基于导向重构与降噪稀疏自编码器的合成孔径雷达目标识别

为解决现有合成孔径雷达（SAR）目标识别算法泛化能力差和算法复杂度高等问题，提出一种基于导向重构与降噪稀疏自编码器的SAR目标识别分类算法。利用导向重构算法对SAR图像进行两尺度融合预处理，生成一维图像矢量并作归一化处理，以降低图像输出特征的维度，提高预处理的速度；采用减少降噪自编码器隐层神经元方式对图像进行低维特征抽取和识别；使用Softmax分类器进行分类处理。实验结果表明，通过导向重构与降噪稀疏自编码器的SAR目标识别算法，不仅提高了目标识别性能以及泛化能力，而且降低了自编码器的隐层神经元数量和计算复杂度，网络结构也得到改进和优化。

A SAR Target Recognition Algorithm Based on Guided Filter Reconstruction and Denoising Sparse Autoencoder

The existing synthetic aperture radar（SAR） target recognition algorithms have the poor generalization ability and high complexity. For the problems above，an algorithm based on the guided filter reconstruction and denoising sparse autoencoder is proposed. The guided filter reconstruction algorithm with two-scale image fusion preprocessing of SAR image is used to generate an one-dimensional image vector and normalizate it in order to reduce the dimension of output feature of the image and increase the speed of preprocessing. The algorithm would extract and recognize the low-dimensional features of images by reducing the hidden layer neurons of the denoising autoencoder， which can effectively reduce the complexity of the algorithm. The Softmax classifier is used for classifying. The experimental results show that the SAR target recognition algorithm based on the guided filter reconstruction and denoising sparse autoencoder can not only improve the target recognition performance and generalization ability， but also reduce the number of hidden layer neurons in the autoencoder and the computational complexity, and the network structure has also been improved and optimized as well.