FWDNet: A Novel Recognition Network for Ferrography Wear Debris Image Analysis

Table 1

The framework of 5 DCNN models.

	AlexNet (8 layers)	VGG16 (16 layers)	InceptionV3 (22 layers)	ResNet50 (50 layers)	DenseNet121 (121 layers)
Input size
CONV block 1	, 96	, 64; , 64; Max. pool, 64	, 32; , 32; , 64; Max. pool, 64	, 64; Max. pool, 64	, 64; Max. pool, 64
CONV block 2	, 256	, 128; , 128; Max. pool, 128	, 64; , 80; , 192		Avg. pool, 128
CONV block 3	, 384	,256; , 256; , 256; Max. pool, 256	Inception A, 288; Inception A, 288		Avg. pool, 256
CONV block 4	, 384	, 512; , 512; , 512; Max. pool, 512	Inception B, 768; Inception B, 768; Inception B, 768; Inception B, 768; Inception B, 768		Avg. pool, 512
CONV block 5	, 256	, 512; , 512; , 512; Max. pool, 512	Inception C, 1280; Inception C, 1280
Dense net	Fc, 4096; Fc, 4096; Fc, 1000	Fc, 4096; Fc, 4096; Fc, 1000	Avg. pool, 2048; Fc, 1000	Avg. pool, 2048; Fc, 1000	Avg. pool, 1024; Fc, 1000