Scholarlink

<p>Tomato is one of the most essential and consumable crops in the world. Leaf disease is the main factor affecting the quantity of crop production. So, it is essential to detect and classify these diseases accurately. Conventional models are unable to detect fine-grained spatial details and cannot selectively pay attention to infected areas of the leaf, which results in false positives and misclassification. Therefore, an optimized disease detection in tomato plants using attention-based neural networks and YOLOv7 for precision agriculture (TADNN-YOLOv7-POA) is proposed in this paper. The process is initiated with obtaining input images from a dataset of tomato leaf disease, which are pre-processed with Dynamic Gradient Filtering (DGF) to eliminate noise and correct distortions in images like rotation and skew. Then the pre-processed images are filtered through the Scale-Invariant Feature Transform (SIFT) to transform RGB images to the Hue, Intensity and Saturation (HIS) model. Following this conversion, segmentation is conducted through the use of edge and spot detection methods to precisely identify the areas of damage in the leaves. Detection of diseases in the tomato crops is then carried out by YOLOv7 augmented with the employment of Exponential Distance Transform (EDT) in extracting energy, contrast, and entropy features from the images. Moreover, YOLOv7 also applies Dual Attention in Attention Model (DAiAM) and Simple, Parameter-Free Attention Module (SimAM) to enhance the detection accuracy. For classification, an Attention-Guided Neural Network (AGNN) is employed to classify the diseases into classes like bacterial spot, early blight, late blight, leaf mold, septoria leaf spot, spider mites, target spot, yellow leaf curl virus, mosaic virus, healthy. To further improve categorization accurateness, the Parrot Optimizer Algorithm (POA) is employed to optimize the weight parameters of AGNN. The proposed TADNN-YOLOv7-POA model is implemented in Python, and its performance is analyzed based on accuracy, precision, F1-score, computational time (CT) and recall. The proposed TADNN-YOLOv7-POA framework achieves 99.5% accuracy and 1.2% error rate, demonstrating superior performance compared to existing techniques.</p>

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

Optimized disease recognition in tomato plants using attention- driven neural networks and YOLOv7 for precision agriculture

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Ravi Kumar Suggala,
  • G. Devi Sriya,
  • D. Lakshmi Sai Manojna,
  • D. Vijaya Laksmi,
  • G. Lalitha Devi,
  • B. Jassika

摘要

Tomato is one of the most essential and consumable crops in the world. Leaf disease is the main factor affecting the quantity of crop production. So, it is essential to detect and classify these diseases accurately. Conventional models are unable to detect fine-grained spatial details and cannot selectively pay attention to infected areas of the leaf, which results in false positives and misclassification. Therefore, an optimized disease detection in tomato plants using attention-based neural networks and YOLOv7 for precision agriculture (TADNN-YOLOv7-POA) is proposed in this paper. The process is initiated with obtaining input images from a dataset of tomato leaf disease, which are pre-processed with Dynamic Gradient Filtering (DGF) to eliminate noise and correct distortions in images like rotation and skew. Then the pre-processed images are filtered through the Scale-Invariant Feature Transform (SIFT) to transform RGB images to the Hue, Intensity and Saturation (HIS) model. Following this conversion, segmentation is conducted through the use of edge and spot detection methods to precisely identify the areas of damage in the leaves. Detection of diseases in the tomato crops is then carried out by YOLOv7 augmented with the employment of Exponential Distance Transform (EDT) in extracting energy, contrast, and entropy features from the images. Moreover, YOLOv7 also applies Dual Attention in Attention Model (DAiAM) and Simple, Parameter-Free Attention Module (SimAM) to enhance the detection accuracy. For classification, an Attention-Guided Neural Network (AGNN) is employed to classify the diseases into classes like bacterial spot, early blight, late blight, leaf mold, septoria leaf spot, spider mites, target spot, yellow leaf curl virus, mosaic virus, healthy. To further improve categorization accurateness, the Parrot Optimizer Algorithm (POA) is employed to optimize the weight parameters of AGNN. The proposed TADNN-YOLOv7-POA model is implemented in Python, and its performance is analyzed based on accuracy, precision, F1-score, computational time (CT) and recall. The proposed TADNN-YOLOv7-POA framework achieves 99.5% accuracy and 1.2% error rate, demonstrating superior performance compared to existing techniques.