Scholarlink

This study proposes real-time disease diagnostic techniques for the farm by using deep learning algorithms and edge IoT devices. Experiment the accuracy, precision, inference time, and edge deployment of plant leaf disease classification using YOLOv5, YOLOv4, and Faster R-CNN models. Three senior and junior horticulturists considered and labeled 50,000 photos of healthy and sick leaves from 14 crop species and 138 categories for training and testing. The dataset preprocesses were carried out using YOLO format for scaling, augmentation and annotation. Real-time performance for three deep learning models—from YOLOv5, YOLOv4, and Faster R-CNN were tested on NVIDIA Jetson Nano and Xavier NX. Land sickness Detections in Agriculture have historically utilized human inspectors or specific to SVM and KNN traits. These approaches are too slow and unscalable for timely detection, The lightweight YOLOv5 models extract more accurate features using IASM, GhostNet, BiFPN architectures, and EIOU loss functions. Performance is measured via model size, inference time, mAP, and F1-score, comparing YOLOv4 and Faster R-CNN. Every model was put through IoT testing and edge deployment optimisation. LMSYOLOv5 is an order of magnitude faster, more accurate and lighter. Not been beaten in edge resource constrained conditions by YOLOv4 and Faster R-CNN (92.57% (F1-score), 92.65% (mAP)). Even though Faster R-CNN was more accurate under static conditions, YOLOv5 achieved better times for the real-time inference on edge devices task. Inference time: 18 ms, F1-score: 92.57% and mAP:  92.65% for YOLOv4 The optimised YOLOv5 does: 226 | R-CNN v2 faster: mAP: 88.4%, F1-score: 89.2%, inference time: 38 ms; results: 90.8%, 90.2%, 112.0 The architecture was modified based on the original YOLOv5 to achieve the optimal algorithm for the real-time plant disease diagnosis method (implemented on edge IoT sensors). Due to low-latency, computational efficiency, and a high detection accuracy, it is advantageous for precision agriculture. Next, we will use our insights to look at smart irrigation and agricultural drones.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Edge IoT–Based Detection of Crop Diseases: A Comparative Study of Deep Learning Algorithms for Real- Time Accuracy and Efficiency

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. Mahaveerakannan,
  • S. Saraswathi,
  • R. Balamanigandan

摘要

This study proposes real-time disease diagnostic techniques for the farm by using deep learning algorithms and edge IoT devices. Experiment the accuracy, precision, inference time, and edge deployment of plant leaf disease classification using YOLOv5, YOLOv4, and Faster R-CNN models. Three senior and junior horticulturists considered and labeled 50,000 photos of healthy and sick leaves from 14 crop species and 138 categories for training and testing. The dataset preprocesses were carried out using YOLO format for scaling, augmentation and annotation. Real-time performance for three deep learning models—from YOLOv5, YOLOv4, and Faster R-CNN were tested on NVIDIA Jetson Nano and Xavier NX. Land sickness Detections in Agriculture have historically utilized human inspectors or specific to SVM and KNN traits. These approaches are too slow and unscalable for timely detection, The lightweight YOLOv5 models extract more accurate features using IASM, GhostNet, BiFPN architectures, and EIOU loss functions. Performance is measured via model size, inference time, mAP, and F1-score, comparing YOLOv4 and Faster R-CNN. Every model was put through IoT testing and edge deployment optimisation. LMSYOLOv5 is an order of magnitude faster, more accurate and lighter. Not been beaten in edge resource constrained conditions by YOLOv4 and Faster R-CNN (92.57% (F1-score), 92.65% (mAP)). Even though Faster R-CNN was more accurate under static conditions, YOLOv5 achieved better times for the real-time inference on edge devices task. Inference time: 18 ms, F1-score: 92.57% and mAP:  92.65% for YOLOv4 The optimised YOLOv5 does: 226 | R-CNN v2 faster: mAP: 88.4%, F1-score: 89.2%, inference time: 38 ms; results: 90.8%, 90.2%, 112.0 The architecture was modified based on the original YOLOv5 to achieve the optimal algorithm for the real-time plant disease diagnosis method (implemented on edge IoT sensors). Due to low-latency, computational efficiency, and a high detection accuracy, it is advantageous for precision agriculture. Next, we will use our insights to look at smart irrigation and agricultural drones.