Search Results (250)

This paper provides a comprehensive review of deep learning in agricultural visual perception. Climate change and land degradation demand a shift from experience-driven to data-driven intelligent agriculture. Traditional manual inspections remain subjective and unscalable. Computer vision offers a non-invasive solution for precision crop and livestock management, while challenges also exist. Its application in actual agricultural scenarios faces specific challenges such as severe occlusion, drastic changes in lighting, and non-rigid deformation of biological targets. To systematically summarize how these perception bottlenecks are being resolved, this review explores deep learning architectures featuring spatial extraction and spatio-temporal modeling. This comprehensive review first constructs a progressive analytical framework from low-level data augmentation and static spatial cognition to high-level dynamic spatio-temporal reasoning, and then systematically classifies existing literature. In-depth analysis reveals that these three core challenges can be computationally resolved using three unified technologies. Domain drift and label scarcity, rather than baseline accuracy, are the main obstacles to practical deployment. We further identified four priority research directions, i.e., architectural efficiency, data-level annotation efficiency, deployment-level privacy and simulation, and trust-level security, to guide future research. Full article

Background: Breast cancer brain metastases (BCBMs) exhibit notable receptor discordance between primary tumors and metastases, yet obtaining intracranial biopsies is rarely practical. We created an interpretable deep learning radiogenomic model to predict estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) status from MRI scans in BCBM cases. Methods: A total of 241 post-contrast T1-weighted MRIs from 142 patients were analyzed. We developed a mask-free 3D Residual Neural Network (ResNet) ensemble trained directly on cropped, bias-corrected brain images, with comprehensive 3D geometric and intensity augmentations. The model was optimized in a multi-label setting using Asymmetric Focal Loss. Hyperparameters were fine-tuned via Bayesian optimization, and class probabilities were calibrated. Additionally, Integrated Gradients (IG) offered voxel-level saliency maps. Results: Our ResNet ensemble model achieved a micro-averaged AUROC of 0.74 and a macro-averaged AUROC of 0.67. Receptor-specific AUROCs were 0.57 for ER, 0.79 for PR, and 0.64 for HER2. The F1-scores were 0.56, 0.62, and 0.88 for ER, PR, and HER2, respectively. The held-out cohort contained only four HER2-negative patients, so threshold-dependent HER2 metrics are strongly prevalence-driven and AUROC is the more appropriate summary. Qualitatively inspected saliency maps were concentrated on enhancing metastases with limited background attribution. Conclusions: This proof-of-concept study shows that a segmentation-free, interpretable 3D Convolutional Neural Network (CNN) can be trained to capture receptor-associated patterns in post-contrast T1-weighted MRI of BCBMs. Accuracy was modest relative to published radiomics models, and the mask-free, single-sequence design is offered as a methodological contribution rather than a performance gain. These findings are preliminary, do not establish clinical utility, and require validation in larger, multi-center cohorts. Full article

Hidden adulteration of stale berries in dried goji berry batches is difficult to detect by manual inspection or batch-level quality assessment. This study developed a high-throughput method for kernel-level spoilage adulteration quantification in dried goji berries. It addressed three practical challenges in the image processing of densely arranged dried-fruits, including scalable label generation for deep learning segmentation without pixel-level manual annotation, separation of densely touching small berries, and full-size quality level distribution map reconstruction. SAM-assisted pseudo-label generation combined with multi-scale image cropping was used to overcome the limitation of manual pixel-level annotation, while YOLO-based instance segmentation was further employed for efficient berry localization in dense scenes. The freshness labels of segmented single berries were assigned by a statistical RGB-HSV grading rule. Specifically, adaptive multi-scale image cropping for segmentation was applied to improve local separability of berries under dense adhesion and occlusion conditions. The crop-level segmentation and grading outputs were subsequently reconstructed into the original image coordinate system to generate complete quality distribution maps. Results showed that YOLO models trained based on the pseudo-labels achieved a precision of 0.953, a recall of 0.951, an mAP₅₀ of 0.960, and an mAP_50-95 of 0.846. The full-size grading map reconstruction method produced a mean duplicate-suppression rate of 4.31%. In the full freshness-grading test dataset, 4850 berries were detected, including 449 stale berries. The mean absolute counting error was 1.61%. The proposed framework reduces manual annotation requirements while enabling berry-level freshness classification and quantitative stale-berry proportion estimation, providing objective information for dried fruit quality screening and adulteration control. Full article

