大数据光谱视频的理解，压缩和交互研究

Hyperspectral or multispectral video applications have influential impacts on every aspects of our life. For instance, it could be used for environmental monitoring, food safety detection, exploration sensing, etc. Even though there are few discussions on hyperspectral image compression (i.e., most of them are based on JPEG2000), it does not have systematic exploration on hyperspectral video yet. On the other hand, recent advances of hyperspectral video are mostly focusing on the real-time, high accurate signal acquisition. It results in the exponential increase of the storage, thus it is demanding to have the high-efficiency hyperspectral video compression as well as the friendly design for exchange system. This work will emphasize on the processing of hyperspectral video including content understanding, compression and exchange, particularly on the content understanding and compression. We propose to develop the high-efficiency hyperspectral video compression framework based on the deep content understanding, including its feature selection and extraction (such as wavelet coefficients), and theoretical distribution (especially inter-spectral correlations). With high-efficiency compressed content, we also propose to explore the media file system with flexible extension capability for further exchange purpose (such as classification, search and retrieval). In addition to the theoretical advances of hyperspectral video processing exploration, together with the world-class signal acquisition, it also promises the brilliant future due to the massive hyperspectral video applications.

光谱视频应用对生产、生活都有着积极重要作用，如环境检测，食品安全保障，物质勘探等。虽然先前有零星的光谱图像压缩工作（主要使用JPEG2000），但还未有系统的针对光谱视频的研究工作。此外，目前国内外主要着眼在光谱视频的高精度实时采集。随着采集的技术进步，光谱视频内容的数据量指数增长，如何高性能压缩和支持后期广泛交互比以往更加急迫。因此本项目主要涉及光谱视频的理解，压缩和交互，重点研究光谱视频理解和压缩。立足于对光谱视频内容的深入理解，包括其特征提取（例如小波系数），以及理论分布（例如谱域相关分布），设计实现针对光谱视频的高性能压缩方案。并且在高效压缩的基础上，通过建立易于扩展的媒体文件系统，融合光谱视频内容及其特征描述，提供开放友好的平台来支持光谱视频的交互（例如分类，检索和比对，等）。集合国内外先进的光谱采集系统，本项目不仅仅具有理论研究意义，也可以深化推广光谱视频在生产生活上的广泛应用。

光谱视频应用对生产、生活都有着积极重要作用，如环境检测，食品安全保障，物质勘探等。相比较传统RGB视频，光谱视频的数据量巨大，需要高效的压缩来赋能潜在的应用。因此，本项目从研制手持式光谱相机入手，实现便利的光谱视频数据采集和数据库构建；提出多通道混合编码方案对光谱和RGB视频进行同步压缩，并应用光谱平均特征峰优化重建光谱视频性能；受生物视觉启发，开拓基于特征域的端到端智能视频编码技术，通过大数据驱动学习非线性变换来高效和紧致的表征视频数据；综合光谱数据的大数据量、敏感数据的安全性和网络带宽的不确定性，提出基于互动视频的交互云系统原型，实现光谱视频应用的低时延高质量远程交互，为光谱视频的大范围应用提供友好平台。本项目的成果可以积极促进光谱视频在工业生产（如有害气体检测）和生活（食品安全保障）的实际应用。