Wood Recognition and Quality Imaging Inspection Systems

Table 2

Comparison of various works focused on wood defect detection methods, based on various features/classifiers.

Reference	Year	Methods	Features	Classifier	Categories	Dataset (train./valid./test)	Accuracy (%)
[26]	2006	—	Unique	BNN/BANN	2	232 (80/20/-)	86.52
[30]	2009	—	Color+size	SVM	5	1200 (66/34/-)	96.50
[31]	2009	Gabor filters+AMMLP	Gabor filters	AMMLP	3	100 (52/48/-)	97.91
[33]	2009	Color+GLCM+own+SVM/MLP	Color+GLCM+own	SVM	10	2200 (50/25/25)	91.39
[29]	2016	LBP/GLCM+NN	LBP	NN	52	220 (82/18/-)	93.30
[27]	2017	PLS/PLS-DA/LS-SVM/BPNN	Nonimage	BPNN	4	360 (66/34/-)	97.50
[34]	2017	DPLS/PCA-DPLS/BPNN	Nonimage	BPNN	4	400 (75/25/-)	92.00
[32]	2018	HOG/FREAK/SURF+BPNN	HOG	BPNN	4	150 (60/40/-)	90.82
[55]	2018	—	—	CNN	18	839 (-)	91.55
[57]	2019	—	—	Faster R-CNN	4	353+aug. (-)	96.10

AMMLP: artificial metaplasticity multilayer perceptron; BANN: Bees Algorithm Neural Network; BPNN: Backpropagation Neural Network; DPLS: Discriminant Partial Least Squares; FREAK: Fast Retina Keypoint; GLCM: gray level covariance matrix; HOG: Histogram of Oriented Gradients; LBP: Local Binary Patterns; LS-SVM: least squares support vector machines; MLP: multilayer perceptron; NN: neural network; PCA-DPLS: Principal Component Analysis-Discriminant Partial Least Squares; PLS: Partial Least Squares; PLS-DA: Partial Least Squares and Discriminant Analysis; PSO-GA: particle swarm-genetic hybrid algorithm; SURF: Speed-up Robust Feature; SVM: support vector machine; : an indication that dataset is available online for free.