Vineyard monitoring requires efficient methods for detecting plant stress and disease while limiting flight time, data volume, and labour effort. This paper presents a satellite-enabled two-tier UAV workflow for semi-automated vineyard inspection. The proposed approach combines high-altitude multispectral scanning, NDVI-based point-of-interest identification, adaptive flight path planning, low-altitude RGB inspection, and downstream vision-based disease analysis. Processing tasks are offloaded to a remote server accessed through an emulated Low Earth Orbit satellite communication environment, allowing the UAV-side system to remain lightweight while receiving multispectral analysis results during the mission. A simulation framework was developed to evaluate mission behaviour under controlled and repeatable conditions, using both pseudo-random point generation and real multispectral vineyard images processed through the satellite emulation testbed. A flight with a real drone was also conducted to validate adaptive flight optimisation. Experimental results focus on the impact of path-adaptation strategies and communication bandwidth on mission efficiency. The results show that route optimisation can reduce mission time by up to $15\%$ when new low-altitude waypoints emerge, while bandwidth bottlenecks affect performance once image transmission can no longer keep pace with acquisition. The findings highlight the need to consider sensing, communication, and mission planning jointly in adaptive UAV-based crop monitoring. Full article

The textile and clothing value chain faces increasing pressure to improve reuse and recycling rates, particularly in post-consumer scenarios, where garments must be rapidly assessed, classified, and routed toward appropriate end-of-life pathways. Post-consumer garment sorting must preserve reusable items while directing non-reusable textiles toward appropriate recycling or inspection pathways. This article presents a two-stage image-only decision-support framework that combines YOLO-based image classification, a locally executed vision–language model (VLM), two ConvNeXt-Tiny textile classifiers, and deterministic routing rules. In Stage 1, YOLO classifiers estimate garment type and dominant color, while Qwen2.5-VL-3B-Instruct VLM assesses visible stains, holes, pilling or lint, tags, dirt or discoloration, intentional distressing, condition, and supporting evidence. The backend validates these outputs and applies explicit precedence and uncertainty rules to assign categories A (resale), B (donation/reuse), C (recycling-oriented pre-sorting), or D (critical review). Stage 2 is triggered only for C/D garments and aggregates predictions from multiple RGB crops to estimate broad material-family hints and visible fabric structure before proposing an initial route, container, color group, recycling mechanism, and validation requirement. The YOLO garment-type classifier achieved 78.6% top-1 and 99.2% top-5 accuracy on the test set. The ConvNeXt-Tiny fabric-structure classifier achieved 78.55% accuracy and 78.64% macro-F1, whereas the material-family classifier achieved 56.39% accuracy and 55.83% macro-F1. In a controlled Stage 1 pilot test, binary reuse-oriented versus additional-processing routing achieved 80.0% accuracy, 75.0% precision, 75.0% recall, and an F1-score of 0.75. A Stage 2 end-to-end pilot test achieved 66.7% correctly recommended final routes, with macro-F1 of 0.767. These results provide evidence that complementary models and explicit validation rules can support preliminary explainable garment triage. However, RGB imagery cannot confirm exact fiber composition, blend percentages, or chemical contamination. Full article

Mulberry (Morus alba L.) is an important perennial woody crop for sericulture, medicinal resource development, and ecological conservation. Accurate disease and pest detection in natural mulberry orchards remains challenging due to dense foliage, severe occlusion, varying illumination, and the presence of small lesions with weak texture features. To address these problems, this study proposes CLH-DETR, a lightweight detection framework improved from the Real-Time Detection Transformer (RT-DETR) for mulberry disease and pest detection under natural field conditions. The proposed model introduces four targeted improvements: Cross-Stage Partial Progressive Multi-Scale Feature Aggregation (CSP-PMSFA) is designed to strengthen multi-scale lesion feature extraction while reducing redundant computation, the Lesion Detail Enhancement Block (LDEB) is designed to enhance weak lesion details and responses related to lesion boundaries, Haar Wavelet Downsampling (HWD) is adopted to preserve texture and structural information during feature downsampling, and Focaler-ShapeIoU is introduced to improve bounding-box regression for irregular disease regions. Experiments on the Mulberry Disease Dataset show that CLH-DETR achieves 80.1% precision, 75.3% $mA{P}_{50}$, and 55.9% $mA{P}_{50:95}$, improving the RT-DETR baseline by 3.9, 2.3, and 2.2 percentage points, respectively. Meanwhile, the number of parameters decreases from 19.9 M to 13.4 M, and the computational cost is reduced from 57.1 G to 46.4 G FLOPs. Compared with representative YOLO- and DETR-based detectors, CLH-DETR provides a favorable balance between detection accuracy and model complexity. When deployed on an iPhone 16 Pro using CoreML with FP16 precision, the model achieved an average latency of 16.5 ms per image, corresponding to 60.6 FPS. This result indicates its potential for edge-assisted mulberry disease inspection under the tested hardware conditions. Full article

