Research On Single Target Detection And Tracking Technology Based On Infrared Multi-spectrum Fusion

In recent years,deep learning has achieved great success in the fields of target recognition,target detection,and target tracking.However,these model algorithms mostly use visible light images as input sources,and there are few researches on infrared target detection and tracking algorithms.Compared with visible light images,infrared data has the characteristics of low resolution and low contrast,and the texture information of infrared targets is not obvious,and the infrared data set required for training is relatively small.These factors bring huge challenges to the detection and tracking of infrared targets.Therefore,it is particularly necessary to carry out research on detection and tracking algorithms for infrared targets.The current mainstream detection and tracking algorithms are difficult to balance speed and accuracy well,and it is difficult to cope with the challenges of scene changes,partial occlusion,and the reappearance of targets after they disappear.Based on these problems,this paper uses multi-spectral fusion technology to perform pixel-level fusion of infrared images of different spectrums,and proposes an improved Center Net infrared target detection algorithm and an infrared single target tracking network based on global re-detection.In this paper,the algorithm is tested by constructing infrared data sets of different scenarios.The main research contents are as follows:Firstly,in view of the different imaging characteristics of multi-spectral images,research on the fusion algorithm of infrared multi-spectral image sequences is carried out.In this paper,the enhancement algorithm and the noise filtering algorithm are used to preprocess the data,and the preprocessed images of the three spectrum bands are fused at the pixel level.This paper selects the optimal fusion strategy by comparing fusion algorithms such as principal component analysis,pyramid transform,wavelet transform,and non-subsampled contourlet transform.Secondly,in view of the difficulty of the existing detection algorithms to balance speed and accuracy,and the difficulty of coping with complex infrared scenes,the research on infrared target detection algorithms based on improved Center Net was carried out.In this paper,starting from the selection and improvement of the backbone feature extraction network,the attention mechanism,feature enhancement module and multi-detection head prediction module are successively added.Aiming at the infrared detection scene,an infrared data set is constructed,and the test work of the improved algorithm is carried out on the data set.The experimental results show that the improved algorithm in this paper can balance various detection indicators and effectively improve the infrared target detection performance in complex scenes.Thirdly,aiming at the problems of partial occlusion of the target in the infrared tracking scene,reappearance after disappearance,and background change,the research of infrared single target tracking algorithm based on global re-detection is carried out.This paper compares the mainstream classification models,chooses Efficient Net as the reference network,and optimizes the structure of the reference network by introducing a cross-stage partial network.At the same time,the characteristics of the template branch and the branch to be searched are used to perform weight coding on the two feature branches at the channel level.The path aggregation network is introduced as a feature enhancement module to fuse feature layers of different resolutions,thereby improving the scale generalization ability of the network.In the prediction layer,the area guided by the template is introduced to generate the network structure,and the network is guided to output the mark box of the specified target.Aiming at the infrared tracking scene,an infrared data set is constructed,and the tracking algorithm testing work is carried out on this data set.Experimental results show that the tracking algorithm in this paper can effectively cope with tracking tasks in infrared scenes,and is superior to other algorithms in terms of accuracy and success rate indicators,and has good robustness.Finally,this paper builds a system test platform to further illustrate the necessity of introducing spectrum fusion.This article uses Py Qt to develop the main interface of the test system.Based on the infrared simulation system,the multi-scene test sequence is constructed and the test sequence is used to quantitatively analyze the impact of the fusion algorithm on the performance of detection and tracking.The experimental results show that the detection and tracking performance of the fusion image sequence is better than that of a single spectrum.