面向高分遥感影像的快速解译系统设计与实现

DESIGN AND IMPLEMENTATION OF FAST INTERPRETATION SYSTEM FOR HIGH-RESOLUTION REMOTE SENSING IMAGES

高分遥感影像可用于快速监察和跟踪突发事件,然而海量影像数据对系统的处理效力及解译的实时性提出更高要求。基于B/S架构设计一套高分影像快速解译系统,采用多并发、多节点、多线程、多任务的计算模式,实现大规模高分影像的快速入库及预处理;通过模块化的方式构建部署深度学习训练平台,以智能学习为主、人工校验为辅,交互可视化地实现高分影像的高效解译;结合影像金字塔、分块瓦片及改进的Nginx服务器技术,实现对目标影像的快速加载及可视。测试结果表明,对单幅12 GB的高分影像进行解译,从入库、信息录入、目标检测算法的运行到最终完成解译,解译时间得到显著缩短,且机器自动识别的正确率达95%。

High-resolution remote sensing imaging can be used to quickly monitor and track emergencies.However,massive image data put forward higher requirements for the processing effectiveness and real-time interpretation of the system.Based on B/S architecture,a fast interpretation system for high-resolution images is designed.It adopted a multi-concurrent,multi-node,multi-thread and multi-task computing mode to realize the fast loading and pre-processing of large-scale high-resolution images.The deep learning training platform was constructed and deployed in a modular way,with intelligent learning as the main and manual verification as the auxiliary,to achieve efficient interpretation of high-resolution images interactively and visually.Combined with image pyramid,tiled mapping and improved Nginx server technology,it could quickly load and visualize the target images.The test results show that the interpretation time of a single 12 GB high-resolution image is significantly shortened from the loading,information storing,target detection algorithm running to the final interpretation,and the accuracy of machine automatic recognition reaches 95%.