A Novel Deep Convolutional Neural Network Based on ResNet-18 and Transfer Learning for Detection of Wood Knot Defects

Table 7

Wood knots classification results of ReSENet-18 and its component networks.

Method	Actual category	Predict category
		Decayed knot	Dry knot	Edge knot	Encased knot	Horn knot	Leaf knot	Sound knot	Total
ReSENet-18 (our method)	Decayed knot	18	0	0	0	0	0	1	19
	Dry knot	0	96	0	0	0	0	0	96
	Edge knot	0	0	91	0	0	0	0	91
	Encased knot	0	0	2	38	0	0	0	40
	Horn knot	0	0	0	0	49	0	0	49
	Leaf knot	0	0	0	0	0	65	1	66
	Sound knot	0	0	0	0	0	2	248	250
ResNet-18	Decayed knot	17	0	1	0	0	0	1	19
	Dry knot	4	86	1	2	0	0	3	96
	Edge knot	0	2	89	0	0	0	0	91
	Encased knot	2	2	0	33	0	0	3	40
	Horn knot	0	0	1	0	47	0	1	49
	Leaf knot	0	3	0	0	19	42	2	66
	Sound knot	4	5	0	0	0	0	241	250
SENet	Decayed knot	16	0	1	0	0	0	2	19
	Dry knot	0	95	1	0	0	0	0	96
	Edge knot	0	1	89	0	0	0	1	91
	Encased knot	0	1	1	36	0	0	2	40
	Horn knot	0	0	1	0	48	0	0	49
	Leaf knot	0	0	0	0	1	65	0	66
	Sound knot	0	1	0	0	0	2	247	250