IEEE Transactions on AgriFood Electronics最新文献

{"title":"Guest Editorial: Special Section on Autonomous Robotic Systems in Challenging Agricultural Settings","authors":"Ali Youssef;Congcong Sun;João Valente","doi":"10.1109/TAFE.2026.3738286","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3738286","url":null,"abstract":"","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"459-461"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11731457","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149131429","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Edge Deployable Weed Detection in UAV Imagery Using YOLOv12 With Scale and Motion Aware Data Augmentation","authors":"Murtiza Ali;Paul Abhisek Mondal;Karan Nathwani;Liang Zhao","doi":"10.1109/TAFE.2026.3690817","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3690817","url":null,"abstract":"Accurate and scalable weed detection is crucial for reducing herbicide use and enhancing the sustainability of soybean production, where input costs are significantly influenced by weed presence. Although uncrewed aerial vehicles (UAVs) equipped with red, green, and blue (RGB) cameras offer a practical and low-cost solution for large-area monitoring, the limited spectral richness of RGB imagery makes reliable weed–crop discrimination challenging. This study proposes an integrated agri-food electronic system for UAV-based weed detection in soybean fields, combining an enhanced you only look once (YOLOv12) detection framework, with a novel preprocessing and augmentation pipeline that introduces directional motion blur and altitude simulation via image segmentation and upsampling. These strategies enhance model robustness to scale variation, motion artifacts, and changes in UAV flight altitude, enabling accurate species-level weed identification under realistic field conditions. The modified YOLOv12 s model achieves 96.8% precision, 89.0% recall, and 94.8% mAP@50, while maintaining deployable inference capability on resource-constrained hardware. The trained model is deployed on a Raspberry Pi 4B edge platform through a custom Streamlit-based graphical user interface that performs fully on-device inference without cloud dependency, with general purpose input/output (GPIO)-driven light emitting diode (LED) and buzzer actuation providing real-time hardware-level feedback. On-device benchmarking confirms a 45.7% latency reduction via open neural network exchange (ONNX) Runtime (3282–1783 ms) with thermally stable operation (36 <inline-formula><tex-math>$^{circ }$</tex-math></inline-formula>C–43 <inline-formula><tex-math>$^{circ }$</tex-math></inline-formula>C) and power consumption within 5–7 W, validating field deployability. Detection outputs are converted into georeferenced shapefiles to support global positioning system-guided, site-specific herbicide application. The results demonstrate a compact, low-cost, and deployable sensing-to-actuation pipeline that enhances decision-making in precision agriculture while supporting sustainable soybean cultivation.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"520-529"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149130937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Vision-Based Perception Framework for Autonomous Robotic Date Palm Harvesting in Challenging Agricultural Environments","authors":"Ahlem Maghzaoui;Naeem Ayoub;Emna Aridhi;Sahbi Bahroun;Abdelkader Mami;Sadok Ben Yahia","doi":"10.1109/TAFE.2026.3721210","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3721210","url":null,"abstract":"Date palm harvesting is a labor-intensive and hazardous agricultural operation due to the height of palm trees, dense canopy structures, and the need for precise cutting of fruiting branches. Enabling autonomous harvesting in such environments requires robust, real-time perception systems capable of operating under occlusion, variable illumination, and embedded hardware constraints. This article presents a two-stage vision-based perception framework designed to support autonomous robotic date palm harvesting. The proposed system combines a lightweight You Only Look Once version 11 nano (YOLO11n) object detector for global localization of date palm fruit bunches with a customized YOLO11n-Pose network for fine-grained estimation of harvest-specific keypoints, including the optimal cutting point on each branch. The cascaded design reflects the operational workflow of a harvesting robot, enabling coarse target detection followed by precise anatomical localization. The framework was evaluated on a custom dataset collected from real Tunisian date palm orchards under challenging field conditions, including 5530 detection images with 24 559 annotated branch instances and a dedicated keypoint subset of 387 images containing 776 annotated branch instances and 3104 labeled keypoints. Experimental results demonstrate. For cutting-point localization, the proposed pose model achieved a test pose mAP@0.5 of 0.860 and a test pose mAP@0.5:0.95 of 0.775. A dedicated cutting-point evaluation yielded end-to-end PCK values of 0.588 at T = 0.10 and 0.676 at T = 0.20, with mean and median image-space localization errors of 41.39and 28.63 px, respectively. Both models were deployed on an NVIDIA Jetson Orin NX embedded platform. In the complete two-stage pipeline, the top-1 ROI mode required 65.16 ms per selected harvesting target, while processing all detected region of interests (ROIs) required 204.95 ms per image.