欢迎访问《麦类作物学报》编辑部网站

为了解无人机图像空间分辨率对倒伏小麦提取精度的影响，选取2019年6月9日冀南地区倒伏小麦农田为研究区，采用最大似然法、人工神经网络、支持向量机和随机森林四种分类方法，以倒伏小麦分类面积和空间一致性为指标，对不同空间分辨率下小麦倒伏的提取精度进行了比较。结果表明，最大似然法存在严重的错分现象，人工神经网络、随机森林和支持向量机的总体分类结果较好，其中人工神经网络对倒伏面积提取的结果最准确；随着像元尺寸的增大，倒伏小麦分类面积相对误差变化趋势缓慢，但像元尺寸大于40 cm时，分类结果与实际倒伏区域的空间一致性迅速降低。综合考虑无人机图像数据量、获取时间和倒伏小麦提取精度，本研究认为20～40 cm是提取冬小麦倒伏面积较为适宜的空间分辨率范围。

英文摘要:

In the past, the influence of UAV image spatial resolution on the extraction accuracy of lodging wheat was seldom considered. In this paper, the lodging wheat field on June 9, 2019 in Southern Hebei Province was selected as the research area. Four classification methods, i.e. maximum likelihood method, artificial neural network, support vector machine and random forest, were used to analyze the influence of different spatial resolution on the accuracy of wheat lodging extraction, taking the total area and spatial consistency as the indices. The results showed that there were serious misclassification phenomena in maximum likelihood method, and the overall classification results of artificial neural network, random forest and support vector machine were better, among which the results of artificial neural network extraction were most accurate. With the increase of pixel size, the relative error of lodging wheat classification area changed slowly, but when the pixel size was larger than 40 cm, the spatial consistency between the classification results and the actual lodging area decreased rapidly. Considering the amount of UAV image data, acquisition time and the extraction accuracy of lodging wheat, this study suggests that 20-40 cm is a suitable spatial resolution range for extracting lodging area of winter wheat.

查看全文查看/发表评论下载PDF阅读器

关闭