Research on target detection of Engraulis japonicus purse seine based on improved model of YOLOv5

The refined monitoring and identification of fishing operations by vessels is great significance value to vessels. In order solve the problem inaccurate statistics current Engraulis japonicus quota classification, this paper proposes an improved algorithm based on YOLOv5. This method introduces SENet attention mechanism into YOLOv5 backbone network structure, integrates target information in different periods operations, reduces interference complex backgrounds, improves precision model detection, ensures real-time detection efficiency. artificially shot videos are used as dataset study, video converted a picture format realize pre-labeling processing. 5550 images divided training set, validation test set according 8:1:1. verify validity data, control experiment, was replaced MobileNetV2 introduced, four models were implemented for comparison. experimental results show that research can obtain mean average (mAP) 99.3%, 98.9%, recall 98.7%, which 1.4%, 1.7% 2.5% respectively compared with original model. match expectations. According some categories, Kalman filter Hungarian matching count main categories baskets, accuracy 96.5% be obtained. threshold nets processing 85.8% 75% accuracy. These shows provide new ideas job auxiliary means statistics.