改进3D-Octave卷积的高光谱图像分类方法

郑宗生, 王政翰, 王振华, 卢鹏, 高萌, 霍志俊

An improved 3D Octave convolution-based method for hyperspectral image classification

ZHENG Zongsheng, WANG Zhenghan, WANG Zhenhua, LU Peng, GAO Meng, HUO Zhijun

表1 不同算法在IP数据集的分类结果

Tab.1 Classification results of different algorithms on IP dataset(%)

类别 SVM 3D-CNN SSRN HybridSN DBDA 本文方法 1 0.00 94.40 88.10 93.18 100.00 100.00 2 65.10 89.22 97.57 98.71 98.75 99.33 3 84.27 93.85 97.83 97.38 99.73 100.00 4 100.00 93.75 96.35 99.01 98.60 100.00 5 95.08 97.97 97.25 99.30 99.52 99.24 6 90.45 97.93 99.07 97.90 99.84 94.60 7 100.00 92.86 100.00 96.00 95.83 99.89 8 93.71 94.81 99.54 100.00 100.00 100.00 9 0.00 66.67 75.01 80.95 94.74 100.00 10 79.78 90.16 97.61 97.74 97.28 97.92 11 48.36 96.40 95.82 97.81 97.69 99.01 12 66.54 92.55 97.14 98.27 99.60 98.51 13 95.71 91.92 100.00 99.45 96.86 97.87 14 88.00 99.39 98.69 99.56 99.30 99.04 15 72.63 96.18 95.47 96.90 98.85 98.96 16 99.56 91.76 92.59 88.76 88.17 97.60 OA 66.94 94.82 97.25 97.92 98.64 99.32 Kappa 61.20 94.10 96.85 97.76 98.43 99.13 AA 45.47 90.82 95.30 97.11 98.31 99.15