关键词: 毒杂草, 高光谱, 光谱特征, 包络线去除, 求导, 马氏距离法, 退化草地, 三江源区

Abstract:

The poisonous weed type degraded grassland is a type of degraded grassland in the "Three-River Headwaters" Region, which has the characteristics of community reverse succession. Its manifestation is that the grassland community with fine herbage as the dominant species evolved into the grassland community with weeds and poisonous plants as the dominant species. Therefore, the key to control this kind of grassland degradation is to monitor the spread of poisonous weeds and the change of grassland community structure. Hyperspectral remote sensing Technology can be used to identify poisonous weeds and fine herbage in a region, and spectual feature analysis is the basis of remote sensing recognition of terrain features. In this paper, the hyperspectral data of eight typical poisonous weeds and four fine herbages in the degraded grassland of toxic weed type in the Three-River Headwaters Region were selected as the research samples. The hyperspectral data in this study are all from field sampling using AvaField-2 portable hyperspectral surface object spectrometer. After Savitzky-Golay convolution smoothing, envelope removal, derivative transformation, and spectral parameterization, the spectral characteristics of poisonous weeds and fine forage species were analyzed, and the characteristic recognition bands were extracted by Mahalanobis distance method. The results showed that: (1) After data preprocessing, the spectral reflectance curves of eight poisonous weeds and four fine forages are similar, but the spectral reflectance differences can be compared through some characteristic bands and parameters; (2) NIR peak refers to the maximum reflectivity of each vegetation in the wavelength range 780~1000 nm. The NIR peak of eight typical poisonous weeds and four fine forages were significantly different. The NIR peak of Potentilla anserine reached 60.07%, while that of Poa pratensis was only 17.53%; (3) After envelope removal, the spectral difference between absorption valley and reflection peak in vegetation spectral curve is more obvious. The maximum values of Caltha and Ligulariavirgaurea with similar NIR peaks were 0.3671 and 0.2157, respectively, in the green band; (4) The sensitive recognition bands of toxic weeds and fine forages based on Mahalanobis distance are mainly distributed in the wavelength range of 680~750 nm and 900~1000 nm. This study can provide an important scientific basis for the use of hyperspectral remote sensing in the identification of large-area poisonous weeds and the monitoring of vegetation community growth. It is of great significance to the monitoring and control of poisonous weeds and the sustainable development of animal husbandry in the Three-River Headwaters Region.

Key words: poisonous weeds, hyperspectral, spectral characteristics, Mahalanobis distance, envelope removal, derivation, degraded grassland, “Three-River Headwaters” Region