Research On Infrared Small Target Detection Based On Multi-directional Ring Gradient

Infrared small target detection in complex environment has attracted more and more attention,because it's the key technology to precision guided weapon system and aerial reconnaissance.Due to the distance between targets and infrared sensor is rather long in the infrared image,the imaging area of targets is apparently tiny,the signal-to-noise ratio(SNR)of image is low,the outline and texture information of the target image is blurred,and the available information is less.At the same time,there are a large number of highlight background areas and background edges in the infrared image in complex environment,which results in a large false detection rate and seriously affects the effect of the detection algorithm.However,the detection of infrared small target in complex environment is extremely challenging.1.Based on the research of the infrared small target detection,this thesis analyses the characteristics of infrared small target,and uses several commonly-used small target detection algorithms to simulate it.This thesis analyzes the gradient characteristics of the target using the gradient algorithm.We find that the target has a 360o gradient rapidly descending characteristics around its neighborhood,while the suspected target has the characteristics of gradient descending only in a few directions.In order to solve the problem of the low detection rate of common detection algorithm,this thesis adopts the multi-cascade combination method.Firstly,a detection algorithm based on multi-directional ring gradient method is proposed to extract ROI.In order to solve the scale problem of the target,the Gaussian Image Pyramid is constructed to extract ROI at multi-scale.Secondly,in order to make the segmentation and feature extraction more accurate,an adaptive multi-directional Top-Hat transform is proposed to suppress the background.Finally,the target is segmented by using an adaptive threshold algorithm.2.Based on the detection result has from the single-frame algorithm from the time-space domain,it is improved that searching a real target from continuous frames and eliminate false targets.Target features are extracted aiming to the segmented target.To increase the effect of inter-frame matching,Multiple features of the target are solved,such as size,gray level,centroid,Fast and so on.In order to complete the inter-frame matching more accurately,the similarity measurement criteria in the process of target matching is discussed in detail.Finally,in order to improve the flexibility of the inter-frame matching algorithm,a set of the evaluation function is designed to ensure the robustness of the inter-frame matching effect.3.Aiming at the overall algorithm flow,the advantages and disadvantages of several commonly used image processing chips are analyzed.Finally,the processor architecture of FPGA + ARM is selected for algorithm transplantation.For the infrared long-wave camera with resolution 640 ? 512,the target contrast is more than 5%,and there are no more than 5 targets in the field of view.This system can process 100 frames of images per second.Finally,several sequential images are selected to simulate the algorithm designed in this thesis and compare it with other algorithms in terms of detection rate and false detection rate.The single frame detection algorithm proposed in this thesis has greatly reduced the false detection rate on the premise of guaranteeing the improvement of detection rate by verification.The multi-frame determination algorithm proposed in this thesis,combined with other detection algorithms,has more greatly reduced the false detection rate of target detection.