A method and system for detecting weak and small targets in infrared images

Abstract

The invention discloses a method and a system for detecting weak and small targets in an infrared image, and belongs to the field of image processing. The present invention solves the difficulty that infrared weak and small target features are difficult to directly learn through the idea of ​​using deep learning to estimate the background and detect the target after suppressing the background. The invention realizes the background estimation of the multi-scale target of the image to be detected by using operations such as down-sampling, up-sampling and shortcut links, thereby realizing the target detection of the multi-scale target and supplementing the information lost by the down-sampling, and solving the problem of background feature learning. The problem that cannot be balanced with local detail learning. The invention improves the accuracy of the background estimation of the target area by proposing the weighted mean square error loss function of the membership degree of the target area. By using depthwise separable convolution, including layer-by-layer convolution and point-by-point convolution, the present invention reduces the amount of parameters and computation of the network model, and accelerates the speed of target detection.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more particularly, relates to a method and system for detecting weak and small targets in infrared images. Background technique [0002] Infrared detection system has the advantages of long detection distance, good for concealment, uninterrupted work, and penetrability of clouds and fog. It has a wide range of applications in military and civilian fields such as military reconnaissance and disaster search and rescue. As one of the infrared target detection methods, remote sensing infrared detection has the advantages of wide detection field, high precision, and is not restricted by geographical location and meteorological conditions. The remote sensing infrared imaging image has a wide range, and the target of interest often occupies a small proportion in the image and has a low signal-to-noise ratio. Therefore, the separation, extraction and identification of the target of interest ...

AI Technical Summary

Problems solved by technology

However, since the MNIST handwritten data set is used as the foreground target, the target is not weak enough, so it is not representative, and the detection performance of the weak target is not good.

Images