan, Russia, and Turkey, which are recognized for their extensive agricultural production. The following are the key information extracted from that report.

• Growth in AgroTech adoption remains limited, with only 4% of farmers planning to adopt precision-agriculture hardware, remote-sensing solutions, or sustainability-related technologies in the next two years. Figure 19 presents detailed data categorized by farm size.
• Among submarkets, precision agriculture products have the highest adoption rate with 39% for software and 15% for hardware, making it one of the most utilized technologies.
• Robotics and automation, which are relevant to precision weeding, have the lowest projected growth rate of 2.5%, probably due to high initial investment.

Figure 19. (a) Change in Agricultural Employment vs. GDP per Capita (1991–2017). [Arrows show countries’ shift from 1991 (base) to 2017 (arrowhead). Most arrows point southeast, indicating that rising wealth correlates with reduced agricultural employment.] [165] (b) Survey results on likelihood of adopting AgroTech solutions by farm size.

Figure 19. (a) Change in Agricultural Employment vs. GDP per Capita (1991–2017). [Arrows show countries’ shift from 1991 (base) to 2017 (arrowhead). Most arrows point southeast, indicating that rising wealth correlates with reduced agricultural employment.] [165] (b) Survey results on likelihood of adopting AgroTech solutions by farm size.

As precision agriculture advances, governments worldwide are increasingly investing in this sector to secure a competitive edge. For instance, the U.S. Department of Agriculture (USDA) already offers grants to support farmers and researchers developing innovative solutions to enhance agricultural efficiency and sustainability [166]. Venture capital investment in the AgroTech sector has been increasing as investors recognize the potential for high returns driven by technological advancements that enhance agricultural efficiency and sustainability. In addition to previously mentioned companies, FarmWise, established in 2016, has developed innovative precision weeding solutions, including the Titan, a self-propelled weed robot, and Vulcan, an intra-row precision weeder [167]. The company has undergone a total of eight funding rounds, securing more than $140 million in investments over the years [168]. Public-private partnerships (PPPs) play a critical role in advancing precision weed removal technologies by facilitating collaboration between research institutions, government bodies, and private enterprises. These partnerships help bridge funding gaps, share risks, and accelerate the deployment of innovative solutions aimed at improving agricultural efficiency.

6.2. Economic Impact at Local and Global Level

The economic impact brought by revolution of smart weeding technologies can have both locally (at the farm and community level) and globally (across broader agricultural markets and supply chain). By fostering the development of agri-tech industries, rural communities can experience economic growth, while simultaneously attracting skilled workers, thereby promoting education, and encouraging investments in local infrastructure. The digitalization of agriculture can offer new avenues for growth, reducing the rural-urban migration trend and improving the standard of living in rural communities. Targeted weed removal minimizes herbicide use, contributing to healthier soil and water systems, thereby promoting long-term environmental sustainability in rural areas.

Plus, improved efficiency and consistency in local production can enhance a country’s food security and reduce vulnerability to global market fluctuations. This can alter international trade dynamics, potentially decreasing the volume of outsourced more expensive agricultural products in favor of domestically produced goods. As technology becomes more widespread, costs may decrease, making the smart tech more accessible to a broader range of farmers. Increased production efficiency may help stabilize prices and reduce vulnerability to global supply chain disruptions, benefiting both local economies and international trade dynamics.

Counterintuitively, by aggressively targeting weeds, these technologies might inadvertently reduce plant diversity. Some weeds support beneficial insects and pollinators; their removal could alter local ecosystems and reduce natural pest control. Moreover, the high initial investment required to adopt robotic technology may lead to a divergence between large-scale and small-scale farming enterprises. As technological advancements progress, larger farms with greater financial resources are more likely to incorporate these innovations, while smaller farms may struggle to keep pace. This dynamic has the potential to exacerbate existing disparities, resulting in a new form of economic inequality and a shift toward consolidation, where agricultural production is increasingly dominated by large corporations. On a national level, these advancements might widen the gap between developed and underdeveloped nations; developed countries, with greater resources, can invest in these technologies, boosting productivity and economic growth. In contrast, underdeveloped nations, heavily reliant on agriculture, may struggle to adopt costly innovations, deepening economic disparities.

