Research On Small Object Detection In Infrared Imagery Based On Human Visual Saliency

With the development of infrared technology,infrared imaging detection systems have played an increasingly important role in missile precision guidance,night navigation,military early warning,and aviation detection,and are widely used in military and civilian fields such as national defense and industrial production.Infrared small target detection is one of the core topics of infrared imaging detection systems.Small infrared targets have a low signal-to-noise ratio and are often submerged in the natural background.How to detect and identify small infrared targets quickly and accurately is particularly important.This thesis studies the small infrared target detection algorithm based on visual saliency,explores and tries to improve the detection performance of the algorithm.The contrast mechanism is one of the important contents of visual saliency.The position of the target is determined by measuring the contrast difference between the target and its surroundings.In this thesis,an improved local contrast method is applied to the infrared small aircraft detection to extract its contrast features.At the same time,the origin moment and center moment of the image are introduced to extract the gray center and axis of the image and define the target shape feature.The proposed method has achieved good results on the image data set in this thesis.The shape feature has played a positive role in suppressing Gaussian spots,but it will increase the processing time.We found in the experiments that the improved local contrast method still over-processes the pixel-level highlight noises.Therefore,this thesis has further optimized the improved local contrast method and proposed a new difference-based local contrast method.Furthermore,combining median filtering and image segmentation,this thesis proposes an infrared small target detection system based on the difference-based local contrast method,and tests it on five representative real-life infrared image sets.Also,the comparisons with other typical infrared small target detection algorithms are also performed.Numerical experiments show that the proposed method in this thesis can not only achieve fast target detection speed,but also achieve the best performance in terms of detection probability,false alarm rate,target enhancement and background suppression.On this basis,we introduced parallel computing into this infrared detection system,leading to above 10 times speedup in image processing for a single image.