宁夏盐池荒漠植物光谱特征分析及光谱识别

宁夏盐池县荒漠草地属于中温带干旱气候，由于过度利用出现不同程度的退化，退化指示种比重增大，造成不同荒漠草地群落组成差异也很大，如何区别不同荒漠草地植物，并据此对退化指示种进行动态监测是了解荒漠草地退化程度的关键。目前随机森林（RF）、支持向量机（SVM）与K-邻近（KNN）分类模型被广泛应用于森林植物和农作物的遥感分类，并取得了较好的分类识别效果，但针对草地尤其是荒漠草地植物的分类识别研究较少。因此使用ASD地物光谱仪于7月在宁夏盐池二步坑、冯记沟、高沙窝、麻黄山不同荒漠草地采集了32种植物作样本获得442条光谱进行光谱特征分析。筛选出7个植被指数：归一化植被指数705（NDVI₇₀₅）、绿通道植被指数（GNDVI）、光化学植被指数（PRI）、土壤调节植被指数（OSAVI）、可视化气压阻抗指数（VARI）、植被衰减指数（PSRI）和归一化水指数（NDWI）作为随机森林模型（RF）、支持向量机（SVM）模型、K-邻近（KNN）模型的原始变量，对32种荒漠草地植物进行分类识别，并通过分类模型精度的比较筛选较优模型。结果表明：(1）不同植物光谱反射率均符合绿色植物特征，但各植物原始光谱不同波段之间存在明显差异，植物原始光谱水分吸收波段差异明显，且有红边蓝移现象；(2）RF，SVM和KNN三个分类模型对32种植物的分类精度分别达到了0.98，0.94和0.98，识别效果较好，但3种分类模型均对白莲蒿与北芸香、虫实与甘草发生了误判；(3）随机森林模型重要性指标中NDWI与PRI为区分荒漠草地植物的关键指标，说明荒漠植物冠层水分与类胡萝卜素含量是影响荒漠草地植物光谱分类的重要因素。试验利用随机森林模型（RF）、支持向量机（SVM）与K-邻近（KNN）分类方法，建立了主要植物的分类模型。

Spectral Characteristic Analysis and Spectral Identification of Desert Plants in Yanchi,Ningxia

The desert grassland in Yanchi County of Ningxia belongs to the mid-temperate arid climate. Due to over-utilization, different degrees of degradation has occurred, and the proportion of degradation indicator species has increased, resulting in large differences in the composition of different desert grassland communities. How to distinguish between different desert grassland plants and determine the Dynamic degradation monitoring of indicator species is the key to understanding the degree of desert grassland degradation. At present, random forest (RF), support vector machine (SVM) and K-neighbor (KNN) classification models are widely used in the remote sensing classification of forest plants and crops. The classification and recognition effect is good, but few studies on the classification and recognition of grassland, especially the desert grassland. Therefore, in July, the ASD ground feature spectrometer was used in Ningxia Yanchi Erbukeng, Fengjigou, Gaoshawo and Mahuangshan. In the desert grassland, a total of 442 spectral data of 32 species of plants were collected for spectral feature analysis, and 7 vegetation indexes were selected: normalized vegetation index 705 (NDVI705), green channel vegetation index (GNDVI), photochemical vegetation index (PRI), soil Adjusted vegetation index (OSAVI), visual pressure resistance index (VARI), vegetation attenuation index (PSRI) and normalized water index (NDWI) as random forest model (RF), support vector machine (SVM) model, K-neighbor (KNN) the original variables of the model, classify and identify 32 species of desert grassland plants, and screen the best model by comparing the accuracy of the classification models. The research results show that: (1) The spectral reflectance of different plants is in line with the characteristics of green plants but there are obvious differences between the different bands of the original spectrum of each plant, and the difference in the water absorption bands of the original spectrum of plants is obvious, and there is a red edge blue shift phenomenon; (2) RF The classification accuracy of the three classification models, SVM and KNN for 32 species of plants reached 0.98, 0.94 and 0.98, respectively, and the recognition effect was good. However, the three classification models all made mistakes in the classification of Artemisia spp. Judgment; (3) NDWI and PRI are the key indicators to distinguish desert grassland plants in the importance of random forest model indicators, indicating that desert plant canopy water and carotenoid content are important factors affecting the spectral classification of desert grassland plants. The experiment uses a random forest model (RF), support vector machine (SVM) and K-neighbor (KNN) classification methods to establish a classification model for main plants, laying the foundation for remote sensing monitoring of desert grasslands.