Table 3. Results of ensemble learning approach and other comparison algorithms.

LoG, DW and CNN represent Laplacian of Gaussian, discrete wavelet transform and convolutional neural network respectively. In Our Approach 1, parameters fine-tuned on our dataset were used while in Our Approach 2, parameters were fine-tuned for this dataset.

	Accuracy (%)	Feature Descriptor	Classification Model
Hassan et al. [26]	79.2	CNN features extracted from the unsegmented images	CNN
Peyret et al. [6]	88.3	Single-scale LBP	SVM
Peyret et al. [6]	91.3	Multi-scale LBP	SVM
Our Approach 1	97.8	[LBP,LPQ,GLCM,Hist, Perception, BSIF]	SVM
Chaddad et al. [7]	98.9	[LoG,DW,GLCM]	LDA
Our Approach 2	99.1	[LBP,LPQ,GLCM,Hist]	SVM
Hassan et al. [26]	99.2	CNN features extracted from the segmented images	CNN
Peyret et al. [24]	99.6	Stacked Multispectral Multi-scale LBP	SVM