The angular distribution of surface-strands in oriented strand board (OSB) is closely associated with board mechanical properties and mat formation quality. By acquiring surface images through vision sensing and combining them with deep learning-based segmentation, the angle classes of OSB surface-strands can be segmented and statistically analyzed automatically. However, OSB surface images contain complex strand textures, blurred boundaries, local adhesion between adjacent strands, and subtle differences among neighboring angle classes. To address these challenges, this study proposes a direction-aware dual-branch semantic segmentation network (DiBiNet) for pixel-level segmentation of surface-strand angle classes. OSB surface images were collected using a Hikrobot MV-CE120-10UC color industrial camera, and an 11-class dataset was constructed, including the background and ten angle classes from 0° to 90°. The samples were cropped to 512 × 512 pixels, and an improved angle-semantic-consistent Copy–Paste strategy was used to augment the training data. DiBiNet enhances directional feature representation through a Directional Strip Detail Enhancement Module, improves semantic feature modeling by combining MobileNetV3-Small with a DS-MobileViT Block, and fuses the two branches through a Bilateral Gated Fusion Module. Considering the continuity among angle classes, Direction Vector Auxiliary Supervision is introduced to map discrete angle labels into continuous direction vectors, thereby improving discrimination among neighboring classes. Experiments on the self-constructed dataset show that DiBiNet achieves a mean Intersection over Union (mIoU) of 0.8532, an overall pixel accuracy (Acc) of 0.8823, and a Dice coefficient of 0.8623, outperforming several representative semantic segmentation models. After 8-bit integer (INT8) + 16-bit floating-point (FP16) mixed quantization, the model achieves a neural processing unit (NPU) inference speed of 34.0 frames per second (FPS) on the RK3588 platform, demonstrating its potential for vision-based sensing and edge AI inspection. Full article

Accurate identification of the hepatopancreas is essential for prawn quality assessment and automated seafood processing. This study presents an explainable deep learning framework for the binary classification of prawn images into “with hepatopancreas” and “no hepatopancreas” categories. A custom convolutional neural network (CNN) was developed using a dataset of 252 annotated images. To improve feature extraction from the limited dataset, an image preprocessing pipeline incorporating automatic contour-based cropping, contrast enhancement, and data augmentation was employed. The proposed model achieved a test accuracy of 97.37% and a calibrated full-dataset accuracy of 97.22%. Five-fold cross-validation yielded a mean accuracy of 96.83% ± 1.94%, indicating stable performance across different data partitions. Receiver operating characteristic analysis demonstrated satisfactory discriminative ability with AUC > 0.99. Gradient-weighted Class Activation Mapping (Grad-CAM) showed that the model primarily focused on biologically relevant hepatopancreas regions, improving the interpretability of the classification results. A graphical user interface was developed to enable rapid image analysis with visual feedback. Although further validation using larger and more diverse datasets is required, the proposed framework demonstrates the potential of explainable deep learning for automated prawn quality assessment and provides a practical foundation for intelligent seafood inspection applications. Full article

Reference labels for crop-type mapping are frequently coarse, and administrative land-use registries such as Japan’s eMAF (electronic Map of Agriculture and Forestry) database routinely group agronomically distinct crops under broad, ambiguous categories. This study addresses that problem for the Nakasatsunai region of Hokkaido, Japan, by combining Sentinel-1 synthetic aperture radar (SAR) time series with field-level optical vegetation-index analysis in a modular processing pipeline. The principal novelty of the work is not the pipeline architecture alone but a three-step, Normalized Difference Vegetation Index (NDVI)-driven label-refinement procedure—automatic removal of non-growing or low-amplitude field samples, Euclidean k-means subclass discovery within each coarse label, and trajectory-based label correction—that converts noisy nine-class eMAF labels into a more reliable training set prior to classifier training. The feature set combines the Radar Vegetation Index (RVI), VV and VH backscatter, the γVH/γVV polarization ratio, and NDVI, together with temporal-shape descriptors (phenological timing, peak magnitude, amplitude, maximum slope, and area under the curve) derived from monthly growth trajectories over the 2018 growing season. A Random Forest classifier, together with a gradient-boosting comparator, is evaluated before and after preprocessing under stratified k-fold cross-validation. Across n = 1208 field samples spanning the nine eMAF classes, classification accuracy improved from an overall accuracy of 71.8% on the raw labels to 82.6% after the three-step refinement; Cohen’s kappa increased from 0.63 to 0.77. Correlation analysis indicates that γVH/γVV tracks field-level NDVI more consistently (mean Pearson r = 0.68) than RVI does (mean Pearson r = 0.43) across the eight classes with sufficient samples, motivating its use as a SAR-only phenological proxy; this comparison is extended to the polarimetric PRVI, DPSVI, and DpRVI indices in the discussion. The underlying 80–90% label-accuracy estimate is derived from NDVI trajectory inspection rather than independent, field-surveyed ground truth, and a factorial ablation is used to characterize, to the extent the cross-validated evidence allows, how much of the reported accuracy gain is attributable to label-error correction as opposed to NDVI–SAR feature fusion; both this attribution and the label-accuracy estimate itself are identified as priorities for field validation in future work. The proposed framework is intended to convert coarse, noisy crop labels into a structured and reliable dataset while producing interpretable, field-level phenological insight for agricultural monitoring. Full article

