基于无人机遥感的玉米株高提取方法

Extraction Method of Maize Height Based on UAV Remote Sensing

摘要:

为在玉米生长周期内，准确、快速地掌握玉米生长信息，通过无人机获取玉米生长阶段4期不同高清数码正射影像（Digital orthophoto map, DOM）及数字表面模型（Digital surface model,DSM），利用K-means算法、遗传神经网络算法和骨架算法分别对DOM中的玉米区域进行提取，生成掩膜，与DSM套和，获取玉米高度信息。与实地测量株高进行对比，3种方法的R2分别为0.853、0.877、0.923，RMSE分别为15.886、14.519、11.493cm，MAE分别为13.743、11.884、8.927cm。结果表明：结合DOM和DSM可以较好地提取生长阶段的玉米高度。与K-means算法、遗传神经网络算法相比，基于骨架算法提取玉米高度具有一定优势，且精度较高。采用DOM和DSM相结合的骨架算法提取植株骨架，为株高提取提供了一种新途径，可为无人机遥感监测作物株高状况提供参考。

Abstract:

In order to accurately and quickly grasp the growth information of maize in the growth cycle, different digital orthophoto maps（DOM）and digital surface model （DSM） in the four stages of the nutritional growth stage of maize were obtained by unmanned aerial vehicle（UAV）. K-means, genetic neural network and skeleton algorithm were used to extract the maize areas in the DOM, generate masks, and combined with DSM sets to obtain the height information of maize. Compared with the field measurement of plant height, the R2 of three methods were 0.853, 0.877 and 0.923, respectively, RMSE were 15.886cm, 14.519cm and 11.493cm, respectively, MAE were 13.743cm, 11.884cm and 8.927cm, respectively. The results showed that combining DOM and DSM can better extract the height value of maize in the nutritional growth stage. Compared with K-means and genetic neural network, the maize height extracted by the skeleton algorithm was highly consistent with the field measurement (R2 was 0.923, RMSE was 11.493cm, MAE was 8.927cm), and the extraction accuracy was high. Skeleton extraction combining DOM and DSM provided a way to extract plant height, which can be used as a reference for monitoring maize height by UAV remote sensing.