Applying Neural Networks to Hyperspectral and Multispectral Field Data for Discrimination of Cruciferous Weeds in Winter Crops

Table 4

Percentage of correct classification of cruciferous weeds, and wheat and broad bean crops for multispectral wavebands and spectral vegetation indices obtained from STEPDISC analysis, and from MLP and RBF neural networks.

		Wavebands (nm) and spectral vegetation indices
Years	Classification methods	All bands (450–900) and spectral vegetation indices	B (450–520)	G (521–600)	R (630–690)	NIR (760–900)	NDVI	RVI	B/G	R/B	R/G	NIR/B	NIR/G
						(% correct classification)
2007	STEPDISC	98.0	52.7	71.6	66.9	40.5	55.1	56.8	68.9	80.1	59.5	56.1	56.8
	MLP	100	55.9	70.4	65.5	60.0	72.0	74.2	55.9	82.1	71.4	75.0	56.7
	RBF	92.1	63.3	53.3	78.9	60.0	70.6	52.6	54.2	84.4	62.1	62.1	42.3
2008	STEPDISC	100	81.5	92.1	92.1	100	100	100	100	60.1	100	100	100
	MLP	100	79.5	94.6	88.2	100	100	100	100	81.8	100	100	100
	RBF	100	77.6	90.6	84.8	100	100	100	100	71.9	100	100	100
2009	STEPDISC	99.3	89.0	92.7	94.1	74.2	75.1	91.2	91.8	94.2	85.1	87.4	93.2
	MLP	99.4	90.5	94.3	95.6	80.1	87.1	94.9	90.3	94.5	86.0	93.3	95.9
	RBF	98.9	89.8	92.9	98.3	83.6	90.6	96.6	89.9	92.0	86.2	92.4	97.8
2010	STEPDISC	98.7	42.1	48.3	48.9	57.3	51.2	54.9	77.5	65.5	66.2	32.6	50.5
	MLP	98.1	58.1	60.1	55.9	63.2	61.6	63.4	75.8	79.1	67.8	54.9	66.2
	RBF	94.4	54.5	56.8	58.4	64.3	60.2	63.9	78.3	78.3	71.5	48.2	66.3