基于改进YOLOv5s的苹果叶片病害检测研究

Improved YOLOv5s-based Method for Apple Leaf Disease Detection

目前苹果叶片病害检测技术仍然存在检测精度低、效率低的问题。对此,本文提出一种基于改进YOLOv5s的算法进行病害识别。首先,增加小目标检测层改进漏检问题,提高检测精度;其次,引入双向特征金字塔结构加强特征提取,融合多尺度特征扩大视野;最后,将损失函数替换为SIo U,解决了预测框和真实框方向不匹配问题。实验结果表明:改进后的算法在Original数据集上mAP0.5为95.4%,比传统的YOLOv5s提升了3.3%。改进后的算法在复杂度没有发生很大变化的基础上明显提升了算法性能。

The current apple leaf disease detection technology still has the problems of low detection accuracy and low efficiency. In this regard, this paper proposes an algorithm based on improved YOLOv5s for disease identification. Firstly, the small target detection layer is added to improve the leakage detection problem and increase the detection accuracy;Secondly, the BiFPN is introduced to strengthen the feature extraction, and the multi-scale features are fused to expand the field of view;Finally, the loss function is replaced by SIoU, which solves the problem of mismatch between the prediction frame and the real frame direction. The experimental results show that the improved algorithm has a mAP0.5 of 95.4% on the Original dataset, which is 3.3% higher than the traditional YOLOv5s. The improved algorithm significantly improves the performance of the algorithm on the basis of no significant change in complexity.