Detection of surfacing white shrimp under hypoxia based on improved lightweight YOLOv5 model

White shrimp typically surface to breathe in the absence of adequate oxygen, and detecting this abnormal behavior can help better exploit the benefits of white shrimp farming. Therefore, an accurate and efficient model for detecting white shrimp surfacing was developed in this study. The proposed method is based on the YOLOv5 model, which utilizes ghost convolution to optimize standard convolution. A ghost bottleneck was constructed to improve the performance of the original bottleneck, and a more efficient detection layer was built based on the data. The model was trained and verified using a self-built white shrimp surfacing dataset. The mAP(@0.5) of this model was 98.139%, while the size and floating-point operations were only 1.76 MB and 2.1 G, respectively. Compared with Faster-RCNN, single-shot multi-box detector (SSD), and YOLOv4-tiny, our model presents higher detection accuracy and speed, as well as lower computation cost and smaller model size. Finally, based on the proposed method, we developed related applications for detecting shrimp surfacing.