Research on Marine Target Detection and Recognition Based on YOLO Algorithm

Abstract

Abstract Marine target detection is a crucial technology for developing and utilizing marine resources, and fast and accurate detection of marine organisms is of great research significance for sustainable development and protection of marine resources. In this paper, we first discuss the convolutional neural network and attention mechanisms in deep learning and then present the YOLO series of algorithms for target detection. Then, we take ResNeXt50 as the backbone network, introduce the global attention mechanism and ASFF module to establish the GA-YOLOv5s model for marine life target detection and recognition, and also design the training strategy of the model through migration learning. Simulation experiments are planned to verify the analysis after considering the feasibility of the GA-YOLOv5s model for marine target detection and recognition. The YOLO algorithm, based on multiple improved strategies, improves the mAP@0.5 of marine target detection and recognition by 5.68%. The detection speed of the model after incorporating the GAM module is 48.51FPS; the model using ASFF mAP@0.5 increased by 3.38%; and the average precision and recall of the model for marine target detection and recognition are 82.79% and 80.17%, respectively. Based on the YOLO algorithm, accurate detection and recognition of marine targets can be achieved, which provides reliable technical support for enhancing the exploitation and protection of marine resources.