**4. Discussion**

A polarized multispectral for low-illumination-level imaging system was developed to monitor the plant status at night and evaluate stress-induced plant changes. The reflectance images of leaves on 680-nm and 760-nm narrow spectral lines were recorded by spectral technology, and the 0◦, 60◦ and 120◦ polarization images of vegetation were recorded by polarization technology. The NDVI, DoLP and AOP images of vegetation were calculated by Formulas (1), (3) and (4), respectively. Then, the NDVI, DoLP and AOP images were fused by Formula (5). For example, a DOLP image can reflect the smoothness and rich texture information of the leaf surface, an AOP image can reflect the smoothness of the leaf surface and a NDVI image contains the pathological information of vegetation. Therefore, if the NDVI, DOLP and AOP images are fused, the generated images will carry more abundant vegetation information.

The research on remote sensing of the vegetation health status based on the vegetation index has become a main research approach for scholars. However, for the study of the health status of vegetation at night, the vegetation index is greatly affected by the illumination [66–68]. As shown in Figure 6, the NDVI value decreases with the decrease of the environmental illumination. Therefore, it is not reliable to only use the NDVI as an indicator of the plant stress degree in a low illumination environment at night.

It can be seen clearly in Figure 8 that, with the increase of illumination, there is no significant change in the DoLP of vegetation. Therefore, the DoLP is hardly affected by the illumination of the experimental environment. In addition, polarized images have the potential to provide additional information about the target's shape, shadow, roughness and surface characteristics [69,70]. There are relatively few studies on polarization technology providing contrast enhancement information for the detection of vegetation health. Therefore, in this paper, we proposed the fusion of the vegetation index (NDVI), DoLP and AOP to better detect the physiological state of plants in a nighttime environment. As shown in Figures 9, 11 and 12d, different health states of plants correspond to different colors, so the proposed enhancement algorithm for nighttime plant detection has the effect of enhancing the contrast.

Under stress, the decrease of the NDVI ratio was related to the decrease of the SPAD content, and this proportional relationship could be explained by the optical characteristics of leaves. Compared with healthy leaves, those under stress had lower chlorophyll contents [71], and the reduced chlorophyll content led to a decreased chlorophyll absorption and enhanced reflectance at F680 nm. Moreover, the reflectance at F760 nm was reduced. According to Formula (1), the NDVI ratio was bound to decrease. In the NDVI images, the NDVI ratio of the stressed leaves was low (Figure 12a). Due to the fusion of polarization, the contrast of the fused images was enhanced, and leaf lesions could be identified more clearly (Figure 12d). It can be pointed out that the increase of the stress degree will lead to a lower vegetation index and leaf chlorosis [72,73].

Three-dimensional scatter plots (Figure 13a,b) were drawn using NDVI and NPSDI data obtained from the polarized multispectral low-light-level imaging monitoring system to classify different health status levels of plants. The classification discrimination accuracy is shown in Table 1. It can be seen from Table 1 that the sensitivity and specificity of the fusion image for the classification of different plant health conditions are generally close to the NDVI image data. It is important that the fusion image shows a color contrast, and the fusion image carries richer object information, so it is excellent to use the proposed fusion image algorithm to monitor the health of plants. In addition, the sensitivity and specificity of the fusion image (NDAI) to classify healthy leaves from stress level-1 were 89% and 92%, which was significant, as they demonstrate the potential of the fusion algorithm for detecting early symptoms of stressed plants at night.

The decrease of the chlorophyll and nitrogen contents under stress was also consistent with previous studies. We believe that the decrease of the chlorophyll content in stressed leaves may be caused by the degradation of chloroplasts in stressed leaves [74,75]. The scatter plots of the chlorophyll and nitrogen contents (Figure 15a,b) were similar to those of the NDVI and NPSDI and tended to decrease with the increase of the leaf stress degree. By drawing the distinguishing lines of different stress levels, the diagnostic accuracy of different stress levels can be determined (Table 1). It can be seen that the NDVI and NPSDI have similar diagnostic accuracies in distinguishing different levels of stress, while the SPAD has a higher correlation with stress and higher diagnostic accuracy. Furthermore, the NDVI determined through a spectra analysis follows a linear relationship with the SPAD and NC (Figure 16a,b), and the NPSDI obtained through a fusion algorithm for nighttime plant detection follows a linear relationship with the SPAD and NC (Figure 17a,b). There was also a linear correlation between the NDVI and NPSDI (Figure 18), which shows the potential of applying the fusion image determined by PMSIS in understanding the health status of outdoor plants at night from proximal sensing platforms.

In contrast to point monitoring devices such as spectroradiometers and hand-held chlorophyll analyzers, imaging devices provide spatial distribution information on the properties of the objects being detected. The advantage of PMSIS imaging is that it is a passive technology. It is a relatively inexpensive and compact instrument that can be easily carried anywhere and deployed on platforms (such as cars and drones) to feel the effects of various types of stress on farmlands and forests at night.
