An application of YOLOv8 integrated with attention mechanisms for detection of grape leaf black rot spots.

Abstract

Aiming at the problems of low recognition rate of small target spots in grape leaf images and low detection accuracy due to low resolution of input images. In this paper, an improved recognition network based on YOLO v8 is constructed. In the constructed network, Spatial Pyramid Dilated Convolution (SPD-Conv) is used to replace each stepwise convolution layer and each pooling layer to better capture the detailed features of small targets. Meanwhile, the Efficient Multi-Scale Attention (EMA) Module is incorporated into the Neck part of YOLO v8 to make full use of the feature information of each detection layer and improve the accuracy of feature representation.The Plant Village dataset and the orchard image set are used to test the network performance of the improved model. The experimental test results show that the improved YOLO v8 has 92.64% precision, 93.28% recall and 96.17% AP. The size of model was a mere 7.1M. Compared to YOLO v8, the improvements are 2.38%, 1.91%, and 1.13%, respectively. Compared with the mainstream networks YOLO v4, YOLO v5, YOLO v6, and YOLO v7, precision is improved by 4.74%, 3.38%, 4.15%, and 4.69%, respectively. Therefore, the improved network proposed in this paper can improve the detection accuracy of small target objects and also identify the black rot disease of grape leaves more accurately.

Plant phenotyping relevance

ブドウ葉の黒腐病斑を画像から検出する改良YOLOv8を開発し、複数データセットで性能評価しており、植物病害状態の画像ベース表現型取得が中心である。

Code and data availability

The paper's own datasets (1070 labelled grape leaf black rot images, Test1/Test2 sets) are available only 'from the corresponding author on reasonable request' (request_only territory, not public). The only public link given is the generic PlantVillage website, a third-party dataset source, not a paper-specific deposit

No evidence-backed public reproduction asset is currently recorded.