An Embedding Skeleton for Fish Detection and Marine Organisms Recognition

The marine economy has become a new growth point of the national economy, and many countries have started to implement ranch project made strategic industry support vigorously. In fact, with continuous improvement people’s living standards, market demand for precious seafood such as fish, sea cucumbers, urchins increases. Shallow aquaculture extensively promoted vigorous development fisheries. However, traditional diving monitoring fishing are not only time consuming but also labor intensive; moreover, personal injury is significant risk factor high. recent years, underwater robots’ matured been applied in other technologies. Marine energy chemical construction opportunity growth. detection organisms an essential part intelligent strategy ranch, which requires robot detect organism quickly accurately complex ocean environment. This paper proposes method called YOLOv4-embedding, based on one-stage deep learning arithmetic organisms, construct real-time target system extract in-depth features, improve backbone’s architecture neck connection. Compared object arithmetics, YOLOv4-embedding was better at accuracy—with higher confidence ratio than EfficientDet-D3. results show that suggested could different varieties organisms. Furthermore, compared original YOLOv4, mAP75 proposed improves 2.92% dataset speed 51 FPS RTX 3090.