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"498-507"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133431","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Geometry-Guided Plant Stand Count Estimation in Densely Populated Lentil Fields","authors":"Numair Nadeem;Muhammad Hamza Asad;Muhib Ullah;Abdul Bais","doi":"10.1109/TAFE.2026.3677798","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3677798","url":null,"abstract":"The agriculture industry faces increasing pressure to enhance food production while ensuring sustainability. One critical aspect of this challenge is to achieve consistency in crop establishment, which ensures optimal resource utilization and uniform crop maturity, ultimately increasing crop yield and reducing chemical inputs to the field. Traditional methods of quantifying the consistency of crop establishment, such as manual plant counting, are labor-intensive, error-prone, and inefficient for large-scale applications. Deep learning-based solutions are developed to automate these methods. However, their applicability is primarily limited to sparsely populated crops where individual plants are visually separable. In densely populated crops, such as lentils, minimal interplant spacing and rapid canopy closure cause seedlings to merge into blobs, where each blob contains multiple plants, and individual plant boundaries disappear. This article presents a novel pipeline to analyze the consistency of crop establishment for densely populated crops. Our proposed approach extracts crop blobs by segmenting crop pixels using a domain-robust semantic segmentation model based on a vision transformer. In parallel, convolutional neural network combined with image processing techniques is applied to identify crop rows, including their orientation and a row-width buffer, thereby enabling the assignment of blobs to their corresponding rows (on-row blobs). Geometric and contextual features are then extracted from the blobs, and a regressor predicts the number of plants within each on-row blob. Finally, statistical postprocessing integrates per-blob counts to quantify plant stand counts per-row and per-image for consistency of crop establishment analysis. The experimental results demonstrate the effectiveness of the method in predicting plant stand count to ensure consistency of crop establishment.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"570-586"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133467","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel Deep Learning Architectures for Soybean Kernel Defect Classification","authors":"Akshay Kumar;Niteen Sapkal;Ankur Miglani;Dheeraj Kumar;Pavan Kumar Kankar","doi":"10.1109/TAFE.2026.3688499","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3688499","url":null,"abstract":"Reliable identification of soybean kernel defects is essential for automated quality inspection in grain processing and export systems. However, current approaches face challenges when dealing with subtle defect patterns and heavily imbalanced class distributions. This study introduces a soybean kernel defect classification framework built on a high-resolution dataset of 9866 red, green, and blue images spanning eight defect categories, including underrepresented classes, such as purple-mottled and stained, stinkbug damage, and tear damage. To address performance degradation in rare categories, we propose a novel unified training pipeline that combines data augmentation and class-weighted focal loss. The proposed pipeline is evaluated using three deep learning architectures: EfficientNet-B0, ResNet-50, and the hybrid convolution–attention model CoAtNet0. Experimental results demonstrate that CoAtNet0 achieves the highest performance, with 83% accuracy and a 6%–9% improvement in F1-score over convolutional neural network baselines. gradient-weighted class activation mapping analysis further confirms improved localization of defect-relevant regions in CoAtNet-0, supporting stronger recognition of minority classes. Overall, this work highlights the benefit of combining hybrid modeling, imbalance-aware training, and visual interpretability to enable practical, scalable deployment for real-world soybean kernel defect inspection.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"601-608"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133525","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Extension of Beerline Cleaning Intervals by Electromagnetic Signals: Analysis and System Verification","authors":"Dagna I. Wójciak;Thomas Ballinger;Maksim Kuznetcov;Robin Guille;David Arnold;Louis Arnold;David Jenkins;Adam A. Stokes;David Clarke;Symon K. Podilchak","doi":"10.1109/TAFE.2026.3686307","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3686307","url":null,"abstract":"This article investigates the electromagnetic field generation and beerline hygiene maintenance performance of the Beerline Wizard RF Module 4 (BLW-4), very low-frequency (VLF) system by Beer Line Wizard and The University of Edinburgh, designed to extend intervals between manual beerline cleaning. In particular, the BLW-4 outputs a 1.6 to 6 kHz swept pulsewidth-modulated square wave, generating electromagnetic fields that couple into liquid-filled beverage lines. Simulations and anechoic chamber measurements showed that the fields are concentrated beneath the coil, with measurable coupling through conductive media such as beer or water (attenuation of approximately 0.1 dB/cm). Air-coupled fields from the module and coils were significantly lower than those coupled through filled lines. A 28-day forcing-test experiment also compared four identical beer taps, two being fitted with BLW-4 devices. Beer quality, assessed via post-incubation light absorbance at 600 nm (OD600), remained stable in treated lines, while untreated lines showed increased microbial growth indicators after four weeks. These results form a demonstration study which suggest that the BLW-4 VLF module can slow beerline fouling and help maintain beverage quality over extended cleaning intervals.