Research on target detection and recognition algorithm of Eriocheir sinensis carapace

Chinese mitten crab is one of China's unique aquaculture species, which has significant economic value to the aquatic product market. In order to recognize different individual Chinese mitten crab, this paper proposes a method of first detection and recognition of carapace combined with YOLOv5 (You Only Look Once v5) and principal component analysis (PCA) and its improved method. First, the image of the Chinese mitten crab is obtained through the camera. Secondly, YOLOv5 and transfer learning method are used to detect the target of the Chinese mitten crab, and then the target is automatic cropped according to the detected target frame of Chinese mitten crab carapace. Finally, four methods of KPCA, one-dimensional PCA (1D-PCA), two-dimensional PCA (2D-PCA), and two-way two-dimensional PCA ((2D)(2)-PCA) were used for matching. The results show that the (2D)(2)-PCA recognition rate can reach 84.42%, which is 18.27%, 9.128% and 8.689% higher than the other three methods respectively. In addition, the matching speed only takes 1.859 s, compared with the other three methods. The method improves by 86.051 s, 2.562 s and 0.784 s, respectively. Therefore, this method has a better experimental effect in the experiment, and the recognition speed is faster. The results of the study can avoid the economic loss of aquatic products and provide a new research method for the recognition of the Chinese mitten crab carapace.