Research On Infrared Typical Target Detection,Recognition And Tracking Method In Airborne Photoelectric System

Funded by the wide application requirements of airborne photoelectric system,the thesis carried out the related research on infrared target detection,infrared target recognition,and infrared target tracking based on saliency theory and sparse representation theory.The proposed algorithm provides new theoretical support for airborne optoelectronic system and improve the performance of airborne optoelectronic system.The detail research contents are as follows:Based on the study of infrared image characteristics,an effective algorithm based on saliency analysis in frequency domain is proposed.The infrared target is usually more significant than backgrounds in spectral domain.Thus,we can obtain the saliency map by some operations in frequency domain.In this way,target signal is enhanced and background clutter is suppressed.Then an adaptive threshold is adopted to segment the region of interest.At last,we measure the saliency of windows in the region of interest to predict the exact position of targets.Under the precondition of guaranteeing the performance of detection,the introduction of salience greatly improves the efficiency of the algorithm.To solve the problem of infrared target classfication,an infrared target recognition method based on model training is proposed.Feature is extracted by using sparse coding and spatial pyramid matching algorithm.At the same time,a feature dictionary is trained by K-SVD method for sparse coding.Then the sparse features are trained by the linear support vector machine(SVM)to get a classifier,completing the target recognition.Fusing the dictionary learning,feature representation and classifier training in a unified framework,achieves better recognition performance.A fast tracking method based on compressed sensing theory is proposed to solve the infrared target tracking.The multi-scale convolution is used to extract the feature of the sample.This feature incorporates the spatial and multi-scale information of the target.Then a very sparse measurement matrix is constructed to reduce the dimensionality and computational complexity.Finally,employing a naive Bayes classifier realizes the infrared target tracking.At the same time,the classifier is updated online during the tracking process,which increases the robustness of the tracking algorithm.Compared with the state-of-the-art methods,the proposed algorithm has obvious advantages in speed of tracking,and can track the target in a robust,accurate and fast way.On the basis of the theoretical research,this paper also studies the application of the proposed algorithm.Transplanting the algorithm into ARM,realizes a theoretical prototype.At last,we verify the effectiveness and engineering feasibility of the proposed infrared target detection,recognition and tracking algorithm.