With the rapid development of artificial intelligence, UAV remote sensing, and agricultural Internet of Things technologies, crop growth monitoring is evolving from manual inspection and single-source analysis toward intelligent decision-making based on multisource perception. However, existing methods still suffer from limited robustness under environmental variations, insufficient integration between UAV imagery and sparse ground sensor observations, and weak capability for transforming predictions into practical agricultural management recommendations. This study proposes a UAV–ground sensor collaborative lightweight framework for crop growth assessment and agricultural decision support. The proposed framework integrates UAV RGB and multispectral imagery with ground sensor observations through a region-level aerial–ground alignment mechanism and a sensor-guided attention fusion module, enabling environmental conditions to enhance visual feature interpretation. Furthermore, a fact-constrained decision module is developed to generate management recommendations based on crop status, environmental risks, and field information. Experimental results demonstrate that the proposed method achieves superior performance in crop growth classification and yield-trend prediction, reaching Accuracy, Precision, Recall, and F1-score values of $92.47\%$, $91.86\%$, $91.39\%$, and $91.62\%$, respectively, with an RMSE of $0.381$ and an $R^2$ of $0.902$. The lightweight framework requires only $6.18$M parameters and $0.91$G FLOPs, achieving $39.56$ ms inference latency and $25.28$ FPS on edge devices. The proposed framework also improves decision reliability, achieving an expert agreement rate of $89.34\%$ and a risk identification accuracy of $90.18\%$. Economic analysis indicates that the proposed framework reduces labor cost, water consumption, and fertilizer input by $49.7\%$, $26.7\%$, and $23.0\%$, respectively, while increasing net benefit by $46.1\%$ compared with conventional field management practices. These results demonstrate that the proposed method provides an accurate, interpretable, and deployable AI-driven solution for intelligent crop management in smallholder and medium-sized farming systems. Full article

Viral diseases are one of the most destructive threats to global agriculture. Among plant viruses, tobacco mosaic virus (TMV) is a highly contagious pathogen infecting numerous economically vital crops. With no curable treatments, the only strategy to mitigate virus spread is early detection and plant removal. The gold standard for TMV detection is reverse transcriptase-polymerase chain reaction (RT-PCR), followed by agarose gel electrophoresis. To reduce TMV diagnosis time and eliminate the requirement for expensive equipment, this study developed and optimized a one-step RT-PCR-coupled cysteamine-functionalized gold nanoparticle assay for the colorimetric determination of the virus. By integrating cDNA synthesis and PCR amplification into a single tube and analyzing the results using cysteamine- functionalized gold nanoparticles (Au@Cys), the diagnosis turnaround time was significantly reduced. Furthermore, this AuNPs-based colorimetric assay enabled straightforward visual inspection with the naked eye, eliminating the need for costly optical devices. Additionally, our regression equation linking RT-PCR product color values, extracted as mean A value based on the CIELAB color space (green-red axis), and viral load allows for the accurate quantification of TMV infection levels in field samples. Our research lays the groundwork for the further development of more cost-effective, quantitative and rapid plant virus diagnosis methods. Full article

