Research On Infrared Weak And Small Target Detection And Tracking Based On Deep Learning And Biological Vision Mechanism

Infrared small target detection and tracking is widely used in military early warning,traffic monitoring,and air security.Although infrared detection has the advantages of long detection distance,working around the clock,and is not easy to be found.However,in the actual environment,there is little prior information about infrared targets under complex backgrounds.The traditional infrared target detection and tracking methods still have the problems of slow detection rate and low detection accuracy,which makes infrared target tracking and classification recognition a very challenging problem.First of all,This paper proposes an infrared target detection algorithm combining deep learning and biological vision mechanism.Construct a generative adversarial neural network and train the generator to generate infrared image samples.Infrared image noise reduction processing is simulated by Gabor kernel function to simulate the response of biological visual neural unit.Then construct a convolutional neural network to detect infrared targets.Convolutional neural networks are trained by minimizing the entropy of the predicted position of the infrared target and the true infrared target position.After the parameter training,input the infrared image to be processed frame by frame to detect the candidate target area.Secondly,in order to overcome the defect that the convolutional network cannot detect unknown targets,further refine the detection results and mine the human visual operating mechanism and.The infrared image pixels are used as optic neurons to construct the optic nerve space.Describe the local probability distribution characteristics of infrared images in high-dimensional optic nerve manifold space,map infrared target image data points in high-dimensional optic nerve manifold space to low-dimensional space for classification.Realize the difference of infrared target detection results,and provide a good data foundation for infrared target trajectory tracking.Finally,for the problem that the infrared target trajectory cannot be accurately tracked,a tracking method based on biological visual matching and prediction mechanism is proposed.This article combined with the focus matching mechanism of local moving targets in the biological vision mechanism,the starting position of the infrared target trajectory is determined.With the help of the trajectory prediction mechanism in the biological vision mechanism,a trajectory prediction and tracking system based on the motion characteristicsof infrared dim targets is designed.Generate corresponding prediction trajectory sets for different types of targets,and use the Hungarian algorithm to optimize the distribution of this target trajectory and the prediction trajectory with Euclidean distance as the weight.Finally,accurate tracking of the infrared target trajectory is achieved.In the experimental part,this article selects the infrared target image data set on the public data set and makes corresponding labels.MATLAB simulation verifies that the algorithm in this paper can effectively detect and track the infrared weak target under the basic requirements of real-time detection,and Realize the classification of different types of infrared targets.