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

Table 5

The prediction results (%) of convolutional neural networks running 10 times.

Data sets	Sample sets	Discriminant accuracy
Upper leaves	Training set	99.75, 99.87, 99.75, 99.75, 100, 100, 100, 100, 100, 100
	Testing set	95.86, 95.86, 96.15, 96.45, 95.86, 96.15, 96.45, 96.15, 96.45, 96.45
Middle leaves	Training set	100, 99.87, 100, 99.74, 99.61, 99.08, 98.95, 99.34, 99.61, 99.34
	Testing set	95.38, 94.77, 94.46, 94.77, 95.69, 95.69, 95.38, 95.69, 95.08, 95.08
Lower leaves	Training set	99.12, 99.75, 99.75, 99.62, 100, 99.75, 99.75, 99, 99.5, 99.75
	Testing set	96.49, 97.08, 97.37, 98.54, 97.66, 97.95, 96.2, 96.49, 97.37, 97.95