基于多尺度 3D-2D卷积神经网络的高光谱图像分类

设计提出了 1种针对高光谱图像分类任务的 3D-MSCNN模型。在 PCA降维的基础上,利用 3D空谱特征提 取网络和 2D多尺度特征提取网络实现高光谱图像特征提取,充分发挥高光谱图像空谱信息价值,增强对不同尺度 地表覆盖的表达能力。最后,利用 Softmax分类损失函数实现高光谱图像分类任务。实验结果表明,本文算法在 In. dian Pines和 Pavia University数据集上都取得了较好的分类效果。与 CD-CNN、3D-CNN、SS-Net和 HybirdSN等方法相 比,本文算法能够有效提升总体精度、平均精度和 Kappa系数等客观评价指标。

Hyperspectral Image Classification Based on Multiscale 3D-2D Convolutional Neural Network

A 3D-MSCNN model for hyperspectral image classification task is designed and proposed. On the basis of PCAdimensionality reduction, hyperspectral image feature extraction is realized by using 3D spatial-spectrum feature extractionnetwork and 2D multi-scale feature extraction network, so as to give full play to the value of hyperspectral image spatialspectrum information and enhance the expression ability of surface coverage at different scales. Finally, the hyperspectralimage classification task is realized by using softmax classification loss function. The experimental results show that the pro.posed algorithm has achieved good classification results on Indian pines and Pavia university datasets. Compared with CD-CNN, 3D-CNN, SS-NET, HybridSN and other methods, this algorithm can effectively improve the objective evaluation in.dexes such as overall accuracy, average accuracy and Kappa coefficients.