基于面向對象與深度學習方法的遙感影像自動提取技術研究

打開文本圖片集

摘要：文章在對面向對象多尺度分割技術和深度學習技術分別進行理論、方法闡述后，開展目標區(qū)建設用地和非建設用地自動提取實例研究。通過建立少量地類樣本庫完成遙感影像自動分類提取，并對提取結果進行分析，得出目標區(qū)總體分類精度達到94.40%，建設用地的制圖精度和用戶精度能夠滿足實際生產(chǎn)需求。

關鍵詞：遙感影像；自動提取；面向對象；深度學習

doi：10.3969/J.ISSN.1672-7274.2023.07.009

中圖分類號：P 237，TP 3             文獻標志碼：A               文章編碼：1672-7274（2023）07-00-03

Research on Remote Sensing Image Automatic Extraction Technology Based on Object Oriented and Deep Learning Methods

DOU Yajuan

（Zhongse Blueprint Technology Co.， Ltd.， Beijing 101312， China）

Abstract： This article conducts a case study on automatic extraction of construction and non construction land in the target area. By establishing a small number of land class sample libraries to complete automatic classification and extraction of remote sensing images， and analyzing the extraction results， it was found that the overall classification accuracy of the target area reached 94.40%， and the mapping accuracy and user accuracy of construction land can meet actual production needs.

Key words： remote sensing images; automatic extraction; object-oriented; deep learning

目前，遙感圖像解譯存在兩大難點：一是不同地物難以分割開，二是地物分類不準確。（剩余4348字）