Small-size target detection and identification method based on deep learning and dual-band fusion

Abstract

The invention relates to a small-size target detection and identification method based on deep learning and dual-band fusion, and the method comprises the steps: 1, judging whether a target in an input image is a weak target or a small target, entering a step 2 if the target is the weak target, and entering a step 3 if the target is the small target; 2, carrying out super-resolution processing on the input image, then carrying out data enhancement processing on the processed image, and entering step 4; 3, segmenting the input image, amplifying each segmented image, performing data enhancement, and entering step 4; and 4, sending the image data after enhancement processing into a YetCNN network, extracting target position features, and further identifying the category and position of the target. According to the invention, high-robustness detection of weak and small targets is realized.

Description

technical field [0001] The invention belongs to the field of target recognition, and relates to a small-volume target detection and recognition method based on deep learning and dual-band fusion, and is especially suitable for detection and recognition tasks of hidden targets or targets with a small proportion of the frame. The small-volume target means that the resolution of the entire image is 1K (1920*1080 pixels), and the target image is less than 64*64 pixels. Background technique [0002] The goal of the object recognition method is to find out all the objects of interest in the image. It usually includes two subtasks of object location and object classification, and simultaneously determines the category and location of the object. These two subtasks are also the core starting point of the detection process and framework. [0003] The current mainstream technology trends are mainly divided into target recognition based on traditional algorithms and target recognition...

AI Technical Summary

Problems solved by technology

[0010] This method statement is mainly aimed at the detection of weak and small infrared targets in the imaging of infrared images, and uses visible light to compensate for the higher image signal-to-noise ratio in the infrared field of view, but it does not take into account the weak and small situations of visible light and infrared imaging. The detection and recognition rate is not high in the case

Images