Author Affiliations
Department of Forensic Science and Technology, Zhejiang Police College, Hangzhou 310053, Zhejiang, Chinashow less
Fig. 1. Infrared transmission spectra of laser printing toner
Fig. 2. ROC and AUC of toner’s PLS-DA model. (a) Testing set ratio is 1/5; (b) testing set ratio is 1/4; (c) testing set ratio is 1/3; (d) testing set ratio is 1/2
Fig. 3. ROC and AUC of toner’s SVM model. (a) Testing set ratio is 1/5; (b) testing set ratio is 1/4; (c) testing set ratio is 1/3; (d) testing set ratio is 1/2
Fig. 4. ROC and AUC of toner’s RF model. (a) Testing set ratio is 1/5; (b) testing set ratio is 1/4; (c) testing set ratio is 1/3; (d) testing set ratio is 1/2
Fig. 5. All training samples’ 3D score plot of PLS-DA model
Fig. 6. S1 and S3 training samples’ 3D score plot of PLS-DA model
Fig. 7. Variable influence on projection (VIP) of toner’s PLS-DA model
Samples’ number | Brand | Type | Type of toner cartridge |
---|
S1 | HP | M202n | C0388C | S2 | Canon | LBP6018L | CRG 925 | S3 | Canon | LBP6230DN | CRG 326 | S4 | FUJI XEROX | P268B | CT202331 | S5 | HP | M226DN | C388A* | S6 | HP | M227 | C0388C | S7 | HP | P1106 | CF230A | S8 | FUJI XEROX | M188w | DR1035 |
|
Table 1. Information of laser printers
Preprocessing method | Number of main components | R2Y | Q2 | Accuracy |
---|
None | 12 | 0.865 | 0.747 | 0.875 | FD | 4 | 0.529 | 0.495 | 0.792 | SD | 10 | 0.921 | 0.810 | 1.000 | MSC | 9 | 0.713 | 0.639 | 0.833 | SNV | 3 | 0.398 | 0.386 | 0.500 | SG | 12 | 0.857 | 0.773 | 0.917 | FG+MSC | 11 | 0.871 | 0.761 | 1.000 | SD+MSC | 10 | 0.921 | 0.810 | 1.000 | FD+SNV | 5 | 0.614 | 0.560 | 0.958 | SD+SNV | 11 | 0.927 | 0.729 | 1.000 | FD+SG | 4 | 0.529 | 0.496 | 0.750 | SD+SG | 11 | 0.938 | 0.814 | 1.000 | FD+MSC+SG | 11 | 0.867 | 0.758 | 1.000 | SD+MSC+SG | 4 | 0.504 | 0.488 | 0.792 | FD+SNV+SG | 5 | 0.614 | 0.559 | 0.958 | SD+SNV+SG | 11 | 0.923 | 0.744 | 1.000 |
|
Table 2. Performance comparison of spectral preprocessing methods
Testing set ratio | PLS-DA | SVM | RF |
---|
1/5 | 1.000 | 0.961 | 0.999 | 1/4 | 1.000 | 0.953 | 0.998 | 1/3 | 0.998 | 0.942 | 0.977 | 1/2 | 0.994 | 0.934 | 0.972 |
|
Table 3. Micro-average AUC of different toner discriminant models
Testing set ratio | PLS-DA | SVM | RF |
---|
1/5 | 1.000 | 0.875 | 0.917 | 1/4 | 1.000 | 0.900 | 0.900 | 1/3 | 0.983 | 0.917 | 0.917 | 1/2 | 0.950 | 0.850 | 0.933 |
|
Table 4. Testing accuracy of different toner discriminant models
Testing set ratio | Full spectrum PLS-DA | VIP spectrum PLS-DA | Full spectrum SVM | VIP spectrum SVM | Full spectrum RF | VIP spectrum RF |
---|
1/5 | 1.000 | 0.995 | 0.961 | 0.984 | 0.999 | 0.998 | 1/4 | 1.000 | 0.994 | 0.953 | 0.980 | 0.998 | 0.998 | 1/3 | 0.998 | 0.993 | 0.942 | 0.975 | 0.977 | 0.996 | 1/2 | 0.994 | 0.984 | 0.934 | 0.972 | 0.972 | 0.994 |
|
Table 5. Micro-average AUC of models based on full spectrum and VIP spectrum