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"648-658"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149131275","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Dairy Antibiotic Sensor: Actuation Through Microchip Electrophoresis and Detection by Lock-in Amplification and Fluorescence Spectroscopy","authors":"Camille A. Leclerc;Christopher M. Collier","doi":"10.1109/TAFE.2026.3675919","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3675919","url":null,"abstract":"This work develops a microfluidic antibiotic sensor based on microchip electrophoresis integrated with lock-in amplification. The microfluidic antibiotic sensor is for dairy antibiotic sensing and is compact, achieves high signal-to-noise ratios (SNR), has precise analyte actuation, uses pico-liter-scale sample volumes, and avoids sample pretreatment. We compare the SNR of ciprofloxacin hydrochloride monohydrate (CPFH) antibiotic fluorescence measurements of this sensor to a benchmark sensor without lock-in amplification. We provide the response curve of CPFH for the developed microfluidic antibiotic sensor. We provide the limit-of-detection of CPFH for this sensor through various methods and compare this limit of detection to other microchip electrophoresis and droplet-based dairy antibiotic sensors. Overall, the developed food contaminant sensor performs favorably in relation to other sensors.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"562-569"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149131724","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"2026 Index IEEE Transactions on AgriFood Electronics Vol. 4","authors":"","doi":"10.1109/TAFE.2026.3745930","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3745930","url":null,"abstract":"","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"790-804"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11743510","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149249527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Tiny Object Detection in Agriculture: Actual Field Dataset Expansion Technique and YOLOv11 Benchmark for Colorado Potato Beetle","authors":"Imran Hassan;Ahmad Al-Mallahi;Alimohammad Shirzadifar;Qamar Zaman;Khalil Al-Mughrabi","doi":"10.1109/TAFE.2026.3704973","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3704973","url":null,"abstract":"Real-time monitoring of insects in agricultural fields is challenged by their tiny target size, motion blur, and cluttered backgrounds. The Colorado potato beetle (CPB), a major threat to potato production, is particularly difficult to detect under operational-spraying conditions. This study introduces a field-realistic dataset and detection framework tailored for CPB monitoring using a sprayer-mounted camera operating at <inline-formula><tex-math>$8-11 text{km}cdot text{ h}^{-1}$</tex-math></inline-formula>. The high-definition images collected were expanded using a five-image crop strategy that preserved insect scale and introduced contextual diversity. From 668 original frames (1052 CPBs instances), 4837 cropped images containing 7286 annotated CPBs were generated. Subsequently, five you only look once v11 (YOLOv11) variants (n, s, m, l, and x) were trained using optimized hyperparameters identified through a grid search. The cross-validation results demonstrated strong performance across all models, whereas Tukey’s posthoc analysis found no significant difference among YOLOv11m, YOLOv11l, and YOLOv11x <inline-formula><tex-math>$(p &gt; 0.05)$</tex-math></inline-formula>. On the independent test dataset, these models achieved comparable accuracy (mAP@0.5:0.95 = 0.806–0.809) with precision and recall exceeding 0.90. Considering real-time deployment constraints, YOLOv11m provided the best accuracy–latency tradeoff for sprayer-based CPB monitoring when considering its inference time of 13.65 ms for an image. The results showed that systematic cropping of real imagery without synthetic augmentation could effectively mitigate tiny object distortion and would enable reliable detection under dynamic field conditions. Beyond CPB, this dataset design and evaluation framework provided a benchmark for small-pest detection and a pathway toward precision spraying systems that reduce chemical inputs while supporting sustainable pest management.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"668-679"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149130991","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FoodPAL 1.1: Novel Methods to Ensure Allergens Free Diet in Smart Healthcare Framework","authors":"Abdulrahman Alkinani;Laavanya Rachakonda;Saraju P. Mohanty;Elias Kougianos","doi":"10.1109/TAFE.2026.3727521","DOIUrl":"https://doi.org/10.1109/TAFE.2026.3727521","url":null,"abstract":"The immune system plays a vital role in defending the body against pathogens and foreign substances. However, food allergies disrupt its normal functioning, leading to immune system related disorders. During early childhood (ages one to three), children begin exploring their environment by placing various objects, including food, into their mouths. While this exploratory behavior is essential for development, it also poses significant risks, especially for children with food allergies. Therefore, increased supervision is crucial during this stage of growth. In response to this challenge, FoodPAL, a smart surveillance device, is introduced to monitor toddlers' food consumption. A hybrid object detection approach is implemented to enhance performance efficiency and optimize resource utilization. A premium dataset was evaluated across four different object detection models, and the most accurate model was selected. The overarching goal of this research is to protect children's immune systems by preventing exposure to allergenic substances.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"4 2","pages":"772-779"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149130997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}