Infrared Dim And Small Target Detection Based On Target Characteristic Analysis And Spatio-temporal Saliency Modeling Method

Infrared target detection system is widely used in many fields,and has good value in military aspects such as aircraft ship detection,guidance,and early warning,as well as civil aspects such as geological detection and industrial defect detection.Whether the target can be found accurately and effectively determines the performance of the infrared target detection system.Therefore,the research on infrared target detection is very important.However,due to the characteristics of infrared imaging,such as low proportion of infrared targets,lack of texture and other information,low contrast and complex background environment,it is difficult for traditional algorithms to achieve detection stability,high detection accuracy,and maintain detection the advantage of efficiency.Therefore,I introduced the analysis of airspace,time domain and visual saliency,and used more information to effectively process the detection of infrared targets.The main work and research contents are as follows:(1)Study the imaging principle of infrared targets,and analyze the infrared target characteristics,background characteristics and noise characteristics;collect infrared sequence images to make data sets,and preprocess the images.(2)A method for modeling infrared target saliency based on space-time entropy weight fusion is proposed.Aiming at the problem that the proportion of weak and small infrared targets in the image frame is too low and the features are lost during the conversion process,in the spatial domain,the Gaussian pyramid is constructed to extract the features of the image layer by layer to obtain the construction of the saliency map.The saliency map of the frame sequence map is used to obtain the time-domain saliency map,and then the final space-time saliency map is obtained through the weighted fusion algorithm to realize the detection of infrared targets,which makes up for the false detection caused by the existing algorithm due to too few target features.(3)A space-time saliency-based uncertainty estimation fusion algorithm is proposed.Aiming at the problem of weak target intensity and complex scenes in the project,the saliency map of image contrast,gradient,and optical flow amplitude and gradient is calculated through the co-occurrence histogram,and then the object map is obtained by capturing top-down target clues,and then synthesize the above-mentioned graphs into the final saliency map,and finally use the method of unsupervised uncertainty estimation to obtain the final result map to realize the detection of infrared targets.(4)An infrared target modeling and detection algorithm based on the space-time entropy flow method is proposed.Aiming at the problem of false detection of small targets in the complex environment disturbed by noise by existing algorithms,the entropy flow field is firstly calculated by the differential method,and then the two-dimensional histogram of the entropy flow is constructed to determine the main peak in the histogram.After determining the main peak,by comparing the background to obtain the difference image,and finally use the pipeline filtering method to realize the detection of small targets.The three algorithms mentioned in this Thesis are compared with other classic vision algorithms in quantitative and visual qualitative control experiments under multiple sets of data,and the comprehensive performance evaluation of the algorithms is obtained.The results show that the algorithm mentioned in this Thesis has better detection performance in the complex environment including edge and false alarm sources and can be effectively applied to the infrared detection system.