Development of a composite 3D imaging technique using LiDAR and image sensors for estimating spatial distribution of plant chlrophyll content

概要

Forest plant function affects forest growth and is the role of energy balance, carbon exchange, and biogeochemical cycles. The function is directly or indirectly related with physiological factors involving leaf surface temperature and structure, water content, organ, and biochemical content. Physiological factors are intensively measured to achieve comprehensive mechanism of plant function and their responds in different environmental condition in forestry field. However, the result of those factors responds are subject to environmental acclimation, as the process of regulating plant function and governed by microenvironment in form of different availability in light, water, and atmospheric temperature. The spatial and temporal dynamics between microenvironments may cause inconsistent response in patterns of physiological factors. This may lead wrong interpretation of the relationship between physiological factors and plant function, as physiological factors were not carefully conducted in the same spatial and time sequence.

Moreover, to date, there is an increase of needs to acquired physiological factors according to spatial attribution consisting of consistent plant physiological responds for plant model, which aims to describe the physiological or physical interaction between the biosphere and atmosphere accurately. However, neither field measurement nor semi-theoretic model approaches were compromise of the assumption of spatial and time consistency, even when it was inconsistent with the facts. Although there are increasing attentions that take those spatial and temporal dynamics into account in forestry works, physical uncertainties in majorities of field measurement limit to approach more details and accurate interpretation of plant function.

In remote sensing communities, 2D image technology is widely used to monitor physiological factors, because sensors can quickly reproduce data, which improves the investigation on physiological factors in the same time sequence. On other hand, there has been attention on using terrestrial LiDAR (Light Detection and Ranging) to capture canopy spatial details and continues spatial attribution of plant down to millimeter-level resolutions. By retrieving spatial information, it allows perceiving physiological factors under the same spatial dynamics.

Despite there have been already some technologies that is possible to improve conduct physiological factors with 3D spatial attribution recoding, they are inelastic and not commercially available. Hence, to overcome primary issues of data acquisition in forestry field and remote sensing communities, the overall goal of the research is to develop a simple method that can simultaneously obtain physiological factors of the plant and 3D spatial attribution of plant for (1) more accurate study of physiological factors that perceive plant function, (2) monitoring temporal dynamics of physiological factors, and (3) proper estimation of plant model parameters This thesis consists of 5 chapters. In chapter 1 contained brief introduction of challenges of conducing appropriately on environmental acclimation factors and the techniques that is possible to apply to overcome issues in forestry field. Moreover, outline and relationship between research approaches would be emphasized in the end of the chapter.

In Chapter 2, the approach of estimating chlorophyll content (Chl) based on spectral images is proposed by considering specular and shading effect. Chl seems to be one of the main contributors to plant function in the light adaptation, rational use of solar energy and water stress by proportional adjustment of quantities based on light availability. The vegetation indices (VIs) are intensively studied to estimate Chl content; however, optically anisotropic reflectance is strongly dependent on the specular component, primarily produced by leaf surface heterogeneity, which leads same leaf to have different VIs value. In addition, on the ground-based measurement, incident light fluctuations and shading, which is inconsistent with the region would make VIs variable under the same leaf. Specular and shading effect might lead inaccurate Chl estimation when VIs are not regularly increase or decrease with changes of Chl content. In our study, we used existed optical model that could describe interaction of light reflectance with Chl content, in order to perceive the specular and shading effect on VIs based Chl estimation. In addition, due to the wide range of chlorophyll content in our samples, a “green” normalized vegetation index (GNDVI) suitable for a wide range of chlorophyll content is used. Two sampled species, Japanese stone oak and Bamboo leaf oak with distinct leaf surface heterogeneity and lighting condition have been selected to verify our approach. As a result, GNDVI-based Chl estimation in both sampled tree has greatly improved by R-squared value from 0.04 to 0.81 (Japanese stone oak) and 0.55 to 0.80 (Bamboo leaf oak). Moreover, it was found that not only the total Chl content but also Chla and Chlb can be estimated by our method.

In Chapter 2, 2D image technology has been introduced to retrieve accurate photosynthesisrelated physiological factors, total Chl, Chla, and Chlb. but it cannot meet needs of obtaining the 3D spatial attributes of plants. Therefore, in Chapter 3, 2D imaging technology was combined with LiDAR to simultaneously acquire Chl-based pigments and plant 3D attributes. To obtain Chl-based pigments composited with 3D attributes from LiDAR projection images, a fusion method based on projective transformation was employed. Besides, the spatial fusion errors were examined in different 2D images taken by different focal lengths. Subsequently, experiments were conduct on a sampled tree Japanese stone oak seasonally compared with another sampled tree, bamboo leaf oak, under different lighting conditions but same season in winter. Moreover, a spatial analysis of the Chl distribution would be performed to observe the behavior of Chl under different light availability. The results showed images taken by focal length 35 mm with less lens distortion has high spatial accuracy for image fusion with LiDAR projected images, and minimal error is 0.55 cm in reality. 3D spatial analysis of seasonal variation of leaf pigments including vertical and horizontal distribution was fulfill; moreover, 3D spatial analysis also helps to reduce spatial fusion errors. Besides, unlike other fusion methods or multispectral LiDAR (limited to the application of a single physiological factor of plants), this fusion method can be combined with various 2D imaging technologies to accomplish research on multiple physiological factors.

Chl with recording of 3D attributes based on the fusion method is accomplished in Chapter 3. In Chapter 4, we associate Chl with leaf angle, a structural characteristic of plants, because leaf angle is also a physiological factor that regulates light availability. Regulatory network based on plant function controls the behavior of arrangement in plant leaf angle and Chl content. Thus, it is critical to take the variability of leaf angle distribution into account in a remote sensing analysis of a system of plant function. Due to the physical limitations of field measurements, it is difficult to obtain leaf angles quickly and accurately, especially with a complicated canopy structure. An application of terrestrial LiDAR (Light Detection and Ranging) is a common solution for the purposes of leaf angle estimation, and it allows for the measurement and reconstruction of 3D canopy models with an arbitrary volume of leaves. However, in most cases, the leaf angle is estimated incorrectly due to inaccurate leaf segmentation. Therefore, the objective of this study was an emphasis on the development of efficient segmentation algorithms for accurate leaf angle estimation. Our study demonstrates a leaf segmentation approach based on a k-means algorithm coupled with an octree structure and the subsequent application of plane-fitting to estimate the leaf angle. Furthermore, the accuracy of the segmentation and leaf angle estimation was verified. The results showed average segmentation accuracies of 95% and 90% and absolute angular errors of 3° and 6° in the leaves sampled from mochi and Japanese camellia trees, respectively. Moreover, the comparison between chlorophyll content and arrangement of leaf angle were compared in the end. It is our conclusion that our method of leaf angle estimation has high potential and is expected to make a significant contribution to future plant and forest research. Moreover, the combination of fusion method in the chapter 3 with leaf angle estimation method in chapter 4 provides a way of perceiving on the behavior of plant function on leaf structural development and their responds of Chl content.

The last chapter, Chapter 5 will be the conclusion and the future works of the study