ESM-YOLOv11: A lightweight deep learning framework for real-time peanut leaf spot disease detection and precision severity quantification in field conditions

文献类型： 外文期刊

第一作者： Zhang, Yapeng

作者： Zhang, Yapeng;Li, Shangzhou;Feng, Sifan;Zhang, Juanjuan;Guo, Wei;Liu, Juan;Cui, Yanan;Liu, Haijiao;Sun, Ziqi;Cheein, Fernando Auat

作者机构：

关键词： Peanut; Leaf spot; YOLOv11; Disease severity assessment

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 238 卷

页码：

收录情况： SCI

摘要： Leaf spot (LS) is a destructive foliar disease in peanuts, severely impacting yield. Compared to typical foliar diseases in crops, peanut leaves are smaller and denser, making the spots on the leaves challenging to detect and quantify accurately. Therefore, real-time precise monitoring and damage assessment of peanut LS disease are critical for agricultural decisions regarding the selection of disease-resistant varieties. In this context, the necessity of a lightweight model is paramount for effective deployment in real-world agricultural environments with hardware constraints. This study introduces a novel and advanced YOLOv11 model for rapid and accurate detection of peanut LS disease. First, an efficient multi-scale attention (EMA) module is incorporated into the backbone network to enhance feature extraction capabilities by enabling the model to focus on critical disease-related features across different spatial scales, thereby improving detection sensitivity for small and densely distributed lesions. Second, the Slim-neck module reconstructs the neck network by replacing original convolutions and kernel size 2 cross-stage component (C3k2) modules, reducing computational complexity and enhancing multi-scale feature fusion for improved speed and accuracy in real-time scenarios. Finally, the minimum point distance intersection over union (MPDIoU) loss function is employed instead of the traditional complete intersection over union (CIoU), which enhances the precision of bounding box regression by better penalizing misaligned or irregularly shaped predictions, ultimately contributing to more accurate localization of disease regions. Experimental results demonstrate that our model achieves notable lightweighting while maintaining superior detection performance: model parameters decreased by 3.87 %, floating-point operations per second (FLOPs) reduced by 7.94 %, detection precision reached 94.30 %, and the mean average precision (mAP) reached 96.90 %. The frame per second (FPS) on a central processing unit (CPU) improved to 26.6, representing a 10.37 % enhancement over the original model. Consequently, an intelligent peanut LS disease assessment system is implemented, enabling precise calculation of lesion area proportion and disease severity within entire images or selected regions. The system achieves a determination coefficient (R2) of 0.98 between calculated and manually measured lesion areas, with a disease severity assessment accuracy of 97.50 %, demonstrating satisfactory performance. The technological advancement offers substantial practical value for optimizing peanut cultivation practices, enabling targeted pesticide application, yield protection strategies, and accelerated screening of disease-resistant cultivars through quantitative phenotypic analysis.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序