A Target Recognition Method Based on Human-Machine Fusion

Abstract

The invention relates to a target recognition method based on human-computer fusion, which belongs to the field of remote sensing image processing and analysis, and solves the problem of target recognition and classification of large-scale multi-target remote sensing images; the method includes acquiring eye movement signals when an image analyst performs image analysis , generating an eye movement heat map corresponding to the image, determining the candidate target area in the image according to the set heat threshold, cropping the image to extract the candidate target; using the pre-trained convolutional neural network to perform feature extraction on the candidate target; according to the candidate target features for object recognition and classification. The invention combines the accurate and efficient target search ability of human eyes and the powerful target classification ability of convolutional neural network to realize target recognition of large-format visible light remote sensing images, and can be extended and applied to infrared, hyperspectral, and synthetic aperture radar remote sensing image target recognition.

Description

technical field [0001] The invention relates to the field of remote sensing image processing and analysis, in particular to a target recognition method based on human-machine fusion. Background technique [0002] Target recognition is an important research content in the field of remote sensing image analysis and application. Its main purpose is to distinguish specific types of targets, such as distinguishing whether a certain aircraft is a Boeing-737 or a Boeing-787. [0003] Existing target recognition methods are usually based on target detection, that is, it is assumed that a slice image containing a single target has been obtained through detection, but it does not meet the requirements of recognizing targets in large-format images containing multiple targets and backgrounds in practical applications. Require. [0004] At present, there are two feasible methods for identifying targets in large-format images: the first method is to detect first and then recognize, which...

AI Technical Summary

Problems solved by technology

However, in order to ensure the recognition performance of the second method, the candidate target extraction method needs not only a high recall rate, but also high efficiency and accurate positioning, that is, the number of candidate targets extracted should not be too large, and the position of the candidate target can be accurately located; However, the existing candidate target extraction methods cannot meet the above requirements, and usually hundreds of candidate targets are extracted, and the positioning is not accurate enough; and in the second method, the subsequent target feature extraction after extracting candidate targets mainly uses traditional artificial design features, Cannot meet the requirements for target recognition

Images