Discrimination of Fresh Tobacco Leaves with Different Maturity Levels by Near-Infrared (NIR) Spectroscopy and Deep Learning

Table 6

The prediction results (%) of convolutional neural networks and other four methods.

Data sets	Sample sets	KNN	BPNN	SVM	ELM	CNN
Upper leaves	Training set	89.87	92.11 ± 0.46	96.2	96.11 ± 1.82	99.91 ± 0.12
	Testing set	84.02	66.39 ± 7.31	91.72	87.57 ± 2.79	96.18 ± 0.26
Middle leaves	Training set	90.79	93.66 ± 0.79	93.03	94.24 ± 2.1	99.55 ± 0.37
	Testing set	84.92	80.18 ± 2.93	89.23	86.71 ± 1.16	95.2 ± 0.44
Lower leaves	Training set	91.99	95.87 ± 0.43	94.87	95.71 ± 2.28	99.6 ± 0.31
	Testing set	89.77	86.81 ± 4.06	93.57	92.51 ± 2.12	97.31 ± 0.75