An Efficient Method for Sea Cucumber Recognition and Sorting Based on Improved YOLOv9 and RepViT

The automatic sorting system for sea cucumbers in food processing plants faces challenges such as high false detection rates, slow processing speeds, and sensitivity to light intensity variations. This paper presents a high-precision, high-efficiency real-time recognition and sorting method for sea cucumbers, based on YOLOv9 and the RepViT network. We improved the YOLOv9 model by introducing auxiliary training modules to help the model better understand the characteristics of sea cucumbers. Additionally, we used the lightweight RepViT network as the backbone to enhance the model’s expressive power and computational efficiency while maintaining a low weight. We replaced the original CIoU loss function with the EIoU loss function to accelerate convergence. Experimental results show that our improved model achieves an accuracy of 98.33% in sea cucumber sorting, with an inference speed of 92.71 fps and a model size of only 42.53 MB, outperforming most detection models. Moreover, the average sorting speed for a single sea cucumber is just 0.92 seconds, meeting the production needs of food processing plants.