一种融合Gabor滤波与3D/2D卷积的高光谱图像分类算法

为了减轻卷积神经网络模型对训练样本的依赖性 和提高高光谱图像的分类性能,本 文提出了一种融合Gabor滤波与3D/2D卷积的高光谱图像分类算法。首先,三维Gabor滤 波器用于处理原始高光谱数据以生成空谱隧道信息；其次,利用三维卷积操作提取生成的 空谱隧道信息的深层特征；然后,再利用二维卷积进一步提取图像的空间信息；最后,通 过Softmax分类器完成高光谱图像分类。为验证模型性能,将提出的方法与CNN、2D- CNN、3D-CNN-LR、SSRN算法在Indian Pines、Pavia University、Salinas数据集上进行 对 比实验。实验结果表明,提出的方法在三个数据集上的总体识别精度分别达到99.51%、99.94%、99.99%,均高于其 他方法,能够有效提高分类性能,是一种简单而高效的高光谱图像分类算法。

A hyperspectral image classification algorithm combining Gabor filtering and 3D/2D convolution

In order to reduce the dependence of the convolutional neural network model on training samples and improve the classification performance of hyperspectral images,this paper propo ses a hyperspectral image classification algorithm combining Gabor filtering and 3D/2D convolution.Firstl y,the three-dimensional Gabor filter is used to process the original hyperspectral data to generate space-spe ctrum tunnel information.Secondly,use the three-dimensional convolution operati on to extract the deep features of the generated space-spectrum tunnel information,and then use the two-dimensional convolution to further extract th e spatial information of the image. Finally,the hyperspectral image classification is completed by the Softmax clas sifier.In order to verify the performance of the model,the method in this paper is compared with CNN,2D-CNN ,3D-CNN-LR and SSRN algorithm on the data sets of Indian Pines,Pavia University and Salinas.Experi mental results show that the overall recognition accuracy of this method reaches 99.51%,99.94% and 99.99%,which are higher than other methods.The method in this paper can effectively improve the classification accuracy,is a s imple and efficient hyperspectral image classification algorithm.