A hyperspectral image classification algorithm based on atrous convolution

Table 2 Classification results of different categories by various methods

Categories	Asphalt	Meadows	Gravel	Trees	Painted metal	Bare soil	Bitumen	Bricks	Shadows
Train samples	5305	14919	1679	2451	1076	4023	1064	2945	558
Test samples	1326	3730	420	613	269	1006	266	737	189
SVM [7]	0.827	0.859	0.000	0.730	0.982	0.196	0.000	0.740	1.000
1D CNN [28]	0.869	0.906	0.385	0.899	0.993	0.604	0.706	0.788	0.997
1D CNN [29]	0.939	0.967	0.748	0.949	0.998	0.916	0.771	0.860	0.932
2D CNN [30]	0.795	0.709	0.607	0.873	0.999	1.000	0.999	0.961	0.898
3D CNN [31]	0.976	0.955	0.947	0.976	1.000	0.981	0.983	0.985	1.000
3D CNN [32]	0.933	0.860	0.854	0.949	1.000	0.999	1.000	0.998	0.978
DLCNN [33]	0.979	0.959	0.966	0.978	0.998	0.995	0.983	0.986	0.997
DSSCNN [34]	0.922	0.967	0.828	0.977	1.000	0.915	0.895	0.840	1.000
RNN [35]	0.961	0.970	0.826	0.970	1.000	0.931	0.903	0.877	0.999
NG-APC	0.983	0.987	0.974	0.976	1.000	0.979	0.973	0.965	1.000
Average	0.918	0.914	0.714	0.928	0.997	0.852	0.821	0.900	0.980