SAR time-sensitive target sample augmentation method for deep learning training

Abstract

The invention relates to an SAR time-sensitive target sample augmentation method for deep learning training, and belongs to the field of image processing technologies and deep learning. The SAR time-sensitive target sample augmentation method comprises the following steps of: step 1, shooting a heterologous SAR image set with a resolution of a meters in a region where targets are located, and converting the heterologous SAR image set into a heterologous SAR image set with a resolution of b meters; step 2, finding out all targets, and making each target into an SAR time-sensitive target slice to obtain a slice set; step 3, intercepting a background image from each image in the heterologous SAR image set to obtain a background image set; step 4, performing optimization processing on each slice in the slice set; step 5, establishing a learning sample set of the time-sensitive target; and step 6, rotating the learning sample of the time-sensitive target to obtain learning samples at different angles. According to the SAR time-sensitive target sample augmentation method, the problem of poor deep learning training effect caused by a small number of samples and lack of consideration of the features of the deep learning network is solved.

Description

technical field [0001] The invention belongs to the field of image processing technology and deep learning, and relates to a SAR time-sensitive target sample augmentation method for deep learning training. Background technique [0002] In recent years, countries have rushed to introduce artificial intelligence technology into the military field, and accelerate the development and deployment of intelligent weapons and equipment and intelligent new concept weapons. Due to the advantages of convolutional neural networks in the field of target recognition and classification, the branch of intelligent perception and recognition has been widely developed in the military field, including intelligent target recognition and classification technology based on spaceborne SAR images, and intelligent information processing technology for missile-borne SAR images. In order to obtain better recognition and classification results for the above technologies, on the basis of selecting an appr...

AI Technical Summary

Problems solved by technology

[0003] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art, to propose a SAR time-sensitive target sample augmentation method for deep learning training, and to solve the problems caused by the small number of samples and the lack of consideration of the deep learning network characteristics The problem of poor learning and training effects

Images