Target identification method based on geometry reconstruction and multi-scale analysis

Abstract

The invention relates to a target identification method based on target geometry reconstruction and a multi-scale analysis theory. The method is characterized by firstly, reconstructing a target attitude, which includes flight attitude normalization and image preprocessing; then using Contourlet conversion to extract low frequency and high frequency characteristic vectors of an image and taking the vectors as a basis input training set and a correction base entry training set of a BP neural network; and finally, designing the BP neural network, designing a BP neural network correction model, using high frequency detail data to correct low frequency contour data, and determining an input and output layer, a middle layer number and an algorithm, wherein a trained network possesses an identification capability. Different types of nonoverlapping picture materials in an infrared picture database are selected to test precision of the identification method, and a result display identification rate is high and identification time is short. The method of the invention possesses high engineering applicability and has a certain meaning and a wide application prospect.

Description

technical field [0001] The invention belongs to the field of maneuvering target recognition in radar data processing. Background technique [0002] For the target recognition of infrared images, the main difficulties are: 1. The flying target images are usually captured, and the attitude diversity and geometric features are different, which increases the difficulty of recognition. How to reconstruct the flight attitude and image preprocessing, 2. The method of extracting feature vectors makes it have good target representation ability; 3, the pattern recognition method makes it have high recognition accuracy and short time. [0003] At present, in the field of target recognition, there are many methods for extracting image feature vectors, such as using wavelet transform, but wavelet transform reflects the point singularity of the signal, which cannot accurately describe the direction of the edge of the image, nor can it realize the sparse representation of the image, thus a...

AI Technical Summary

Problems solved by technology

[0003] At present, in the field of target recognition, there are many methods for extracting image feature vectors, such as using wavelet transform, but wavelet transform reflects the point singularity of the signal, which cannot accurately describe the direction of the edge of the image, nor can it realize the sparse representation of the image, thus affecting Recognition accuracy; there are many recognition methods, some apply corner feature and kernel clustering algorithm, some use wavelet transform for feature, some are based on closed contour feature, some use template matching, etc.

Images