Research on Improved YOLOv5 Rice Leaf Disease Detection Algorithm

Authors

  • Hongyi Xia, Lixin Tan College of Information and Intelligence Science and Technology, Hunan Agricultural University, Hunan 410125, China

Keywords:

DEFFN-YOLOv5, rice leaf disease detection, PixelShuffle

Abstract

In view of the complex characteristics, multi-scale and low efficiency of rice diseases, deep learning was used to construct the DEFFN-YOLOv5 rice leaf disease algorithm and study rice bacterial blight, rice blast and brown spot. In order to improve the accuracy of disease detection, the original YOLOv5 algorithm was improved and the PixelShuffle upsampling module was introduced to restore image details. In addition, feature extraction capabilities are enhanced, and deformable convolution and lightweight ECA channel attention modules are introduced. By using BiFPN to improve the PAN module, information interaction is enhanced and the model’s understanding and positioning capabilities are improved. Experiments have shown that the average accuracy (mAP) of the improved DEFFN-YOLOv5 algorithm in target detection reaches 86%, which is 3% higher than the original YOLOv5 algorithm. At the same time, the computational requirements are reduced by 4.6 GFLOPs, which is 27.85% less than the original YOLOv5 algorithm. These improvements make DEFFN-YOLOv5 perform better in rice disease detection.

Downloads

Published

2024-04-30

How to Cite

Hongyi Xia, Lixin Tan. (2024). Research on Improved YOLOv5 Rice Leaf Disease Detection Algorithm. Frontiers in Interdisciplinary Applied Science, 1(1), 1–7. Retrieved from https://fias.com.pk/index.php/journal/article/view/1

Issue

Section

Articles