基于基元结构表示模型的高分辨率遥感图像典型中小型目标解译技术研究

With the improvement of the resolution of remote sensing images, the number and species of identifiable targets increase greatly.The requirements for the medium and small sized target interpretation technique also improve. Based on the high resolution remote sensing data,this project focus on the medium and small sized target such as aircraft, ship and so on,which is import in military and civilian significance.This project starts research on how to build effective target representation model with various basic elements to improve the performance of target interpretation in complex scene of remote sensing images，based on which we will research further on key algorithms about feature extraction, similarity measure, target detection and target recognition, including feature extraction algorithm based on structure mapping strategy, similarity measure algorithm based on reconstruction for templates with multi-scale segments, and target localization and recognition algorithm based on deformable template representation. By the study of these algorithms, the target are detected and recognized automatically. The project will enhance the application on auto target interpretation based on high resolution remote sensing data, and promote the industrial level of our remote sensing applications.

随着遥感图像分辨率的提高，可辨识目标的种类和数量大大增加，从海量遥感数据中快速获取目标信息的需要也更加迫切。本项目以高分辨率可见光图像为数据源，针对遥感图像中飞机、舰船等具有重要军用、民用意义的典型中小型目标，结合多种类型的基元结构构建分等级目标表示模型，对目标进行有效表示；在此基础上进一步研究有效的遥感图像中目标特征提取、相似度度量、目标检测和识别关键算法，包括基于极坐标空间信息映射的目标特征提取算法、基于多尺度基元目标重构的相似度度量算法以及基于形变建模的目标定位识别算法等，最终实现大数据量、复杂场景的遥感图像中小型目标的准确、快速、智能检测和识别。本项目的研究将推动高分辨率遥感图像在典型中小型目标解译方面的应用，促进我国航空航天遥感的产业化应用水平。

本项目研究的重点是面向大场景复杂背景高分辨率遥感图像，基于基元结构表示模型对飞机、 舰船等典型中小型目标的精细化解译技术进行研究 ，主要包括对大场景复杂背景下中小型目标快速检测和精确定位、遥感影像目标精细化参数信息提取等关键问题进行深入研究，在此基础上，结合大场景复杂背景下典型中小型人造目标的特点，搭建大场景目标检测定位-精细化参数提取框架，将其应用于实际遥感图像目标解译任务中，以提高具备实用能力的遥感目标精细化解译技术的性能。本项目将充分利用高分辨率遥感图像的优势，研究目标解译流程中涉及的关键算法，促进我国航空航天遥感的产业化应用水平。