Research On Detection And Tracking Of Infrared Targets Based On Spatio-temporal Saliency

Limited by the visible imaging mechanism,lens can not effectively carry out the work all the time.So the infrared imaging model with its unique advantages makes up for the lack of visible light has been widely used.However,infrared images are vulnerable due to background noise and clutter.Therefore,the result of infrared image detection can not be satisfied.In this paper,the method of detection and tracking of infrared image small targets is discussed.Based on the spatial domain and time domain feature extraction algorithm,an algorithm for infrared target detection and tracking based on spatiotemporal saliency is proposed.The improved algorithm improves the robustness of dim target detection in infrared image.On this basis,the network model of visible light target recognition is improved,and the infrared target recognition based on convolution neural network is realized.In this paper,the classical infrared target detection algorithm is studied.From the basic mathematical model of infrared image,two kinds of common infrared image background suppression algorithms are introduced,and the visual attention model is selected to detect the infrared target.In this paper,a dual channel tunnel residual method is proposed to extract the spatial domain saliency of infrared targets.Firstly,the saliency of the coefficients of the contourlet decomposition coefficients is used to propose the maximum median filtering for the coefficients of the subbands of the contour wave.After the fractal dimension is transformed to the Fourier frequency domain,the fractal dimension and the original image are computed by spectral residual method(SR),so the salient image of the image in the space domain is finally generated.For the time domain saliency of infrared images,a new method of infrared target motion information extraction based on optical flow is proposed in this paper.When the background and the small target both exist in the image at the same time,the background motion optical flow field is used to calculate the motion of target,so as to extract the small target motion information and obtain the significance in the time domain.In this paper,a new rule of image fusion with weighted dynamic summation and multiplication algorithm is proposed.The improved spectral residual and infrared small target are multiplied by multiplicative fusion,the significant target of the space domain is filtered,and the dynamic weighted addition of the frame difference between adjacent frames is fused,thus the stability of the algorithm is guaranteed.The recall and precision of the algorithm are higher than those of other infrared target detection algorithms.For infrared target recognition problem,a convolution neural network structure suitable for infrared targets is designed in this paper.Firstly,7 categories of infrared target data sets and test sets are constructed by manual annotation,after that the network training and testing are completed.The experiment results show that the improved network structure improves the efficiency of image training and improves the accuracy.The experimental results show that the algorithm proposed in this paper can improve the accuracy rate of infrared target detection and can stabilize the detection and tracking of the weak and small targets in the infrared video sequence,under the premise of guaranteeing the robustness.