高光谱信息的农林植被种类区分

为了能够更加快速、准确对粮食主产区的作物与树木进行种类区分，以黄淮海地区三种主要植被(玉米、小麦和杨树)为研究对象，获取该三种植被原始反射率光谱，并对原始光谱进行特征点提取、一阶微分变换、二阶微分变换以及植被指数计算四种方法的分析处理，提取三种植被各自的光谱特征点、特征波段、蓝黄红边微分值和、位置、振幅以及面积四个特征指标以及植被指数的数值区间。基于特征值在不同植被种类间数值重叠范围越小区分精度越高的原理，比较分析植被光谱在不同处理方法下的植被区分精度，并且最终选取重叠范围最小的特征指标作为区分不同植被的识别指标。结果显示：相较于原始光谱特征点提取、二阶微分变换以及植被指数计算，一阶微分变换对于玉米、小麦和杨树的识别分类具有较高的精度，其中黄边振幅、黄边面积以及黄边微分值和具有较高的识别精度，黄边振幅的识别精度达到97.5%，黄边面积以及黄边微分值识别精度达98.1%，用另外167组数据对该结果进行验证，显示黄边振幅的识别精度达96.4%，黄边面积以及黄边微分值和的识别精度达97.6%。该结果与用平均光谱曲线区分单种植被不同生长状态选取的特征值结果不同，这种方法能有效的保留个体光谱反射曲线的差异，从结果可见通过一阶微分变换提取黄边参数的方法能有效的用于树木和粮食作物共同种植区域的植被区分, 并且黄边面积以及黄边微分值和的识别精度最高。

An Approach to Distinguishing Between Species of Trees and Crops Based on Hyperspectral Information

In order to distinguish between the crops and trees in the main grain producing areas more quickly and accurately, maize, wheat and poplar which are the main vegetation planted in the Huang-Huai-Hai Plain are used as the research object. Obtain the original spectral reflectance and calculate the data by using the original spectral feature point extraction, first derivation, second derivation and vegetation index. Extract the range of feature points, characteristic bands obtained by analyzing the original spectral reflectance, position, amplitude, area and differential value sum in blue, red and yellow edges by first derivation, and vegetation index by empirical formulas. Compare the accuracy of four methods in distinguishing between the three vegetable types based on the principle that the smaller the overlapping range is, the higher the accuracy of parameter will be, and choose the most suitable characteristic index as the identify indicator which has the smallest overlap in different vegetation types. The results showed that among the four methods of manipulation spectral data, first derivation had the highest accuracy in identifying corn, wheat and poplar compared with the original spectral feature point extraction, second derivation and vegetation index. Among the indexes obtained by the first derivation, the amplitude, area and differential value sum in yellow edge region had the higher recognition accuracy. The recognition accuracy of the amplitude in yellow edge was up to 97.5%, and the area and differential value sum in yellow edge was up to 98.1%. The results were verified with 167 other sets of data, and the verification results showed that the recognition accuracy of the amplitude in yellow edge was up to 96.4%, and the area and differential value sum in yellow edge was up to 97.6%. The result was different from that result which was obtained by average reflectance curve of spectrum from the single plant in different growth state, and this method could effectively preserve the difference between individual spectral reflectance curves. Thus, it could be seen that extracting the yellow edge parameters through the first derivation was effectively used in distiguishing vegetation where the crops and trees were planted in the same place, and among all the parameters, area and differential value sum in yellow edge had the highest recognition accuracy.