Using spectral features, morphological features and Hough transformation from UAV images, a method of mapping corn seedling is proposed. First, spectral features and morphological features are extracted from UAV images and then are separately classified using an improved one-class random forest for mapping of corn seedling. Second, Hough transform is used to extract the corn seedling rows from the classification results with morphological features. Third, the corn seedling classification results with spectral features and the corn seedling rows from Hough transform are combined to obtain the final seedling mapping result. The proposed mapping method is evaluated in two study areas. The results demonstrated that the proposed method, effectively combines morphological features and Hough transform in mapping of corn seedling, thus obtaining better results compared with the existing methods.

提出一种利用光谱特征、形态学特征和霍夫变换的无人机图像玉米幼苗分布信息提取方法。首先, 从无人机图像中提取光谱特征和形态学特征, 利用改进的单类随机森林算法, 分别得到基于光谱特征和基于形态学特征的玉米幼苗图像分类结果。然后, 利用霍夫变换方法, 从基于形态学特征的玉米幼苗图像分类结果中提取玉米幼苗行线。最后, 利用得到的玉米幼苗行线, 优化基于光谱特征的玉米幼苗图像分类结果, 得到最终的幼苗分布信息提取结果。两个研究区提取结果的对比表明, 所提方法有效地结合了形态学特征与霍夫变换的特点, 与现有方法相比, 可以得到更好的玉米幼苗分布信息提取结果。