AUTHOR=Zhu Shisong , Ma Wanli , Wang Jianlong , Yang Meijuan , Wang Yongmao , Wang Chunyang 

TITLE=EADD-YOLO: An efficient and accurate disease detector for apple leaf using improved lightweight YOLOv5

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2023.1120724

DOI=10.3389/fpls.2023.1120724

ISSN=1664-462X

ABSTRACT=Current detection methods for apple leaf diseases still suffer some challenges, such as the high number of parameters, low detection speed and poor detection performance for dense small spots, which limit the practical applications in agriculture. Therefore, an efficient and accurate method for apple leaf disease detection based on YOLOv5 is proposed and named EADD-YOLO. In the EADD-YOLO, the lightweight ShuffleNet Inverted Residual (SNIR) module is utilized to reconstruct the backbone network, and the proposed novel DWC3 module redesigned through efficient depthwise convolution (DWConv) is introduced into the neck network to replace the original C3 module. The aim is to reduce the number of parameters and floating point of operations (FLOPs) during feature extraction and feature fusion, thus improving the operational efficiency of the network with less impact on detection performance. In addition, the coordinate attention (CA) module is embedded into the critical locations of the network to select the critical spot information and suppress useless information, which is to improve the detection accuracy of diseases with various sizes from different scenes. Furthermore, the SIoU loss replaces CIoU loss as the bounding box regression loss function to improve the accuracy of prediction box localization. The experimental results indicate that EADD-YOLO can achieve the detection performance of 95.5% in mAP and a speed of 625 FPS on a dataset composed of images of five common apple leaf diseases (ALDD). Compared to the latest research method MEAN-SSD on the ALDD, the detection accuracy and speed of the proposed method were improved by 12.3% and 596 FPS, respectively. In addition, the number of parameters and FLOPs of EADD-YOLO were much less than those of MEAN-SSD. In summary, the proposed method not only has a satisfactory detection effect, but also has fewer parameters and high calculation efficiency compared with the existing approaches. Therefore, the proposed method provides a high-performance solution for the early diagnosis of apple leaf disease and can be applied in agricultural robots.