Quality Detection Method of Penaeus vannamei Based on Lightweight YOLOv5s Network

The appearance and meat quality of Penaeus vannamei are important indexes in the production process, and the quality of the product will be reduced if the defective shrimp is mixed in during processing. In order to solve this problem, a quality detection model of Penaeus vannamei based on deep learning was put forward. Firstly, the self-made dataset of Penaeus vannamei was expanded to enhance the generalization ability of the neural network. Secondly, the backbone of YOLOv5 (you only look once v5) is replaced by the lightweight network PP-LCNet that removes the dense layer at the end, which reduces the model parameters and calculation. Then, the 7 x 7 convolution DepthSepConv module is embedded in a PP-LCNet backbone, which effectively strengthens the feature extraction ability of the network. Ultimately, SiLU activation function is used to replace the Hardsigmoid and Hardswish activation functions in the PP-LCNet backbone to enhance the regularization ability and detection speed of the network. Through comparative experiments, the all-round performance of the Shrimp-YOLOv5s network is higher than the current mainstream classical model and the lightweight model. The mAP@0.5, mAP@0.5:0.95, detection speed, parameters, and calculation of Shrimp-YOLOv5s are 98.5%, 88.1%, 272.8 FPS (frames per second), 4.8 M, and 9.0 GFLOPs (giga floating point operations) respectively.

Automated summary

To solve the problem of decreased product quality due to the mixing of defective shrimp during processing, a quality detection model of Penaeus vannamei based on deep learning was proposed. By expanding the self-made dataset and replacing the YOLOv5 backbone with the lightweight network PP-LCNet, the feature extraction ability of the network was effectively strengthened. The Shrimp-YOLOv5s network showed higher overall performance compared to current mainstream classical models and lightweight models.