7. Recommendations for Enabling Smart Farming at Scale

To mitigate the aforementioned effect of “Tech-driven Consolidation”, governments should provide support or low-interest loans to help small farmers access advanced technologies. Plus, the small farmers themselves can form cooperatives to share the cost and benefits of precision technologies by investing in shared devices. Governments and trade organizations should reconsider currents agricultural policies affecting tariffs and subsidies. Overcoming these challenges will require targeted efforts in education, infrastructure, financial, and cooperative frameworks to facilitate the transition toward data-driven, technology-enhanced agriculture.

7.1. Infrastructure Investment

Limited internet connectivity in rural areas further restricts real-time data processing, delaying actionable insights. While 5G and future 6G networks may address this, the current infrastructure remains inadequate. Ensuring adequate mobile network coverage and access to digital tools, such as mobile phones, is essential for widespread technology adoption. Without these foundational elements, the impact of digital farming technologies remains limited [169]. Moreover, a lack of awareness and distrust in automation slows adoption among traditional farmers [170]. Another significant barrier is the high initial investment required for precision farming technologies. Larger farms, with greater access to capital, are more likely to integrate advanced solutions, while smaller farms struggle with affordability [171,172]. Technologies such as dead reckoning systems, which use previous positioning data and in-field markers to estimate movement, can reduce costs for small, regularly shaped fields.

7.2. Policy Considerations

With increased digitization, cybersecurity risks pose a significant challenge to the adoption of smart farming technologies. Threats such as cyber-attacks, data breaches, and privacy violations endanger sensitive agricultural data during storage, transfer, and access. These risks can create distrust among farmers and act as a barrier to adopting digital agricultural solutions [169]. Plus, potential biases in algorithms, that inform farming decisions, needed to be addressed to ensure fairness, transparency, and equitable outcomes for all stakeholders. Since all these technologies rely intensively on data, further research is needed to understand the motivations of transacting parties in providing accurate data. The cost of data collection, such as monitoring weed resistance in crops, can be expensive. Although sampling reduces costs, larger farms benefit more due to economies of scale, potentially increasing income disparities between small and large farms [152]. These policy and economic challenges highlight the critical areas that future research must address to fully realize the potential of smart farming technologies.

8. Strategic Directions for Future Research

As smart systems continue to evolve, agriculture is expected to transition from a labor-intensive industry to a highly automated system where farmers can oversee operations remotely from the convenience of their homes. In order to do so, plant detection systems must achieve precise identification of crops and weeds at various growth stages, as well as distinguish between different species. Integrated weed management is expected to become the dominant strategy, leveraging multiple control methods to address the limitations of single-method approaches. Furthermore, the adaptability and operational efficiency of weeding robots in complex agricultural environments must be enhanced. The integration of advanced sensor technologies will enable these robots to perform diverse functions, effectively simulating human capabilities in agricultural tasks [173].

The future of smart weeding technologies also lies in their integration with next-generation advancements. Quantum computing has the potential to revolutionize the speed and complexity of data analysis in agriculture, enabling real-time, highly accurate decision-making for weed detection [174] and crop management [175,176]. Additionally, genetic engineering could play a crucial role in developing plants that are more resistant to weeds or pests [29]. The convergence of these cutting-edge technologies will drive the next phase of precision agriculture, making farming more efficient, sustainable, and less dependent on human labor. The development and refinement of smart weeding technologies will require global collaboration across research institutions, technology giants, and farming communities. By sharing knowledge, data, and best practices, stakeholders can accelerate innovation and ensure that these technologies are tailored to the diverse needs of farmers worldwide. International partnerships and open-source platforms could help democratize access to these advanced tools, fostering global agricultural sustainability.

9. Conclusions

Smart precision weeding supported by 5IR technologies symbolizes a substantial innovation toward intelligent, sustainable, and scalable agricultural systems. The integration of AI, robotics, IoT, and intelligent sensing enables real-time smart weed control that minimizes environmental impact and reduces chemical dependency. Despite its enormous potential, the wide-spread adoption of these technologies is impeded by high implementation costs, infrastructure limitations, and globalized farmer-centric design. In order to address these practical gaps, there is a necessity for multi-stakeholder collaboration, encouraging policy frameworks, and focused investments in education, training, and infrastructure. This review explains a global shift in the dynamics of agricultural power, where technologically advanced economies may contribute to better infrastructure and frameworks, but reasonable access remains critical to ensuring global food security. Future research in the AgTech domain should focus on improving the interoperability between smart farming systems, improving AI models, and improving the availability of open source, low-cost precision tools. In addition to innovation, factors such as inclusive strategies that empower small economies in fostering resilience across agricultural ecosystems are responsible for the success of smart precision farming in the era of 5IR.