Research On Infrared Target Detection In Complex Scenes

Infrared target detection algorithms have a wide range of applications in many fields such as military,security,transportation,etc.Therefore,research on infrared target detection algo-rithms has great practical value.According to the nature of infrared targets,this paper divides them into three categories: infrared small targets,infrared surface targets and infrared moving targets,and conducts the research on infrared target detection respectively.For the detection of small infrared targets,this paper proposes a fast infrared small target detection algorithm based on the Cen Sur E scale space.The scale space model performs detec-tion in multiple scale on small targets.The Cen Sur E operator uses the integral graph to greatly accelerate filtering calculations.In order to further reduce the amount of calculations,this paper applies rasterization on images,successfully improved real-time performance of the algorithm.As to infrared surface target detection,this paper made several improvements based on the YOLO v3 model.Specifically,this paper proposes a feature enhancement module based on soft attention mechanism and a feature fusion module for multi-scale features,strengthing the algorithm's capability to detect small and medium-sized targets.In addition,the loss function has been appropriately adjusted.In order to train and evaluate the model,this paper introduced a self-collected and annotated dataset of infrared drones.This dataset was used for verification of the proposed model.In terms of moving target detection,this paper proposes an infrared moving target detection algorithm in dynamic scenes.First,multi-frame images are used for global motion estimation,and the dynamic scene is transformed into a static scene setting.Then inter-frame difference method and the motion history image concept were used to detect the moving parts of the image sequence,and finally locating accurate positions of the targets.In this paper,all the proposed algorithms are verified by experiments using actual infrared images.The experimental results show that the infrared target detection algorithms proposed in this paper for different types of targets have good detection results and real-time performance.