Authors

Abstract

To achieve mobile deployment of corn leaf disease detection, this study proposes a lightweight method, YOLOv8n-LSCSBD. The Lightweight Shared Convolutional Separable Batch normalization Detection (LSCSBD) is used to achieve cross-scale feature sharing convolution and independent normalization, thereby reducing computational complexity and preserving detection accuracy. Comparisons of YOLOv8 training strategies show that using YOLOv8n as the initial model, with a learning rate of 1e-2 and an optimizer of SGD, yields the best performance. Comparisons of different detection head schemes show that YOLOv8n-LSCSBD reduces the model size by 20.6% (to 5.0MB) compared to the original YOLOv8n model. When compared to YOLOv10n and YOLOv11n, the model size decreased by 13.8% and 9.1%, respectively. Notably, YOLOv8n-LSCSBD achieves P of 97.6%, R of 95.4%, mAP@0.5 of 97.7%, and mAP@0.5:0.95 of 87.3%. This method provides an efficient lightweight solution for mobile device deployment.

Abstract in Chinese

为实现玉米叶病的移动端部署检测，本研究提出YOLOv8n-LSCSBD轻量化检测方法。该方法通过轻量级共享卷积可分离批量归一化检测头（LSCSBD），实现跨尺度特征共享卷积与独立归一化，在减少计算量的同时保留检测精度。对比YOLOv8不同训练策略显示，使用YOLOv8n作为初始模型，学习率和优化器分别设置为1e-2和SGD时性能最佳。对比不同检测头方案显示，使用LSCSBD较原始YOLOv8n模型大小下降了20.6%至5.0MB。与YOLOv10n和YOLOv11n相比，模型大小分别下降了13.8%和9.1%。YOLOv8n-LSCSBD的P、R、mAP@0.5和mAP@0.5:0.95分别达到了97.6%、95.4%、97.7%和87.3%。该方法为移动端高效部署提供了轻量化解决方案。