摘要
为改善传统土壤监测过程效率低、成本高及扩展性差等缺点,在分析OBIA技术和深度学习图像分割算法基础上,设计了一种端对端的多分辨率遥感土壤监测算法。该算法使用OBIA作为训练标签,通过U-Net提供像素级别属于特定土壤侵蚀类别的概率信息,并结合阈值产生最终分割结果图。仿真结果表明,该方法准确率和召回率结果明显优于传统OBIA方法。
In order to improve the low efficiency,high cost and poor scalability of traditional soil monitoring process,an end-to-end multi-resolution remote sensing soil monitoring algorithm was designed based on OBIA technology and deep learning image segmentation algorithm.The algorithm used OBIA as training tag,provided probability information of pixel level belonging to a specific soil erosion category through U-Net,and generated final segmentation result map combined with threshold.The simulation results showed that the accuracy and recall of the proposed method were significantly better than the traditional OBIA method.
作者
赵明海
黎飞明
王栋
牛丽娟
ZHAO Minghai;LI Feiming;WANG Dong;NIU Lijuan(Shaanxi Railway Institute, Weinan, Shaanxi 714000, China)
出处
《林业调查规划》
2022年第3期27-31,共5页
Forest Inventory and Planning
基金
陕西铁路工程职业技术学院科研基金(KY2020-38).
