基于随机森林算法的航空高光谱数据分类方法研究

高光谱数据具有光谱范围广,光谱分辨率高等优势,可以用于不同地物的分类识别,为近年来遥感领域的研究热点。采用随机森林算法对机载高光谱数据进行了地物分类识别研究,首先选取不同种类的地物样本,并对每类样本打上类别标签,每个像素包含的波段数即为样本的特征数,送入随机森林分类器进行训练;然后将训练好的分类器对待分类的高光谱影像数据进行分类,待分类的数据初始化为统一的类别标签;并根据袋外数据自变量的扰动对分类精度的影响,计算不同波段特征对分类效果的重要性系数。实验采用C++语言结合Intel Open CV计算机视觉库,编写了高光谱影像分类识别程序,对机载AISA高光谱传感器获取的甘肃省张掖市农村与城市影像数据进行分类,结果表明本文算法具有较高分类精度和可靠性。

Study on the Classification Method of Aerial Hyperspectral Image Based on the Random Forest Algorithm

The hyperspectral image data has the advantage of wide spectral range and high spectral resolution, is widely applied in the surface feature classification and recognition. The hyperspectral remote sensing is the research focus in recent years. In this paper, the random forest algorithm is employed in the classification and recognition research of hyperspectral image. Firstly, the different surface features are selected, and tagged with class labels. The bands of each pixel are the features of the sample. The Random forest classifier is trained by selected samples, and then applied in the hyperspectral image to be classified. The hyperspectral image to be classified is tagged with the same label. Through the classification precision effected by the variable of the out-of-bag data changing, the importance coefficient of each band to the classification result is acquired. We adopt the C++ language with Intel OpenCV library to write the hyperspectral image classification program. The hyperspectral image of village and city in Zhang ye of Gansu province collected by AISA hyperspectral sensor is used in the classification experiment. The result indicates that the algorithm adopted in this paper with the higher classification precision and the reliability.