Construction and Application of ReliefF-RFE Feature Selection Algorithm for Hyperspectral Image Classification

Copy Citation Text

show less

Fig. 1. Hyperspectral images of each standard dataset
(a): Indian pines; (b): Salinas-A; (c): KSC

Fig. 2. References of real features for each standard dataset
(a): Indian pines; (b): Salinas-A; (c): KSC

Fig. 3. Classification results of three feature selection algorithms for hyperspectral images of Indian pines dataset

Fig. 4. Classification results of three feature selection algorithms for hyperspectral images of Salinas-A dataset

Fig. 5. Classification results of three feature selection algorithms for hyperspectral images of KSC dataset

Fig. 6. Comprehensive comparison of three feature selection algorithms
(a): Classification accuracy; (b): Feature dimension and runtime

数据集	Indian pines	Salinas-A	KSC
特征维数	200	204	176
样本数	10 249	5 348	5 211
影像尺寸	145×145	86×83	512×614
分辨率/m	20	3.7	18
地物类别	16	6	13

Table 1. Hyperspectral dataset description

算法	OA /%	F-measure /%	Kappa 系数	特征维数	t /h
ReliefF	85.53	85.15	0.834 0	157	58.56
RFE	86.41	86.10	0.844 2	36	107.33
ReliefF-RFE	86.28	86.00	0.842 2	45	91.58

Table 2. Classification results based on the best feature subset of the Indian pines dataset

算法	OA /%	F-measure /%	Kappa 系数	特征维数	t /h
ReliefF	99.31	99.32	0.991 4	180	6.34
RFE	99.31	99.32	0.991 4	35	11.35
ReliefF-RFE	99.38	99.38	0.992 2	71	8.63

Table 3. Classification results based on the best feature subset of the Salinas-A dataset

算法	OA /%	F-measure /%	Kappa 系数	特征维数	t /h
ReliefF	93.22	93.14	0.924 0	145	8.22
RFE	93.41	93.31	0.936 2	72	14.98
ReliefF-RFE	93.16	93.06	0.923 8	64	9.58

Table 4. Classification results with the best feature subset of the KSC dataset

Download Citation