DWTFormer: a frequency-spatial features fusion model for tomato leaf disease identification

Table 5 Results of different Bnecks and activation functions

Activation Function	Bneck-DSM	Accuracy/%	F1-score/%	Params/M	FLOPs/G
Dynamic Shifted Max	(2,2,3,2,1)	99.28	99.10	33.69	7.96
	(2,2,3,4,2)	99.51	99.44	43.83	13.45
	(2,2,3,6,2)	99.62	99.68	45.26	16.75
Leaky ReLU [45]	(2,2,3,2,1)	96.72	96.15	33.69	7.96
	(2,2,3,4,2)	98.91	98.77	43.83	16.75
	(2,2,3,6,2)	99.65	99.02	45.26	9.25
ReLU [46]	(2,2,3,2,1)	95.01	95.77	33.69	7.96
	(2,2,3,4,2)	97.99	97.11	43.83	13.45
	(2,2,3,6,2)	99.87	98.20	45.26	16.75

“(2,2,3,2,1)” represented the Bneck-DSM number in each stage was set to 2,2,3,2,1. Similarly, “(2,2,3,4,2)” and “(2,2,3,6,2)” represented the Bneck-DSM number in each stage, which was set to 2,2,3,4,2 and 2,2,3,6,2. The results of Bneck-DSM with Dynamic Shifted Max and the best results of Bneck-DSM with ReLU and Leaky ReLU are highlighted in bold

ISSN: 1746-4811