太赫兹图像复原和自动目标识别算法研究

项目名称： 太赫兹图像复原和自动目标识别算法研究

项目编号： No.61501297

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 黄慧

作者单位： 上海师范大学

项目金额： 19万元

中文摘要： 太赫兹（THz）波具有较强的穿透力且对人体无害的特性，使得太赫兹成像技术作为一种新兴的安全检查技术而受到广泛关注。然而，由于太赫兹波物理特性的限制，太赫兹图像普遍分辨率较低并且带有大量噪声，因此，提高图像质量是目前太赫兹图像系统应用中亟待解决的问题。本项目拟利用数字图像处理技术和机器学习相结合的方法，揭示太赫兹图像退化的规律，从软件上实现太赫兹图像高分辨率复原，弥补太赫兹硬件成像系统的不足，进而建立准确识别藏匿物品的方法。本项目主要研究内容有：1）拟采用机器学习的方法，探索太赫兹成像系统物理特性造成图像降质的规律，从成像系统本身特性出发提高图像质量；2）拟在小波多尺度层面分别建立不同的图像复原算法，降低图像噪声的同时保存图像细节；3）建立太赫兹安检图像自动目标识别算法，探索对尺度、旋转鲁棒的图像特征提取方法，以实现藏匿物品的准确识别。本项目的实施将对推动太赫兹安检系统的广泛应用有重要意义。

中文关键词： 图像复原；退化图像复原；目标识别；机器学习；稀疏建模

英文摘要： Terahertz (THz) wave has strong penetration and harmless, which makes the Terahertz imaging technology as a new safety inspection technology attracted widespread attention. However, due to limitations of the physical properties of Terahertz waves, Terahertz image resolution is generally low and with a lot of noise. Therefore, improve the image quality is currently very important for promoting the Terahertz application. The project intends to take the advantage of digital image processing and machine learning method to explore Terahertz image degradation rules, so as to achieve Terahertz high-resolution image restoration , thereby to detect accurately concealed objects. The main contents of this project are: 1) use machine learning methods to explore the relation between the physical properties of Terahertz imaging system and image degradation, so as to improve image quality; 2) develop different image restoration algorithm in multi-scale wavelet levels, reducing image noise while preserving image details; 3) establish target recognition algorithms in THz image, looking for scale, rotation invariant image features, in order to achieve accurate classification result. The execulter of this project will be significant for promotion of Terahertz imaging system application.

英文关键词： image restoration;Degraded image restoration;target recognition;machine learning;Sparse modeling