Research on mushroom classification and recognition system based on improved YOLOv8

With the advancement of artificial intelligence technology, deep learning has demonstrated significant potential in the recognition and classification of agricultural products. This study, based on the YOLOv8 deep learning framework, designs and implements an efficient and accurate mushroom classification and recognition system. The model integrates the detection structure of Swin Transformer into the head network to reduce computational redundancy; incorporates the C2f_Deformable Convolutionv4 (C2f_DCNv4) structure with deformable convolution to reconstruct the YOLOv8 backbone network, optimizing and enhancing its feature transmission and extraction capabilities; employs an Efficient Layer Attention (ELA) mechanism to improve network detection speed and accuracy; replaces the original loss function with Wise-Intersection over Union Loss (WIoU) to enhance training effectiveness and convergence speed; and achieves precise recognition of nine common mushroom varieties through the optimization of the YOLOv8 network structure combined with transfer learning strategies. Experimental results indicate that the improved system achieves an average recognition accuracy ([email protected]) of 98.49\% on the Fungi dataset, demonstrating substantial practical value. ()