基于深度学习神经网络的SAR图像目标识别算法

摘要

提出了一种有效的SAR图像目标识别新方法。首先采用改进后的增强Lee滤波和HOG变换对SAR图像进行特征提取，然后通过层叠RBM和GRNN相结合的混合神经网络对SAR图像进行目标分割和目标识别。利用测试图像库的MATLAB算法仿真，结果表明该方法可以明显提高目标识别率，正确率可以达到97%。%A new effective target recognition method for SAR images is proposed. First of all ,take the improved enhanced Lee filtering and HOG transformation for feature extraction of SAR images ,then through a hybrid neural network by cascading RBM and GRNN combination to operate object segmentation and target recognition of SAR images. Using MATLAB algorithm simulation of the test image database ,in this paper ,the method of object recognition based on the deep learning neural network algorithm can obviously increase the recognition rate ,and accuracy reaches 97%.