Table 4.

Classification performance comparison of our proposed method with the other handcrafted feature–based methods.

Feature descriptor and classifier			Accuracy, %		F1, %		mAPa, %		mARb, %
Local binary pattern [67]
	AdaBoostM2	35.74		27.70		35.74		22.61
	Multi-SVMc	43.84		42.35		42.99		41.72
	RFd	57.10		53.85		54.79		52.95
	KNNe	50.46		47.36		46.86		47.87
Histogram of oriented gradients [68]
	AdaBoostM2	39.35		32.86		39.35		28.22
	Multi-SVM	49.84		53.80		67.39		44.88
	RF	61.41		63.19		68.66		58.55
	KNN	53.20		54.68		58.41		51.45
Multilevel local binary pattern [69]
	AdaBoostM2	44.02		37.45		44.02		32.59
	Multi-SVM	55.47		53.10		54.75		51.55
	RF	61.40		57.57		59.08		56.13
	KNN	55.40		52.20		52.06		52.33
Proposed feature descriptor (DenseNet + LSTMf +PCAg)
	AdaBoostM2	93.39		93.66		94.35		92.98
	Multi-SVM	95.50		96.43		97.98		94.96
	RF	81.16		82.96		84.48		81.55
	KNN	96.19		96.99		98.18		95.86

^amAP: mean average precision.

^bmAR: mean average recall.

^cSVM: support vector machine.

^dRF: random forest.

^eKNN: k-nearest neighbor.

^fLSTM: long short-term memory.

^gPCA: principal component analysis.