Computer vision-based wood identification and its expansion and contribution potentials in wood science: A review

Table 6 Performances of deep learning models reported in CV-based wood identification studies

References	Dataset	Image type	#SP/#IMG	CNN model	CLS %
Hafemann et al. [29]	UFPR	Macro	41/2942	3-ConvNet^a	95.8
Hafemann et al. [29]	UFPR	Micro	112/2240	3-ConvNet^a	97.3
Kwon et al. [96]	Softwoods	Macro	5/16,865	LeNet	99.3
Kwon et al. [64]	Softwoods	Macro	5/33,815	Ensemble of LeNet2, LeNet3, and MiniVGG4	0.98^b
Ravindran et al. [81]	Meliaceae species	Stereo	10/2303	VGG16	88.7^c
Ravindran et al. [81]	Meliaceae species	Stereo	10/2303	VGG16	97.5^d
Tang et al. [49]	FRIM collection	Stereo	100/101,446	SqueezeNet	77.5
Lopes et al. [51]	FWRC collection	Stereo	10/1869	InceptionV4_ResNetV2	92.6
Lens et al. [72]	UFPR	Micro	112/2240	ResNet101	96.4
de Geus et al. [55]	Brazilian species	Stereo	281/–	DenseNet	98.8
Ravindran [21]	Melaceae species	Stereo	10/–^e	ResNet34	81.9
Ravindran [21]	Melaceae species	Stereo	10/–^e	ResNet34	96.1
Fabijanska et al. [65]	European species	Macro	14/312	Residual convolutional encoder network	98.7

#SP: number of species; #IMG: number of images; CLS %: classification accuracy; FWRC: Forest and Wildlife Research Center at Mississippi State University; –: no number specified
^a3 layers deep CNN
^bF1 score
^cSpecies identification
^dGenus identification
^eAt least 5 images per specimen (total 193 wood specimens)

ISSN: 1746-4811