Deep learning based research on quality classification of shiitake mushrooms

The classification and processing of shiitake mushrooms is inclined to a labor-intensive task, which needs to pick shiitake mushrooms of high quality by labor force for a long time. In this paper, a high-efficiency channel pruning mechanism is proposed to improve the YOLOX deep learning method that is the latest version of YOLO serials algorithm for identification and grading of mushroom quality. Firstly, the YOLOX model is built by transfer learning after the image data set was expanded. Secondly, the built model was optimized by channel pruning algorithm. Finally, the pruned model is further fine-tuned by knowledge distillation, and the image data set was used to train the YOLOX network model optimized by channel pruning. The experimental results indicate that the improved YOLOX method proposed in this paper can inspect the surface texture of shiitake mushrooms effectively that mAP and FSP are respectively 99.96% and 57.3856, and the model size was reduced by more than half. Compared with Faster R–CNN, YOLOv3, YOLOv4, SSD 300 and the original YOLOX, the improved method proposed in this paper owns better comprehensive performance that it can be effectively applied to the rapid quality classification for shiitake mushrooms in production process. • YOLOX that the latest version of YOLO serial algorithms is applied in the quality classification of shiitake mushrooms. • The channel pruning algorithm is introduced into the YOLOX model and greatly reduces the number of model parameters. • The insufficient dataset samples are expanded by data enhancement method effectively. • The distillation method is adopted in the process of fine-tuning of model for restoring accuracy.