The white peach scale, Pseudaulacaspis pentagona (Targioni-Tozzetti) (Hemiptera: Diaspididae), is a highly polyphagous insect pest of global phytosanitary relevance affecting fruit trees, perennial crops, and ornamental plants in temperate and subtropical regions. Its invasive success is largely driven by high thermal tolerance, a wide host range, and strong survival capacity under postharvest handling and cold-chain storage, which greatly enhances its potential for long-distance dispersal through international trade of fresh fruit and plant material. A systematic review and meta-synthesis of 105 scientific studies published between 1958 and 2026 identified key research areas, including geographic distribution and invasion pathways, host plants and varietal susceptibility, temperature-dependent life history, agricultural impact and quarantine risk, chemical and biological control strategies, and the development of phenology-based monitoring and prediction tools using thermal models and pheromones. Although recent official reports do not confirm the presence of P. pentagona in Ecuador, climate suitability modeling and evidence of long-term survival during cold storage indicate a high risk of introduction and establishment, particularly in inter-Andean valleys. Consequently, preventive phytosanitary surveillance and integrated pest management strategies tailored to Ecuadorian Andean fruit production are proposed, emphasizing phenological monitoring, nursery and agro-urban inspections, and the conservation of natural enemies as a foundation for sustainable management of this emerging pest. Full article

Maize (Zea mays L.) is a major staple crop vulnerable to post-harvest deterioration caused by fungal infection and mycotoxin contamination. Visual defects such as discoloration, breakage, insect damage, and mold growth are commonly associated with reduced grain quality and increased contamination risk. This article presents a publicly available RGB image dataset of white maize kernels deposited in Harvard Dataverse for the development of computer vision models for automated grain quality assessment. The dataset contains 5143 high-resolution RGB images acquired using Samsung Galaxy A12 and Samsung Galaxy A54 smartphone cameras under semi-controlled imaging conditions. Images contain either single or multiple kernels and were annotated at the instance level using the YOLO format, resulting in 13,533 labeled kernel instances. Annotations were assigned by experts experienced in mycotoxin-related grain quality inspection. Labels are based solely on visual surface characteristics and do not represent direct chemical measurements of aflatoxins, fumonisins, or other mycotoxins. The dataset provides a practical resource for developing and evaluating machine learning models for maize kernel defect detection, quality screening, and risk-oriented grain inspection applications. Full article

In Chinese solar greenhouses, row spacing of 40–80 cm and persistent 80–95% RH impose two fundamental constraints: conventional wheeled platforms exceed the inter-row turning radius, and unprotected electronics show <200 h MTBF under continuous humidity. These constraints define the design envelope for integrated inspection–detection–spraying robots. This review synthesizes 139 studies (2020–2026) from China, Europe, and North America. Beyond narrative synthesis, we contribute two quantitative tools: (1) a latency budget analysis decomposing the perception-to-spray pipeline—perception (30–50 ms), inference (50–140 ms), decision (5–10 ms), actuation (10–20 ms), and nozzle response (5–15 ms)—revealing a 100–235 ms total delay, translating to 3.0–7.1 cm spray-target displacement at 0.3 m/s; and (2) a standardized validation protocol with eight metrics for multi-site field trials. Four technical gaps persist. Rail-mounted platforms dominate, but autonomous systems lack validated high-humidity reliability. YOLO-based models achieve 93–94% mAP at 10–20 FPS on edge devices, yet cross-crop generalization is unproven, while Transformers are too slow (2–5 FPS) for real-time deployment. PWM-controlled spraying saves 20–40% of pesticide in controlled trials, but drift control and deposition uniformity under real canopies are rarely quantified with sufficient engineering detail. The quantified latency causes systematic spray-target misalignment that open-loop controllers cannot correct. We propose five future directions with quantifiable targets: low-cost modular platforms (50,000–80,000 RMB/unit vs. current 180,000–250,000 RMB); edge-optimized perception (>15 FPS on <15 W hardware); closed-loop latency compensation (<2 cm displacement); adoption of the proposed protocol for cross-system comparison; and multi-robot collaboration for >1 ha greenhouses. Full article

Soybean variety recognition supports germplasm management and intelligent agricultural inspection, but remains challenging because varieties often show subtle inter-class differences, large intra-class imaging variation, and imbalanced samples. This study adopts a cascaded pipeline that combines YOLO-based object localization, manual verification and cropping, and fine-grained classification, allowing the classifier to learn from standardized cropped soybean seed images rather than original whole images. To improve Swin Transformer for this task, we propose Swin-Diff-DCA, which introduces Dynamic Channel Aggregation (DCA) in Stage 3 for middle-level local feature reuse and a differential attention branch in Stage-4 window attention for deep discriminative enhancement. On a dataset containing 25 soybean varieties and 1511 test images, Swin-Diff-DCA achieves average results of 87.18% Accuracy, 82.33% Macro-F1, and 87.44% Weighted-F1 across three random seeds under the current split and evaluation protocol. It outperforms RegNet, ResNet-50, ViT, the Swin Transformer baseline, and single-module variants. The results show that combining middle-level feature reuse with deep differential enhancement improves cropped soybean variety classification, while low-sample and visually similar classes remain the main sources of misclassification. Full article