Dual-Band Maritime Imagery Ship Classification Based on Multilayer Convolutional Feature Fusion

Table 4

Comparison of classification accuracy (%) with other state-of-the-arts on VAIS dataset.

Methods	VIS	IR
CNN [21]	81.9	54.0	82.1
Gnostic field [21]	82.4	58.7	82.4
[21]	81.0	56.8	87.4
ME-CNN [20]	87.3	—	—
[34]	87.6	—	—
-CLBP [19]	88.0	—	—
Multimodal CNN [15]	—	—	86.7
DyFusion [22]	—	—
SF-SRDA [4]	87.6	74.7	88.0
Proposed Combination 3-SUM (C2C5F6)
Proposed Combination 3-SUM (C3C5F6)
Proposed Combination 2-CON (C3C5F6)
Proposed Combination 3-CON (C2C5F6)

CON and SUM represent concatenation and summation feature fusion methods, respectively. Abbreviated symbol C2C5F6 represents that C2 layer, C5 layer, and F6 layer features for each band image are concatenated, the same as to others. Bold indicates the best one.