Grape Disease Detection Model Based on an Improved YOLOv8

Given the complexity and diversity of grape diseases, along with the vast quantity and intricate associations of grape disease images, matching these images with specific disease types has become exceedingly challenging. To address this issue, this study delves into grape disease detection using an improved YOLOv8 model. We introduce the FocalModulation module into the YOLOv8 framework, incorporating Hierarchical Contextualization and Gated Aggregation steps to construct a novel model architecture named FMYOLO. Images collected from Baidu PaddlePaddle, were processed using mosaic data augmentation techniques before being analyzed with both the YOLOv8 and FMYOLO models, identifying disease types and characteristic regions. Experimental results demonstrate that the enhanced FMYOLO model improves the recall rate for black measles from 0.873 to 0.878 and for black root from 0.892 to 0.895, while overall accuracy is boosted from 0.937 to 0.939, marking increases of 0.5%, 0.3%, and 0.2% respectively. This provides a significant reference and guidance for further research and application in the field of grape disease detection.