The silkworm industry holds great potential for intelligence and automation. This study aims to enhance the intelligence of cocoon processing and increase economic benefits, exploring the application of deep vision technology in high-precision automated classification of silkworm cocoons. By deploying a dual-camera system, it captures the features of both sides of cocoon clusters simultaneously, improving the comprehensiveness and efficiency of detection. The study uses the Faster R-CNN algorithm for cocoon positioning and the Mask R-CNN algorithm for identifying diseased areas on the cocoons. First, Faster R-CNN is employed for object detection, followed by Mask R-CNN to generate precise masks of the diseased areas. These models are installed at the cameras on both sides of the conveyor belt to achieve real-time cocoon positioning and disease area identification. Experiments on a self-built cocoon dataset show that the object detection model achieved an accuracy of 94.9%, and the instance segmentation model achieved an average precision (mAP) of 92.9%. The developed system has significantly improved cocoon classification, providing key technical support for the intelligent production of the cocoon industry. It helps manage and protect cocoon resources more effectively, holding broad application prospects.