[Retracted] Advances in Hyperspectral Image Classification with a Bottleneck Attention Mechanism Based on 3D-FCNN Model and Imaging Spectrometer Sensor

Table 5

Classification effects of different modules on the IP dataset.

Class	3D-FCNN	SE+3D-FCNN	BandAM+3D-FCNN	BAM+3D-FCNN
1	53.33	100	52.27	100
2	82.74	98.10	99.19	95.49
3	59.61	98.04	88.09	98.66
4	64.68	100	80.89	97.65
5	67.78	27.93	94.12	97.47
6	99.03	99.11	98.70	98.93
7	0	96.15	74.07	100
8	94.29	100	100	100
9	0	88.89	73.68	94.44
10	94.24	94.27	79.74	97.60
11	90.09	99.25	97.13	99.91
12	67.12	95.79	82.77	98.88
13	99.01	100	91.79	100
14	97.60	99.05	99.50	99.03
15	89.79	97.45	92.64	99.42
16	65.22	100	100	98.81
OA (%)	82.29	93.01	93.66	98.54
AA (%)	71.00	93.36	88.13	98.51
Kappa	79.64	91.98	92.75	98.33