Detection of wheat scab fungus spores utilizing the Yolov5-ECA-ASFF network structure

Abstract

Rapid detection and identification of Fusarium germinate spores play a vital role in the early prediction and effective management of wheat scab disease. This study proposed an improved Yolov5-ECA-ASFF target detection algorithm that addressed the challenges of small size and precise localization of spore image targets. The algorithm incorporated the attention mechanism module (ECA-Net) and adaptive feature fusion mechanism (ASFF) into the feature pyramid structure of YOLO, effectively tackling issues related to small size, limited characteristics, and unclear attributes of F. germinate spores. The results demonstrate that the proposed model achieved an average recognition accuracy of 98.57% for F. graminearum spores, surpassing the original Yolov5s algorithm’s mAP value by 6.8%. The proposed method outperformed other mainstream target detection networks like Yolov4 and Faster-RCNN. It also exhibited excellent recognition outcomes in scenarios involving multiple targets and complex backgrounds, while maintaining model robustness even when faced with similar appearance, morphology, and color characteristics of various scab spores. In conclusion, this method accurately detected and identified wheat scab spores in the presence of a variety of mixed spores, providing crucial technical support for automated detection of wheat scab spores and early prediction of wheat scab outbreaks under complex field environments.