Fig. 1. Schematic diagram of analysis of different spectral bands and resolution
Fig. 2. Parameter selection of genetic algorithm
(a): Iteration parameters; (b): Population size; (c): Mutation probability; (d): Crossover probability
Fig. 3. Model prediction results of different spectral bands at the same resolution
Fig. 4. Model prediction results of different resolution at 1~2.5 μm spectral band
Training algorithm | Training function | RMSEP | Time /s |
---|
Bayesian regularization algorithm | trainbr | 1.62 | 130 | Quasi-Newton algorithm | trainbfg | 1.64 | 94 | Levenberg-Marquardt | trainlm | 3.93 | 5 | One-step secant algorithm | trainoss | 9.78 | 2 | Quantitative Conjugate Gradient | trainscg | 43.13 | 2 |
|
Table 1. Test results of different training functions
Hidden layer function | Output layer function | R | RMSEP | training time/s |
---|
logsig | logsig | 0 | 9.99 | 116 | purelin | logsig | 0 | 9.99 | 124 | tansig | logsig | 0 | 10 | 168 | logsig | purelin | 0.68 | 3.2 | 170 | tansig | purelin | 0.69 | 3.16 | 171 | logsig | tansig | 0.78 | 2.61 | 167 | tansig | tansig | 0.78 | 2.61 | 170 | purelin | purelin | 0.92 | 1.63 | 130 | purelin | tansig | 0.92 | 1.55 | 130 |
|
Table 2. Test results of different node transfer function