基于CNN的SAR图像目标分类优化算法

随着合成孔径雷达技术的成熟，传统方法已经难以满足海量SAR数据的分类精度和速度需求。为解决上述问题，采用卷积神经网络对海量SAR数据进行分类。针对SAR图像数据的特点，对卷积神经网络结构参数进行调整，提高网络训练速度，克服权重更新中的梯度消失，改善网络训练过程中收敛慢的问题，提升目标分类准确率。同时提出了一种ZCA白化与主成分分析相结合的方法对SAR图像进行预处理，进一步提升了网络的训练速度以及目标分类的准确率。实验采用的是美国MSTAR数据库，通过上述优化方法得到了较好的分类效果。

SAR Images Target Classification Algorithm Optimization Based on Convolutional Neural Network

With the development of synthetic aperture radar technology, the traditional methods are difficult to meet the classification accuracy and speed requirements on the massive SAR data processing. In order to solve the problem, the convolutional neural network is used to classify massive SAR data. According to the feature of SAR image data, the parameters of convolutional neural network are adjusted. The training speed is improved, the gradient disappearance during the weight update is overcome, and the rate of convergence and the classification accuracy are improved. Meantime, a pre-processing method based on the combination of ZCA whitening and principal component analysis is proposed. The training speed and the accuracy of target classification get a further improvement. The MSTAR database was used in the experiments. The classification results are better than others in this way.