Infrared weak target detection method of utilizing sample characteristic learning classification

Abstract

The present invention relates to an infrared weak target detection method of utilizing sample characteristic learning classification. The method is characterized by setting an area containing a target as a positive sample and a background area as a negative sample, and obtaining enough positive and negative samples with labels from an actual infrared image, and comprises the steps of firstly analyzing the samples at the four aspects of gray distribution, edge, information entropy and energy, and extracting seven characteristics of fitting residual, center surrounding contrast ratio, the radius of an edge fitting circle, the center offset amount of the edge fitting circle, the center distance variance of the edge fitting circle, the reference information entropy contrast ratio and the texture energy contrast ratio; then selecting an optimal characteristic subset from all characteristics by the modes of package type selection and forward search and by taking the area below a subject performance curve as an evaluation index; and then extracting the optimal characteristic subset of the positive and negative samples to train a support vector machine classifier, and supervising the learning; and finally, carrying out the top-hat transform pre-processing on the image to obtain a candidate target, and discriminating and screening via the classifier to obtain a final detection result.

Description

(1) Technical field [0001] The invention relates to a method for detecting weak and small infrared targets using sample feature learning and classification, which belongs to the field of digital image processing and mainly relates to mathematical morphology and machine learning. It has broad application prospects in various image-based application systems. (2) Background technology [0002] The infrared early warning system detects suspicious targets by passively receiving thermal radiation. The detection of small and weak targets has always been a key link. If the target can be detected as early as possible, countermeasures can be taken in advance to gain more preparation time. For the guidance system, Acquisition and tracking can be performed when the target is far away from the imaging system. Since most infrared scenes are relatively complex, and there are many interferences such as noise and edges, it is very easy to cause a high false alarm rate. Researchers have pro...

AI Technical Summary

Problems solved by technology

Some adaptive filters (see literature: Hahoud et al. Two-dimensional adaptive minimum mean square error algorithm, Transactions of the Institute of Electrical and Electronics Engineers Circuits and Systems, 1988,35(5):485-494. (Hadhoud M M,Thomas D W.The two-dimensional adaptive LMS (TDLMS) algorithm[J].IEEE Transactions on Circuits and Systems,1988,35(5):485-494.)) can automatically Adjust filter parameters, but require background statistics to change slowly, which is difficult to achieve in many cases

A detection method based on human vision (refer to the literature: Dong Xiabin et al. Infrared weak and small target detection and tracking method inspired by human vision system, Infrared Physics and Technology, 2014,62:100-109.(Dong X, Huang X, Zheng Y, et al.Infrared dim and small target detecting and tracking methodinspired by human visual system[J].Infrared Physics&Technology,2014,62:100-109.)) use DoG for saliency detection, but simply weighted by center and surrounding The comparison cannot effectively distinguish the target from the background, making it more sensitive to the edge area, resulting in too many false detections

Some methods based on feature classification establish a complete and comprehensive description criterion by extracting multiple features of the target and background, but some common features such as standard deviation, frequency domain energy, etc. (see literature: Jin Shanguo. For infrared search and tracking Analysis of small infrared target features and learning-based false detection removal for infraredsearch and application, 2014,17(4):883-900.(Kim S.Analysis of small infraredtarget features and learning-based false detection removal for infraredsearch and track[J].Pattern Analysis and Applications,2014,17(4):883-900.)) has limited descriptive ability, and it is easy to be confused with the correct target in the interference background clutter, so that it is impossible to distinguish the target from the background

[0003] Most detection algorithms only use one metric to detect and distinguish the target from the background, but in most cases, even if the interference such as edges is considered, the description cannot be guaranteed to be comprehensive enough, and there will always be various background noises in complex environments. Wave Impact Detection

Images