基于PGBN模型的SAR图像目标识别方法

特征提取是合成孔径雷达图像目标识别的关键步骤，也是难点之一。该文提出一种基于PGBN(Poisson Gamma Belief Network)模型的SAR图像目标识别方法。PGBN模型作为一种深层贝叶斯生成网络，利用伽马分布具有的高度非线性，从复杂的SAR图像数据中获得了更具结构化的多层特征表示，这种多层特征表示有效提高了SAR图像目标识别性能。为了获得更高的训练效率和识别率，该文进一步采用朴素贝叶斯准则提出了一种对PGBN模型进行分类的方法。实验采用MSTAR的3类目标数据进行了验证，结果表明通过该方法提取的特征有更好的结构信息，对SAR图像目标识别具有较好的性能。

SAR Image Recognition Method with Poisson Gamma Belief Network Model

Feature extraction is a key step and difficult point in SAR image target recognition. This paper presents a novel method based on Poisson Gamma Belief Network (PGBN) for SAR image target recognition. As a deep Bayesian generative network, the PGBN model obtains a more structured multi-layer feature representation from the complex SAR image data using the high nonlinearity of the Gamma distribution, and the multi-layer feature representation effectively improves SAR image target recognition performance. In order to obtain a higher recognition rate and efficiency of training, this paper further proposes a method for classifying PGBN model based on the Naive Bayes rule. The experimental results about MSTAR dataset show that the feature extracted by this new method has better structure information, and it has better performance for SAR image target recognition.