SAR target identification method based on ensemble learning

Abstract

The invention discloses a SAR target identification method based on integrated learning, which mainly solves the problem of low SAR target identification performance in the prior art when the distribution of training data categories is unbalanced. The scheme is: 1. Extract the bag-of-words model features from the given training slice and test slice; 2. Randomly downsample the clutter class training samples, and train the obtained subset together with the target class training samples to obtain cost-sensitive Dictionary; 3. Randomly downsample the clutter class training samples, and train the obtained subset together with the target class training samples to obtain the SVM classifier; 4. Use the cost-sensitive dictionary and SVM classifier to classify the test samples, record The classification decision value of the test sample; 5. Use the maximum voting method to judge the classification decision value of the test sample, and determine the final category label of the test sample. The invention improves the performance of identification, and can be used for SAR target identification when the categories of training data in complex scenarios are unbalanced.

Description

technical field [0001] The invention belongs to the technical field of radar, in particular to a SAR target identification method, which can be used to provide important information for vehicle target identification and classification. Background technique [0002] Synthetic aperture radar SAR uses microwave remote sensing technology, is not affected by climate and day and night, has all-weather and all-weather working capabilities, and has the characteristics of multi-band, multi-polarization, variable viewing angle and penetrability. With the emergence of more and more airborne and spaceborne SARs, a large number of SAR data in different scenarios are brought. An important application of SAR data is automatic target recognition (ATR). Target identification in complex scenes has also become a current research direction. one. [0003] SAR target discrimination refers to a classifier learned from the training data set, which can be used to predict the category label of unkno...

AI Technical Summary

Problems solved by technology

[0005] 1. Since the traditional classification method basically takes the maximum overall classification accuracy of the training data as the classification criterion, when the distribution of training sample categories is unbalanced, the target class data accounts for a small proportion in the training data set, and the classifier usually tends to use the target class The class is judged as clutter class, so the detection rate of target class samples is low, which leads to poor performance of SAR target identification

[0006] 2. Because in the process of SAR target identification, more attention is usually paid to the accuracy of the target class. When the distribution of training sample categories is unbalanced, the classifier trained by the traditional classification method will produce a high detection rate for clutter samples. However, the detection rate of the target class samples is very low. When facing the SAR target identification of the unbalanced data set, it is necessary not only to maintain the original classification accuracy of the clutter class, but also to greatly improve the classification accuracy of the target class. Therefore, the existing These traditional SAR target identification methods cannot meet this requirement

Images