基于UNet模型的智能海洋遥感分类框架研究

Research on Intelligent Marine Remote Sensing ClassificationFramework Based on UNet Model

借助遥感影像和人工智能方法自动提取空间地物是遥感领域的热点方向,本文基于PyQGIS和改进的UNet模型,设计开发了智能海洋遥感分类程序。该程序可以实现面向大尺度遥感影像的用户自定义智能分类、裁剪、拼接、分类评价、影像可视化和地图操作(包括影像放大、缩小、属性查询等)。并以三类典型海洋地物为案例,基于开发的程序测试,实现了总体精度92%、Kappa系数0.87、平均交并比82%的改善结果。文中提出的智能海洋遥感分类框架可为研究人员和工程人员提供参考,为沿海地区规划、监管和综合治理提供决策支持。

Automatic extraction of spatial features with the support of remote sensing images and artificial intelligence methods is a hot direction in the field of remote sensing.In this paper,an intelligent marine remote sensing classification program was developed based on PyQGIS and an improved UNet model.The program can realize user-defined intelligent classification,clipping,mosaicing,classification evaluation,image visualization and map operations(including image zooming in and out,attribute query,etc.)for large scale remote sensing images.The study took three types of typical marine features as case studies,and based on the developed program testing,it achieved an overall accuracy of 92%,a Kappa coefficient of 0.87,and a mean intersection over union of 82%as improvement results.The proposed application framework can provide a quick practical reference for researchers and engineers,and provide decision support for coastal area planning,regulation and integrated management.