Context-constraint-based target identification method

Abstract

The invention discloses a context-constraint-based target identification method, and belongs to the field of remote sensing image processing. The method is used for classifying a remote sensing image scene and detecting and identifying a target. The method comprises the following steps of: filtering an image, performing region segmentation to segment the image into a plurality of connected regions, and marking each connected region; calculating characteristic vectors of each connected region, inputting the characteristic vectors into a classifier which is trained in advance for scene classification calculation, and outputting a class mark sheet; determining the scope of a local region in which a target possibly exists on the mark sheet according to the target to be identified on the basis of the class mark sheet, preprocessing the local region, and calculating a region of interest in the region; and extracting characteristics, and inputting the characteristics into the classifier for identification. The scene classification method is quick and effective, and aims to provide effective context constraints for target identification and improve the identification efficiency and the identification accuracy.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing, and more specifically relates to a target recognition method based on context constraints. Background technique [0002] Target detection and recognition of remote sensing images has developed rapidly in recent years because of its important military and civilian values. However, due to the large amount of satellite remote sensing image data, it is a difficult and time-consuming task to manually analyze and extract the information of interest one by one, which urgently requires us to use computer-aided technology to analyze remote sensing images. [0003] Automatic object recognition is one of the important and challenging research directions. At present, the research methods for this problem are mainly to provide target templates or establish target models, and search and match in the whole image. For images, the calculation is heavy and the speed is slow. The scene classificati...

AI Technical Summary

Problems solved by technology

For example, the classification accuracy of neural network and decision tree is generally higher than that of maximum likelihood method, but they are not stable classifiers; for example, the method of introducing fuzzy theory into neural network can be divided according to the degree of membership by constructing a remote sensing fuzzy classification model. However, there are no mature theories and rules for the determination of the membership function, which often requires expert experience and is subjective, which is the biggest shortcoming of the fuzzy classification method

Images