5.2. Pattern classification
The impact of polymer selection was examined by applying principal component analysis (PCA) to the sensor array response matrix (vapors in rows, sensor array outputs in columns). For this analysis the sensor array responses were normalized with respect to vapor concentration and logarithmically scaled for the reasons explained in [35,119]. The logarithmic scaling of SAW sensor responses has been found to be always enhancing the separability of vapor classes in principal component space, hence improve classification efficiency [35]. Fig. 1 shows the principal component score plots for all headspace VOCs over the milk as listed in Table 1. The regions demarcated by close curve boundary contain data points mostly associated with the fresh milk VOCs barring a few exceptions near the boundary. Most notable exceptions are the spoilage markers hexanoic acid and octanoic acid points in PC1-PC2 plot Fig. 1(a). These are seen to be within the fresh VOCs boundary even though they appear as compact clusters. However, these clusters are well separated from the freshness markers VOCs in PC3 direction as clearly seen in Fig. 1(b). The three other clusters (octanal, nonanal, 2 nonanone) which occur outside the freshness marker boundary in PC1-PC2 plot are seen to be inside the boundary in PC1-PC3 plot. That is, in 2D principal component projections some spoilage VOCs appear which to occupy overlapping space with the freshness VOCs are actually well separated. This becomes quite evident if we examine 3D plot PC1-PC2-PC3 in Fig. 2. It can be seen that the entire freshness marker VOCs are separated from the entire spoilage marker VOCs. This plot is shown with a camera view in MatLab from an angle where the full separation between the two types of VOCs look convincing. The freshness VOCs are shown encircled. These results suggest that by pooling the entire freshness marker VOCs as a single class the state of milk freshness can be ascertained by discriminating it against the spoilage marker VOCs. A striking feature in these plots is the occurrence of indole (a freshness indicator) as a well formed cluster distinctly separate from the rest. This suggests that the discrimination of indole
5.2 รูปประเภทThe impact of polymer selection was examined by applying principal component analysis (PCA) to the sensor array response matrix (vapors in rows, sensor array outputs in columns). For this analysis the sensor array responses were normalized with respect to vapor concentration and logarithmically scaled for the reasons explained in [35,119]. The logarithmic scaling of SAW sensor responses has been found to be always enhancing the separability of vapor classes in principal component space, hence improve classification efficiency [35]. Fig. 1 shows the principal component score plots for all headspace VOCs over the milk as listed in Table 1. The regions demarcated by close curve boundary contain data points mostly associated with the fresh milk VOCs barring a few exceptions near the boundary. Most notable exceptions are the spoilage markers hexanoic acid and octanoic acid points in PC1-PC2 plot Fig. 1(a). These are seen to be within the fresh VOCs boundary even though they appear as compact clusters. However, these clusters are well separated from the freshness markers VOCs in PC3 direction as clearly seen in Fig. 1(b). The three other clusters (octanal, nonanal, 2 nonanone) which occur outside the freshness marker boundary in PC1-PC2 plot are seen to be inside the boundary in PC1-PC3 plot. That is, in 2D principal component projections some spoilage VOCs appear which to occupy overlapping space with the freshness VOCs are actually well separated. This becomes quite evident if we examine 3D plot PC1-PC2-PC3 in Fig. 2. It can be seen that the entire freshness marker VOCs are separated from the entire spoilage marker VOCs. This plot is shown with a camera view in MatLab from an angle where the full separation between the two types of VOCs look convincing. The freshness VOCs are shown encircled. These results suggest that by pooling the entire freshness marker VOCs as a single class the state of milk freshness can be ascertained by discriminating it against the spoilage marker VOCs. A striking feature in these plots is the occurrence of indole (a freshness indicator) as a well formed cluster distinctly separate from the rest. This suggests that the discrimination of indole
การแปล กรุณารอสักครู่..
