A modulated sparse random matrix for high-resolution and high-speed 3D compressive imaging through a multimode fiber

基于稀疏随机矩阵光场调制的高分辨高速多模光纤三维压缩感知成像

Multimode fiber(MMF)possesses intrinsic,ultrathin,noninvasive geometric characteristics and high mode density optical characteristics,having great potential in endoscopic imaging[1,2].A typical MMF endoscopic imaging system involves a camera placed behind the fiber distal end to calibrate the transmission matrix(TM)of the fiber,and when imaging into narrow channels,such as bronchus,or performing insertion into brain tissues,the camera should be replaced by a single-pixel detector placed outside in a reflection mode to record the signal.A traditional MMF imaging system adopts the point scan mode,whose resolution is limited by the numerical aperture(NA)of the fiber,and the imaging speed is restricted by the refresh rate of the modulator and the Nyquist sampling law.The need for high resolution,high speed,and three-dimensional(3D)imaging for observation of biological processes,such as neural activity,blood flow detection,and cellular dynamics,calls for original imaging approaches.

多模光纤内窥镜能在如发丝般纤细的探头上容纳数千个模式,在低侵入和高分辨在体成像方面具有广阔的应用前景.传统的点扫描成像方式速度受限,而基于散斑采样的压缩感知方法只能对二维稀疏样品成像.本文提出了一种在多模光纤中生成具有高正交性的分区域稀疏随机压缩感知采样矩阵的方法,在提升8倍采样速度的同时可对复杂样品进行高对比度成像.分区域采样方法的引入赋予了在各目标区域内对成像参数进行灵活调节的能力,既可以保证图像强度的均匀分布,又能提高图像的局部分辨率.结合并行计算,整个重构速度比传统方法提高了约10倍.基于三维稀疏随机采样和体重构算法,首次实现了对分布在100μm×100μm×200μm体积中的荧光颗粒的高保真度压缩感知成像,可以有效去除离焦面噪声的干扰.此方法为多模光纤内窥镜以微创或无创的方式对狭窄空间内动态过程的高速在体三维成像奠定了基础.