[Estimating polysaccharide content with hyperspectral in Dendrobium of ficinale].

The objectives in the present paper were to study the relationship between the polysaccharide content and the spectrum, aimed at finding a non destructive testing method for the measurement of polysaccharide content Materials in this study were from tissue culture seedlings under different treatments and domesticated plants in different growth stages. In the study, 36 samples were used to build estimation models and another 11 as test samples to examine the models. The relationship between the spectrum and polysaccharide content was investigated by partial least squares regression (PLSR) and factor analysis. The results show that (1) there was stronger correlation between derivative reflectance and polysaccharide content, and the sensitive wavebands mainly concentrated in the visible region. (2) The PLSR model has a high precision, while its prediction accuracy is lower. The models of factor analysis have higher prediction accuracy, the RPD of the model based on spectral reflectance is 2.269 and that of derivative reflectance is 2.305.

[Hyperspectral feature band selection based on mean confidence interval and tree species discrimination].

In the present study, based on the leaf-level hyperspectral data of BaiMu, LeiZhu and WuHuanZi, the authors come up with two solutions through the theory of statistics; the first one is that optimal discriminating band between tree species is extracted by mean interval confidence, the other one is that tree species is discriminated by the Manhattan distance and the Min Max interval similarity. The research results showed that (1) the optimal discriminating bands between BaiMu and LeiZhu are around 350-446, 497-527, 553-1 330, 1 355-2 400 and 2 436-2 500 nm; the optimal discriminating bands between BaiMu and WuHuanZi are around 434-555, 580-1 903, 1 914-2 089, 2 172-2 457 and 2 475-2 500 nm; the optimal discriminating bands between LeiZhu and WuHuanZi are around 434-555, 580-1 903, 1 914-2 089, 2 172-2 457 and 2 475-2 500 nm; and this result is helpful for us to find maximum difference to identifying tree species respectively. (2) In these optimal discriminating bands, we find that the Manhattan distance between the same species is far less than the different species; but the Min-Max interval similarity between the same species is far more than the different species, so this result could help us to discriminate and identify different types of tree species effectively.

[Hyperspectral Bambusoideae discrimination based on Mann-Whitney non-parametric test and SVM].

In the present study, based on the leaf-level hyperspectral data of MaoZhu, LeiZhu and XiaoShunZhu, We come up with two solutions to discrimination through the theory of non-parametric test and pattern recognition; the first one is that optimal discriminating band between bambusoideae species is extracted by Mann-Whitney non-parametric test, the other is that bambusoideae species is discriminated by the support vector machine. The research results showed that (1) the optimal discriminating band between MaoZhu and LeiZhu is around 503-655, 689-732, 757-1 000, 1 038-1 084, 1 238-1 311, 1 404-1 591, 1682-1 800, 1 856-1 904, and 1 923-2 500 nm; the optimal discriminating band between MaoZhu and XiaoShunZhu is around 350-386, 731-1 430, 1 584-1 687, and 1 796-1 873 nm; the optimal discriminating band between LeiZhu and XiaoShunZhu is around 355-356, 498-662, 689-745, and 1 344-2 500 nm; and it can eliminate 30.0%, 57.7%, and 35.8% of the invalid distinction between bands by Mann-Whitney non-parametric test method. (2) In these optimal discriminating bands, we found that the accuracy of bambusoideae discrimination is 98.4%, 93.5%, and 95.1%, the generalization accuracy is 93.3%, 90.0%, and 86.7% by sequential minimal optimization algorithm. It indicates that this method is valid for selecting feature band and discriminating bambusoideae species.

[Using fractal dimensions of hyperspectral curves to analyze the healthy status of vegetation].

The reflectance spectral curves of leaves can reflect many information of vegetation growth, and its variation maybe means that the healthy status of vegetation will change. Many spectral feature parameters such as red edge position, height of green peak, depth of red band absorption, the area of red edge and some vegetation index have been used to describe this change. However, the change of vegetation healthy status is not some feature parameters, but a comprehensive variation of the whole curve. So, a comprehensive index maybe has more value to describe the change of hyperspectral curve of vegetation and indicates its healthy status. Fractal is an appropriate mathematical tool, and fractal dimension can be used to explain the comprehensive variation of a curve. Therefore, in the present study, fractal theory was used to analyze the healthy status of different vegetation. Firstly, analytical spectral devices (ASD) were used to measure the hyperspectral curves of different vegetations with different healthy status. Secondly, spectral curves were analyzed, and some parameters which can really reflect different healthy status were obtained. Finally, the fractal dimension of reflectance spectral curves inside a spectral band zone between 450 and 780nm was computed by variation method, and the relationship between fractal dimensions and spectral feature parameters was established. The research results showed that (1) the hyperspectral curves of vegetation have fractal feature, and their fractal dimensions gradually decrease with the health deterioration of leaves, (2) fractal dimension has positive correlation with the height of green peak, the depth of red band absorption and the area of red edge, (3) multivariate analysis showed that fractal dimensions have a significant linear relationship with the three spectral feature parameters just mentioned above. So, the fractal dimension of hyperspectral curve can serve as a new comprehensive parameter to analyze quantitatively the healthy status of vegetations.

[Study on relationships between total chlorophyll with hyperspectral features for leaves of Pinus massoniana forest].

In the present study, the authors built the relationships between the total chlorophyll and hyperspectral features of P. massoniana. The research results showed that (1) chlorophyll content has a good linear relationship with spectral reflectance around 527, 703, 1 364 and 1 640 nm, and this result is helpful for us to select some important bands when monitoring P. massoniana by remote sensing image; (2) all of the nine kinds of spectral feature parameters including red edge position, mean reflectance of red edge, mean reflectance around red edge position, red edge slope, red edge area, absorption depth of red band, green peak height, red edge normalized difference vegetation index and red edge vegetation stress index, have exponential function relationship (r = 0.5-0.7) with the total chlorophyll; (3) the total chlorophyll content can be predicted by multivariate model by the nine spectral feature parameters, and partial least-squares regression model have higher prediction accuracy than the traditional multivariate linear model. The model's root mean square (RMS) is 0.008 8, and mean absolute percentage error is 0.761 7%. During the growth of vegetation, biochemical parameters such as chlorophyll have vital function, for example, it can indicate the health status or pathological feature. So, the models mentioned just above will help us understand the ecological process of P. massoniana forest and provide valuable reference for monitoring P. massoniana and pine wood nematode disease by remote sensing technique.