| Crop growth parameters are important indicators for monitoring the growth of crops, and meanwhile, they are also important for predicting the yield of crops. Therefore, how to obtain these information of crops quickly and accurately becomes the basis of agricultural production. Traditional chemical method for crop growth parameter acquisition has the disadvantages of time consuming and high costs, furthermore, it can only obtain the point-source information and difficult to extend to the macro-scales, that greatly influenced the comprehensiveness, timeliness and objectivity of agronomic decision-making. The improvement of hyperspectral and imaging technology and its great advantage in quantitative analysis brought a new opportunity to solve the problem. However, the spectral information of upper leaf layer was usually obtained easily during the monitor of crop growth parameters by using remote sensing technology, whereas the diagnosis of these parameters in the middle and lower layers was ignored. This paper takes the wheat and corn as the main object of study, and extracted the vertical distribution of crop growth parameters by using remote sensing technology. Thus, the field hyperspectral data, hyperspectral image data, multi-angle observation and airborne remote sensing data were acquired based on the leaf, canopy and region scales, respectively. To extract the vertical distribution of crop growth parameters more quickly and accurately, we improved some methods and models, which can provided the technical support for the agricultural management department to make decisions. The major contents and results in this dissertation are as follows:(1) At leaf scale, we analyzed the vertical distribution of chlorophyll, carotenoid and nitrogen, systematically. On this basis, the spectral response mechanism of different leaf layers and the potential of vegetation indexes for extracting the vertical distribution of crop growth parameters were also investigated. The continuous wavelet transform as a noval method was used to extract the biochemical parameters of corn leaves in different layers. Our results showed that the highest estimation accuracy of Chla, Chlb and Chla+b was lower leaf layers, the coefficient of determinations (R2) were94.45%,92.77%and94.85%, compared with vegetation indexes, which were improved9.71%,9.92%and12.11%, respectively. To extract the vertical distribution of carotenoid and nitrogen, a few of new algorithms such as discrete wavelet transform, extraction of wavelet feature vector, genetic algorithm and neural network were adopted. These results showed that the estimation accuracy (R2) of nitrogen in different leaf layers were90.87%,91.75%and94.85%, compared with vegetation indexes, which were improved16.18%,15.68%and13.89%. Similar results were also acquired for the carotenoid, the coefficient of determinations (R2) were88.39%,94.90%and93.67%for different leaf layers, compared with vegetation indexes, which were improved15.14%,13.77%and13.27%, respectively.(2) At canopy scale, we employed multi-angle observation to extract the vertical distribution of wheat’s growth parameters in different leaf layers. For this purpose, we selected17vegetation indexes designed based on the red edge position to extract the LAI and chlorophyll content, and specially investigated the effect of observation angles on the estimation accuracy. For the first layer, REP had the highest estimation accuracy and its corresponding observation angles were0°and-40°. For the second and third layers, Chlred edge and MTCI were the good candidates to estimate the LAI and chlorophyll content, their observation angles were0°and-20°, respectively. Because the spectral absorption features of nitrogen and carotenoid overlapped with chlorophyll and water, we selected a few of vegetation indexes which had promising potential for estimating these two growth parameters. For the first layer, ARI and TBVI had good estimation accuracy, their corresponding observation angles were0°and-60°. For the second layer, RGI and TBVI obtained better estimation accuracy when the observation angle was0°. For the third layer, CRI and TBVI were the good candidates to estimate the nitrogen and carotenoid, their corresponding observation angles were0°and-60°. From such analysis, the conclusion can be drawed that zenithal and backward observation played an important role in the extraction of crop growth parameters.(3) At canopy scale, a multi-angle acquisition system of hyperspectral image developed by ourself was used to extract the corn’s LAI and chlorophyll content of different sowing dates. Hyperspectral image data had such advantage that it not only contains the spectral information, but also can generate images. Therefore, we extracted the hyperspectral reflectance from these images under different observation angles. After that, the BRDF of corn’s canopy was analyzed by ACRM model. It showed that the BRDF of blue and red wavebands had an obvious bowl-shape, while the near-infrared waveband presented a mound-shape. On this basis, we used three vegetation indexes constructed with simulation data to estimate corn’s LAI in different layers, and optimized them with combination of observation angles. Finally, such previous results were validated by the hyperspectral image data and the best estimation accuracy could be acquired for the upper layer, its coefficient of determination (R2) reached to0.80. To estimate the chlorophyll content in different layers, ACRM model were uesd to analyze the sensitivity of TCARI. After that, we proposed a new vegetation index HD-TCARI to estimate the chlorophyll content by improving TCARI with hot-dark spot index. The new vegetation index can reduce the effect of LAI by the chlorophyll and its estimation accuracy of different layers were0.67,0.58and0.42, respectively.(4) At regional scale, we attempted to monitor the crop growth parameters by CASI aerial remote sensing image. First, we take advantage of the high spatial resolution of CASI and extract the cultivated area of crops by supervised classification. The broadband and narrowband vegetation indexes were constructed with the spectral reflectance extracted from CASI image. And then, these vegetation indexes were compared with partial least square regression models for estimating the crop growth parameters. The result showed that broadband vegetation indexes presented more advantages on the estimation of crop growth parameters. For obtaining the optimal waveband to estimate the LAI, Chla, Chlb, Chla+b and N, an algorithm with combining any of two wavebands was used in this study. Finally, the monitor of crop growth parameters by CASI image was conducted by using the functions established between the vegetation indexes and growth parameters. |
PDF Full Text Request