基于机器学习方法的高光谱遥感图像目标检测研究

Research on target detection in hyperspectral remote sensing images based on machine learning methods

针对高光谱遥感图像目标检测的难题,提出基于机器学习的高光谱遥感图像目标检测方法。首先,通过动力演化算法找到满足偏度、峰度最大化的投影方向,将高维图像数据投影至低维子空间,从而提取图像的光谱信息。然后,通过线性判别法将提取的信息转换为直方图形式,再利用自动标记分水岭算法和KNN方法进行目标区域的初分割和分类,以去除非目标光谱像元。测试结果表明,对目标信息进行检测处理后,图像像元整体分类精度与平均分类精度数值区间为[0.97,0.98],信息熵数值仅有0.01,说明该方法具有较高精度,结果可信度较高。

Aiming at the problem of target detection in hyperspectral remote sensing images,a machine learning based method for target detection in hyperspectral remote sensing images is proposed.Firstly,the dynamic evolution algorithm is used to find the projection direction that maximizes skewness and kurtosis,and high-dimensional image data is projected onto a low dimensional subspace to extract spectral information from the image.Then,the extracted information is transformed into histogram form through linear discriminant analysis,and the target area is initially segmented and classified using automatic labeling watershed algorithm and KNN method to remove non target spectral pixels.The test results show that after detecting and processing the target information,the overall classification accuracy and average classification accuracy of image pixels have a numerical interval of[0.97,0.98],and the information entropy value is only 0.01,indicating that the method has high accuracy and high credibility of the results.