Study On Ship Detection Method Based On YOLOv5

Ship transportation is the most important mode of transportation in the world logistics.achieving more accurate and efficient detection of ship targets has important practical significance for ensuring the safety of ship navigation and improving the efficiency of shipping.At the same time,it has very important strategic significance for safeguarding national maritime rights and territorial sea security.The previous ship target detection model algorithms have many shortcomings in practical applications.In order to solve these problems,the thesis adopts deep learning methods to study ship target detection and proposes a model with better detection performance.The model extracts features with better robustness,higher detection accuracy,and faster detection speed.The main research content of the thesis is as follows:1.The labels of the satellite image segmentation dataset from Airbus company were converted and produced to create the experimental dataset for this thesis.2.The YOLOv5s＿CBAM＿Bi＿F algorithm model was proposed.Experiments on ship target detection were carried out on the basis of the YOLOv5s model.Aiming at the characteristics of data set with many interference backgrounds and small ship targets,the CBAM model of attention mechanism is added on the basis of the original model to highlight the key features of the target and enhance the detection effect.By introducing the idea of Bi FPN,the network structure is improved,the fusion of the characteristics of the Neck network is strengthened,and the detection ability is improved.paaper The loss function is improved to Focal-EIo U,to minimize the difference in length and width between the anchor and the ground truth box,while suppressing the influence of low-quality anchors,to make the model converge faster and achieve better localization.The experimental results of the model on the experimental dataset in this thesis showed that YOLOv5s＿CBAM＿Bi＿F algorithm model was improved by 0.94%than before.3.The MN-YOLOv5s algorithm model was proposed.Based on the YOLOv5s model,by changing Backbone to MobileNet v2 networks,the amount of parameters is reduced,and the detection speed is increased.By introducing Soft-NMS to penalize anchors,the detection effect in overlapping and occlusion situations is effectively improved.The experimental results of the model on the experimental dataset in this thesis showed that the MN-YOLOv5s algorithm has significantly improved the detection speed of the model,making it suitable for detecting requirements with high real-time performance or low-performance devices.