【作者】 宋雪岩；

【导师】 石光明；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2018， 硕士

【摘要】 光谱成像技术在使用传感器获取目标图像的同时,也能获取反映目标特点的连续、光滑的光谱曲线,是一种既能成像又能获取目标光谱曲线的“谱像合一”的技术。在诸多光谱成像技术中,使用LED主动光照和灰度相机构建的多光谱成像系统,由于其结构简单、成本低等优点,在医疗诊断、物质检测等领域得到了较好的应用。目前的基于LED主动光照的多光谱成像系统大都使用多种不同谱段的窄带LED进行多次光照,并进行相应的多次曝光成像。其中每种LED的光谱分布是一个高斯型的函数,这使得成像光谱的波段数受到LED数量的限制,造成这种方法在光谱维的信息采集能力较差。近年来出现的压缩感知理论,由于其较强的信息感知能力为成像技术带来了新的机遇,基于压缩感知理论的计算成像技术得到了快速发展,例如典型单像素相机、编码孔径快照式光谱成像仪(Coded Aperture Snapshot Spectral Imagers,CASSI)系统等。本文将压缩感知理论应用于LED主动光照的多光谱成像,提出了基于随机宽谱带编码光照的多光谱成像方法。该方法将成像过程分成两个过程,首先对成像场景进行编码混叠采样,然后再利用图像的空间、谱间的稀疏性,通过优化算法重构出图像。在编码混叠采样时,该方法不再使用单个类型LED进行光照采样,而是将几种LED随机编码组合,形成宽带光源,通过较少的曝光次数完成光谱维的混叠采样,光谱仿真实验结果表明,本文方法具有更高的PSNR和光谱精度,相比于窄带LED光照成像,能够采集到更多光谱维信息,重构光谱准确度更高,观测次数更少。为进一步验证本文方法的可行性,我们还构建了硬件系统平台,实验结果验证了该方法的有效性。更多还原

【Abstract】 The spectral imaging technology can obtain the image of the target scene,and can also obtain continuous,smooth spectral curve that reflects the characteristics of the target scene.Among many multispectral imaging techniques,the systems using LED illumination and a monochrome camera have been widely applied in medical diagnostics and object detection,due to their simple structure and low cost.Current multispectral imaging systems based on LED illumination mostly use multiple spectrally distinct LEDs and need multiple exposures.The spectral distribution of each LED is a Gaussian-like function,which makes the number of bands of the spectral image limited by the number of LEDs and results in poor information collection capability in spectral dimensions.Recently,the compressive sensing theory brought new opportunities for spectral imaging technology.The computational imaging technology based on the compressive sensing theory has been developed rapidly,such as single pixel camera,coded aperture snapshot spectral imager,etc.In this paper,we proposed a multispectral imaging method using random broadband coded-illumination.The proposed method divides the spectral imaging process into two stages:the observation process and the data restoration process.First,the target scene is coded with aliased sampling.Then the high-resolution spectral image is obtained with an optimized algorithm.In the observation process,no longer a single type of LED but several types of LED are randomly coded and combined for light sampling.Simulation result shows that the proposed framework has higher PSNR and spectral accuracy with fewer exposures than traditional methods.To further verify the feasibility of our method,this paper also sets up a hardware system platform,and the experimental results verified the effectiveness of the method.更多还原