Research On Foggy Ship Detection Method Based On PSD And Improved YOLOv5

Sea fog is a common natural phenomenon that often occurs near ports,posing great challenges in areas such as transportation and safety regulation.Existing object detection algorithms are greatly limited in accuracy due to the impact of sea fog.Therefore,achieving accurate object detection under sea fog weather has become a very important and challenging problem.This article analyzes visible light images from real marine environments and finds several drawbacks:(1)sea fog weather at sea lasts for a long time and occurs frequently,and most visible light images are interfered with by sea fog,resulting in blurry imaging;(2)if the distance between the ship and the shooting device is far,the proportion of the ship in the entire picture will be very small,and the features will not be clear.The above-mentioned image characteristics seriously affect the accuracy of detection.Therefore,effective defogging measures need to be taken before object detection to obtain relatively clear images and reduce the adverse effects of sea fog.At the same time,in order to avoid information loss during the detection process,internal improvements need to be made to the detection network.Finally,by enhancing the network’s attention to targets,the ability to detect small ship targets can be further improved.This article mainly focuses on the research of image dehazing and object detection.Regarding the problems mentioned above,this article’s innovations and contributions include the following points:1.This article uses the PSD dehazing method to train a model specifically for sea fog,which effectively improves the situation where previous image dehazing models are not suitable for sea fog.The PSD dehazing method first trains with synthetic images and then finetunes with real sea images.The purpose of this operation is to generalize the trained model from the synthetic domain to the real sea environment,making the model’s dehazing effect better in the sea environment.2.This article designs an improved YOLOv5 object detection algorithm-SE-YOLO,which effectively solves the problem of poor detection performance in complex environments.In the first step,to avoid information loss inside the network,the SPPF module inside the YOLOv5 network is redesigned and soft pooling is introduced.The new SPPF module is called SOFT-SPPF,which reduces the loss of feature information during the pooling process.In the second step,to address the problem of insufficient network attention to small targets,the ECA attention mechanism is added to the path aggregation section,which makes the network more attentive to the ship’s features and reduces attention to useless background features.This can effectively improve the detection performance of the network.3.This article created a maritime-haze dataset specifically for research on foggy weather at sea,and combined the dehazing model and detection network on this dataset to verify the effectiveness of the proposed model.The images in this dataset were obtained from real sea environments,and the clear weather images were processed with fog and combined with real foggy images,named the maritime-haze dataset.Through designed comparative experiments,this article demonstrates the impact of each improvement on the results,proves the effectiveness of the proposed method,and also proves that the proposed method is superior to the currently popular methods,significantly improving the accuracy and efficiency of object detection in sea fog environments.