Abstract:The vision-based automated recognition systems encounter substantial challenges in interspecies discrimination, primarily due to the diversity in mushrooms, particularly the intricate morphological characteristics and large similar species observed in wild varieties. Therefore, this study focuses on improving the accuracy of automatic detection and wild mushroom classification by optimizing the YOLOv9 deep learning model. With the above focuses, this study aims to achieve accurate differentiation of different species and provide species identification assistance for field picking. In specific, the study integrates DSConv, tailored for diversified mushroom morphology, and CoordConv, which boosts spatial information processing. Then, a deep learning model is successfully developed to optimize diversified mushroom morphology and spatial positioning, significantly improving the accuracy of mushroom species detection. After a series of experimental assessments, the recognition accuracy and the recall rate of the improved YOLOv9 model in the mushroom detection task are increased by 3.5% and 2.3%, respectively, compared with those of the YOLOv9-C model. Additionally, other performance indexes are also greatly improved. Compared with the current mainstream algorithm and the original version, the improved model shows obvious advantages, which proves its strong potential in practical applications.