Fig. 1. Focus structure

Fig. 2. Problems of underwater images. (a) Elongated and obscured objects; (b) small objects

Fig. 3. Overall structure of the proposed model

Fig. 4. Distribution of sea cucumber aspect ratio

Fig. 5. Different ways of convolution. (a) Conventional convolution; (b) deformable convolution

Fig. 6. DC-BoT structure

Fig. 7. Dilated convolution with different expansion coefficients. (a) 3×3 convolution; (b) dilated convolution with an expansion coefficient of 2; (c) dilated convolution with an expansion coefficient of 3

Fig. 8. FP-ASPP structure

Fig. 9. Feature fusion network with three different features. (a) Original YOLOv5 feature fusion network; (b) weight BiFPN; (c) WEVC-FPN structure

Fig. 10. CA module structure

Fig. 11. Results of different detection algorithms

Fig. 12. Visualization of feature maps. (a) Visualization analysis for elongated objects; (b) visualization analysis for occluded objects