基于统一计算设备架构的干涉成像光谱快速反演技术研究

Fast inversion techniques of inteferogram imaging spectrum based on CUDA

摘要: 在基于干涉光谱成像的气体成分实时遥测应用中，为了对推扫获取的原始干涉数据进行快速、有效的反演处理，提出一种结合计算统一设备架构（CUDA）的并行时空混合调制型长波红外干涉光谱反演算法。通过分析自主研制的时空混合调制型干涉光谱仪的数据获取模式，结合CUDA平台实现了并行反演算法。实验结果表明，基于CUDA平台的并行计算技术比仅使用CPU进行计算在效率上提升了5至20倍，为后期进一步做光谱识别打下了基础。

Abstract: In the realtime remote sensing of gas based on inteferogram imaging, fast inversion processing of original data is required.But the data acquired in this way is a large quantity,and the inverse algorithm is also complicated while using CPU for processing is time-consuming.A parallel inversion algorithm using compute unified device architecture (CUDA) was realized by analyzing the patten of data obtained by time-space mix-modulation long-wave infrared interference spectrometer. The results show that the parallel computing technique using CUDA can enhance the efficiency of spectrum inversion by 5 to 20 times,which provides foundations for further application like spectral recognition.