Visual Detection of Portunus Survival Based on YOLOV5 and RCN Multi-Parameter Fusion

Single-frame circulation aquaculture belongs to the important category of sustainable agriculture development. In light visual-detection problem related survival rate Portunus in single-frame three-dimensional aquaculture, a fusion recognition algorithm based on YOLOV5, RCN (RefineContourNet) image residual bait ratio, centroid moving distance, and rotation angle was put forward. Based three-parameter identification LWLR (Local Weighted Linear Regression), model each parameter established, respectively. Then, softmax used obtain classification judgment Portunus’ rate. YOLOV5 centroid, EIOU (Efficient IOU) loss function improve accuracy target detection. RCN, edge detection recognition, optimized binary cross-entropy double thresholds successfully improved clarity contour. The results showed that after optimization, mAP (mean Average Precision) improved, while precision (threshold 0.5:0.95:0.05) between were by 2% 1.8%, training set reduced 4%, obtained using experiment shows 0.920, 0.840, 0.955 under single parameters angle, respectively; multi-feature 0.960. multi-parameter 5.5% higher than single-parameter (average accuracy). relative (average) percentage.