相同地物多光谱遥感影像分类误差分析及其性能对比分析

Analysis of Classification Error of Multispectral Remote Sensing Images of the Same Terrain and Its Comparative Analysis of Performance

摘要: 为了探究易混地物波谱的相近程度对分类的影响，利用Landsat-8影像，选取青海湖和洞庭湖作为研究区域。采用ROI剪裁，选取同一地区(青海湖区)和跨区(青海湖区和洞庭湖区)遥感影像中的水，阴影，云，植被，陆地和建筑6种目标地物的三组样本集合，以6种目标地物样本集合中的平均反射率作为各类目标地物的参照光谱曲线，对比欧式距离，曼哈顿距离，余弦相似度三种不同衡量方式的分类结果精度，实验结果表明：1) 同一区域三种分类方式的分类精度中，曼哈顿距离最好；2) 不同区域波谱相近的地物中水的整体分类精度最高，阴影和陆地较差。

Abstract: In order to explore the influence of the similarity of the spectrum of easily mixable terrain on classification, Qinghai Lake and Dongting Lake were selected as the study areas by using Landsat-8 images. Using ROI clipping, three sets of samples of water, shadow, cloud, vegetation, land and buildings in the same area (Qinghai Lake area) and cross-region (Qinghai Lake area and Dongting Lake area) remote sensing images were selected, and the average reflectivity of the six target fea-ture sample sets was used as the reference spectral curve of various target features, and the accu-racy of the classification results of three different measurement methods of Euclidean distance, Manhattan distance and cosine similarity was compared, and the experimental results showed that: 1) the classification accuracy of the three classification methods in the same area Manhattan is the best distance; 2) the overall classification accuracy of water in features with similar spectrum in different regions is the highest, and the shadow and land are poor.