*3.3. Outdoor Experiment at Night*

The developed polarized multispectral for a low-illumination-level imaging system was used in outdoor experiments, and as measured by the illuminator, the ambient illumination was 0.22 lux. As shown in the figure below, Figure 12a is the NDVI image calculated by Formula (1), Figure 12b the AOP image by Formula (4), Figure 12c the DoLP image by Formula (3), Figure 12d the NDAI image by Formula (5) and Figure 12e the gray-scale image. Figure 12e1–e4 are four different health state regions of plants, respectively.

**Figure 12.** (**a**) NDVI image, (**b**) AOP image, (**c**) DoLP image, (**d**) fusion image (NDAI), (**e**) gray-scale image, (**e1**) healthy leaf region, (**e2**) level-2 stress leaf region, (**e3**) level-1 stress leaf region and (**e4**) withered leaf region.

The spectral image, polarization image and fusion image (NDAI) were obtained through PMSIS. Four regions: healthy leaf region (Figure 12e1), level-2 stress leaf region (Figure 12e2), level-1 stress leaf region (Figure 12e3) and withered leaf region (Figure 12e4) were selected to draw the scatter graph of the discriminant function.

Figure 13a shows the 3D scatter plot of the NDVI image obtained from the F680 nm and F760 nm bands of the selected location points in a specific region, in which the X-axis stands for the health status level of vegetation, the Y-axis the number of groups and the Zaxis the NDVI value. The change trend of vegetation health can be seen intuitively through the change of color and trend of the curved surface. The curved surface is projected on the XY plane and the YZ plane. The cut-off lines are drawn from the arithmetic mean value of the status grades of adjacent vegetation on the YZ plane. For example, a cut-off line with a value of 0.196 distinguished a withered leaf from a level-2 stressed leaf, one with a value of 0.494 distinguished a level-1 stressed leaf from a level-2 stressed leaf and one with a value of 0.773 of distinguished a healthy leaf from a level-1 stressed leaf. According to the misclassification rate of the adjacent plant states, the sensitivity and specificity of the classification of different plant health states were determined, and the positive predictive value (PPV) and negative predictive value (NPV) of the classification are shown in Table 1.

**Table 1.** Discrimination accuracy between different grades of the plant health states obtained with the NDVI, NPSDI, nitrogen content (NC) and PSAD.


Notes: Se (Sensitivity) = true positive/(true positive + false negative), Sp (Specificity) = true negative/(true negative + false positive), PPV (positive predictive value) = true positive/(true positive + false positive) and NPV (negative predictive value) = true negative/(true negative + false negative).

According to the NDVI, DoLP and AOP image data, the image fusion was carried out, and the 3D scatter plot of the NPSDI was made in the selected four feature regions; the X-axis stands for the health status level of vegetation, the Y-axis the number of groups and the Z-axis the NPSDI value, as shown in Figure 13b. In the YZ plane, the cut-off line between a withered leaf and level-2 stressed leaf laid at 0.153 and that between a level-1 stressed leaf and level-2 stressed leaf and that between a healthy leaf and level-1 stressed leaf laid at 0.399 and 0.574, respectively. Table 1 also contains the sensitivity and specificity for classifying different levels of the plant health status according to the scatter plot shown in Figure 13b, as well as the corresponding PPV and NPV.

In the four plant health state regions, 51 pixels were selected and averaged to draw the NDVI, NPSDI and DoLP histogram of the four regions. As shown in Figure 14, the NDVI values show an increasing trend with the improvement of the plant health status, and the NPSDI values show the same trend. This also indicates that there is a correlation between the NDVI and NPSDI.

**Figure 13.** (**a**) 3D scatter plot of the NDVI. (**b**) 3D scatter plot of the NPSDI.

The reason for the higher value of the DoLP in the level-1 stress leaf region may lie in that it is also the leaf shadow region. Since the target was inversely proportional to the DoLP and surface reflectance [65], the DoLP of the shadow area is higher, which is helpful for monitoring the shadow area. The value of the DoLP in the withered leaf region was higher than that in the healthy leaf region. The reason may lie in that the completely withered leaves with more diffuse reflections lead to a higher DoLP value. Under low night illumination, the healthy leaves are smooth, but the mirror reflection is low, so the DoLP value is lower.

The NC and SPAD were measured for spotted laurel with leaves at varying levels of health status. It was observed that the NC of healthy leaves was above 16.8 mg/g, and it decreased to a value between 10.61 and 16.8 mg/g for level-1 stress leaves. The NC of level-2 stress leaves was in the range of 3.53–10.61 mg/g and that of withered leaves was below 3.53 mg/g. In comparison, the SPAD content of healthy leaves was above 44.71, that of level-1 stress leaves in the range of 25.22–44.71, that of level-2 stress leaves in the range of 7.01–25.22 and that of withered leaves below 7.01. The three-dimensional scatter plots of the NC and SPAD contents in the leaves of different health states are shown in Figure 15a,b. The sensitivity and specificity of using the NC and SPAD to distinguish different levels of the plant health status, as well as the respective PPV and NPV values, are shown in Table 1.

**Figure 14.** The NDVI, NPSDI and DoLP histograms of the plant in the four health status regions.

The NDVI image obtained through PMSIS system imaging and the calculated fusion image (NDAI) have a significant positive linear correlation with the NC and SPAD, and so do the NDVI and NPSDI (Figures 16a,b and 17a,b). The results of the different health states showed that the correlation coefficient between the NC and NDVI was the highest (R<sup>2</sup> = 0.931) and that of the SPAD under the same NDVI was 0.911. In comparison, the R<sup>2</sup> values of the NPSDI with the SPAD and NC of the plant leaves were 0.882 and 0.916, respectively.

The relationship between the NDVI and NPSDI obtained using PMSIS was studied under different health states of plants, separately. As seen from Figure 18, there is a positive linear relationship between the NDVI and NPSDI (R<sup>2</sup> = 0.968), which can be used to evaluate stress-induced changes in plants.

**Figure 15.** (**a**) 3D scatter plot of the chlorophyll content. (**b**) 3D scatter plot of the nitrogen content.

**Figure 16.** (**a**) Correlation between the NDVI and SPAD. (**b**) Correlation between the NDVI and NC.

**Figure 17.** (**a**) Correlation between the NDVI and SPAD. (**b**) Correlation between the NDVI and NC.

**Figure 18.** Correlation between the NDVI and NPSDI.
