Remote Sensing Image Scene Classification Based on Fusion Method

Table 5

Model size and efficiency.

Mode	Method	Size (M)	Training (s)	Testing (s)
Front + back	Random crop+average	168	12.6	0.35
	Multiscale+average	168	12.6	0.35
	Multichannel+average	168	12.6	0.35
Front+mid	Random crop+single backbone	138	5.16	0.17
	Random crop+multibackbone	168	8.73	0.26
	Multiscale+single backbone	138	5.16	0.17
	Multiscale+multibackbone	168	8.73	0.26
	Multichannel+single backbone	138	5.16	0.17
	Multichannel+multibackbone	168	8.73	0.26
Mid + back	Single backbone+average	168	12.6	0.35
	Multibackbone+average	168	12.6	0.35
Front+mid + back	Random crop+single backbone+average	168	12.6	0.35
	Random crop+multibackbone+average	168	12.6	0.35
	Multiscale+single backbone+average	168	12.6	0.35
	Multiscale+multibackbone+average	168	12.6	0.35
	Multichannel+single backbone+average	168	12.6	0.35
	Multichannel+multibackbone+average	168	12.6	0.35