基于非精确计算的高光谱图像目标实时探测方法研究

项目名称： 基于非精确计算的高光谱图像目标实时探测方法研究

项目编号： No.41301384

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 天文学、地球科学

项目作者： 吴远峰

作者单位： 中国科学院遥感与数字地球研究所

项目金额： 25万元

中文摘要： 基于高光谱图像进行目标实时探测在军事侦察、灾害监测、应急救援等领域有着重大的应用需求。目标探测的实时性对算法提出了很高的要求，不仅需要"准"而且需要"快"，一直是研究的难点。本研究创新性地将非精确计算思想引入高光谱图像目标实时探测中，从算法逻辑和计算复杂度层面力图找到非精确计算理论与高光谱图像目标实时探测的契合点，通过恒虚警率下高光谱目标探测算法中误差显著性及传递机理的研究，提出实时处理约束下的目标探测算法非精确计算优化，实现计算规模的几何级数降低。并通过地面仿真实验，基于不同高光谱数据和目标类型对非精确计算方法的适应性和鲁棒性进行测试验证。本研究为解决目标实时探测中的时间及精度约束提供一种有效的新思路。

中文关键词： 高光谱遥感；近似计算；目标探测；图像分类；实时处理

英文摘要： Real-time target detection in hyperspectral images is very important in the time critical applications such as military reconnaissance, disaster monitoring and emergency rescue. It's a difficult problem for the target detection algorithm, not only in accuracy but also in processing time. In this study, approximate computing is applied as an innovative way in real-time target detection algorithms and computational complexity analysis. The Error tolerance and error propagation mechanisms of the target detection algorithms are analyzed under the constant false-alarm-rate. And a real-time hyperspectral image target detection algorithm using approximate computing is proposed, which can simplify the computational complexity significantly. Simulation experiments are conducted to verify the algorithm adaptability and robustness, by testing different hyperspectral images and targets. This study provides an effectively way to solve the real-time target detection both in process time and accuracy.

英文关键词： Hyperspectral Remote Sensing；Approximate Computing；Target Detection；Image Classification；Real-time processing