Identification of Tilletia foetida, Ustilago tritici, and Urocystis tritici Based on Near-Infrared Spectroscopy

Table 4

The optimal results except those with both the identification accuracy of the training set and the identification accuracy of the testing set equal to 100% when the BPNN models for Data treatment method 1 were built with different modeling ratios in each modeling spectral region.

Spectral region (cm^−1)	Modeling ratio of training set to testing set	The number of neurons in the hidden layer	Identification accuracy of training set (%)	Identification accuracy of testing set (%)
4000–12000	4 : 1	50	99.83	100
4000–6000	4 : 1	10	97.98	97.99
4000–6000	5 : 1	10	97.99	97.99
5000–6000	4 : 1	10	97.98	97.99
5000–6000	5 : 1	100	99.92	100
5000–9000	3 : 1	10	98.03	97.86
6000–7000	3 : 1	100	98.03	97.59
6000–7000	4 : 1	50	97.98	97.99
6000–7000	5 : 1	10	97.99	97.59
6000–8000	3 : 1	50	98.03	97.86
6000–9000	4 : 1	10	97.98	97.99
6000–10000	3 : 1	10	98.03	97.86
6000–10000	4 : 1	10	97.98	97.99
7000–8000	4 : 1	10	97.98	97.99
8000–9000	3 : 1	50	98.03	97.86
8000–9000	4 : 1	50	97.98	97.99
8000–9000	5 : 1	50	97.99	97.99
8000–10000	3 : 1	100	98.03	97.86
8000–10000	4 : 1	50	97.98	97.99
8000–10000	5 : 1	10	97.99	97.99
9000–10000	3 : 1	50	92.11	91.98
9000–10000	4 : 1	10	87.66	88.29
9000–10000	5 : 1	50	97.58	97.19