Towards Artificial Intelligence Applications in Precision and Sustainable Agriculture

Agriculture is the backbone of many economies across the globe. More than one billion people today make living from agriculture [1]. Agricultural development is among the most effective strategies for eradicating extreme poverty and feeding a projected population of ten billion people by 2050 [2,3]. In order to meet the world’s demand for food and other agricultural commodities while utilizing the limited amount of land available, crop production must be boosted during the coming years [4,5]. Studies have indicated that the growing impacts of climate change have complicated attempts to raise overall crop yield due to rising mean temperatures and increases in the frequency of extreme weather events [6,7,8,9]. Consequently, farmers must devise farming solutions to tackle the challenges of augmenting crop productivity and guaranteeing food security considering the escalating food demands. Artificial intelligence (AI), a machine-based system involving machine learning, deep learning, and natural language processing, has emerged as a powerful tool for performing complex tasks that historically required human intelligence and has played an important role in revolutionizing precision agriculture [10,11,12,13].

Precision agriculture, a key application of AI in smart farming, involves the utilization of sensors, drones, and satellite imagery to monitor crop health, soil conditions, and weather patterns. As agriculture globally faces the challenges of a growing population and changing climate conditions, the integration of AI into farming techniques can revolutionize traditional methods behind intelligent and data-driven systems and hence improve sustainability, productivity, and resilience in the agricultural sector. Such a data-driven approach allows farmers to precisely administer water for irrigation, fertilizers, and pesticides where and when they are needed, therefore minimizing environmental impacts. Recent advances in AI have witnessed an extensive number of applications in precision farming. For instance, studies coupling AI with machine learning, which is able to process vast amounts of data from various sources (e.g., satellite imagery and sensors), have been conducted to predict crop health, disease outbreaks, potential threats, and yield potential through image recognition and predictive modeling [14,15,16,17,18,19,20,21]. This predictive power enables early intervention, reducing crop losses and improving overall productivity. AI-powered autonomous vehicles and robotic systems can also contribute to tasks, such as transplanting, harvesting, and weeding, thus reducing labor costs and increasing overall efficiency [22,23,24,25,26].

This Special Issue (SI) of Agronomy, entitled “Application of Artificial Intelligence in Agriculture: Cultivation, Management and Harvest”, features five unique articles that focus on applying AI-based algorithms to solve problems in smart and precision farming. Methodologically speaking, these research contributions comprise studies into both experiments [27,28,29,30] and yield modelling [31] at multiple spatial scales, spanning from the location-within-field scale up to regional scales. As such, the authors offer new insights about farming technologies currently applied in China and Vietnam, emphasizing the three key areas of small target detection, digital image processing, and yield modeling. In particular, Kong et al. [27] introduces a cost-effective approach for automated fruit-picking using monocular images instead of expensive datasets acquired from depth-sensing cameras or LiDAR. The approach incorporates an advanced YOLOv8s detection package, leveraging the BiFormer block to attain enhanced accuracy in detecting small targets. Furthermore, it also introduces a fused depth estimation technique, integrating high- and low-resolution depth information for comprehensive and structured depth results. The experimental results with citrus as the target demonstrate an improved YOLOv8s network mAP of 88.5% and a recognition accuracy of 94.7%. The proposed method demonstrates excellent potential for enhancing fruit-picking automation efficiency.

To address the challenge of accurate pear sorting, Xie et al. [29] uses an extremely compressed lightweight model for pear object detection (ECLPOD) based on YOLOv7. This model addresses the problem of pear occlusion, as well as the limited computational resources available on embedded devices. The proposed system includes a hierarchical interactive shrinking network (HISNet) for efficient feature extraction, a bulk feature pyramid (BFP) module for enhanced pear contour information extraction, and an accuracy compensation strategy (ACS) to improve detection, especially for pear calyxes and stems. The ECLPOD achieves 90.1% precision and 85.2% mAP50 with only 0.58 million parameters and 1.3 GFLOPs. It outperforms YOLOv7 and other methods, providing an optimal trade-off between accuracy and complexity for embedded device deployment in pear sorting. To solve issues with jujube fruit identification, such as maturity phases and environmental variables, Xu et al. [30] introduced the YOLO-Jujube technique for target detection. By integrating networks, including a complete intersection-over-union (CIoU) loss, cross-stage boundary (CBS), spatial pyramid pooling fusion (SPPF), residual convolutional block (RCC), max pooling (Maxpool), CSPDarknet53-C3 (C3), and path aggregation network (PANet), the method provides automatic detection for ripeness fruit inspection and outperforms YOLOv3-tiny, YOLOv4-tiny, YOLOv5s, and YOLOv7-tiny with recorded params of 5.2 m, GFLOPs of 11.7, AP of 88.8%, and fps speed of 245, respectively.

To tackle the issue of mechanized harvesting for hydroponic lettuces in plant factories, Ma et al. [28] provides an image processing technique to estimate the height of lettuce and the degree of its expansion based on upper contour features. Following contour extraction and segmentation, the method uses the maximum ordinate for lettuce height and fits the upper contour to a quadratic function for leaf expansion size. The findings show a maximal relative error of 5.6% for lettuce height and 8.6% for leaf expansion size measurements. By intelligently adjusting grabbing parameters based on lettuce height and leaves’ expansion size measurements, the harvesting success rates exceeded 90%, with specific injured leaf areas identified for the left, middle, and right lettuces in each image. Lastly, Pham et al. [31] analyze how soybeans (Glycine max. L. Merr.) react to the effects of climate change using the SIMPLE crop model, concentrating on Vietnam’s Mekong Delta. Two experiments conducted in different seasons provided data for the model input and assessment, emphasizing the effects of temperature and CO₂ on soybean biomass and yield. Results revealed the model’s efficacy, with an overall relative root-mean-square error (RRMSE) of 9–10%. Soybean growth was adversely affected by drought stress, especially in the winter–spring of 2021. Furthermore, the higher temperatures accelerated biomass development but reduced yield, while the increasing CO₂ concentrations (600 ppm) enhancing biomass and production compared to 350 ppm. This research highlights the significant influence of impacts of climate change on soybean yields, which is effectively modeled using SIMPLE.

In summary, the use of AI in precision farming represents a radical turn toward knowledgeable decision making. Through technologies like machine learning and data analytics, farmers can optimize resource allocation, monitor crops in real time, facilitate efficient fruit picking and harvesting, and predict yield accurately. This synergy not only enhances productivity but also promotes sustainable practices, crucial for the future of agriculture. Despite challenges, the ongoing evolution of AI applications in precision agriculture promises resilient and efficient farming practices that can enable farmers to make informed decisions, allowing them to meet the growing demands of a dynamic and ever-changing agricultural environment.