Research On Key Technologies Of Infrared Thermal Imaging Detection And Identification Of People Falling Into The Water At Sea

With the continuous occurrence of maritime accidents,maritime search and rescue has become an essential link,among which personnel search and rescue is the top priority.On the one hand,infrared imaging technology has developed rapidly in military and civil applications with its own advantages.On the other hand,with the continuous updating of image recognition technology,target detection technology has outstanding performance in visible light image recognition task.If the two technologies are combined for surface target detection,infrared can repair the defect of messy background of target detection.Target detection can fill the gap of low efficiency of independent detection of infrared technology,and can quickly,accurately and efficiently search for people falling into the water.In view of the fact that the person falling into the water conforms to the characteristics of weak and small targets at sea,based on the research of YOLO network,this paper analyzes and studies the improvement of backbone network and modular improvement,which improves the detection efficiency of target detection network for weak and small targets to a certain extent.The main research work and results include:1)Firstly,this paper takes preprocessing operations on the data set,including generating images,enhancing data and labeling images.Secondly,it enriches the data set and distributes it according to the proportion of 8:1:1.2)In order to realize the real-time search and rescue of people falling into the water,this paper introduces the lightweight network theory into the existing target detection algorithm,and improves the backbone feature extraction network with complex operation into Ghost Net with simplified parameters.The network module composed of ordinary convolution and deep separable convolution has less parameters.The introduction of this module reduces the parameter call frequency during training,effectively speeds up the network training speed and improves the network detection efficiency in theory.3)In this paper,the attention mechanism is introduced into the target detection algorithm,and the mixed attention mechanism module is inserted into the main feature extraction layer of YOLOv4 network,which suppresses some useless background information and highlights the target information,improves the efficiency of the network in feature extraction,and improves the detection success rate of weak and small targets such as drowning personnel to a certain extent.4)In order to further realize the idea of lightweight,this paper designs a detection system to further improve the efficiency of target detection.5)This paper verifies the effectiveness of the improved method through several groups of comparative experiments,uses user-defined data sets to verify each network model,compares and analyzes the training result indicators.Experiments show that the improved method is effective and feasible,and it also proves that the improved network has good detection performance in the detection task of people falling into the sea.