本项目研究结构信号压缩感知的联合观测方法及压缩域的目标直接分类方法。 它针对压缩感知研究中非自适应观测阵与采样信号特性失配，观测域直接分析/分类目标困难的现状，采用多观测灵活重构和压缩域信号结合投影矢量特征的表示方法，直接描述和识别目标。在低分辨率成像及其图像分析，特别是移动探测中发挥重要作用。预期创新： 1. 实现结构信号的Chirp矩阵观测和基于此线性调频模型的组原子计算重构。要求能对可变尺度结构的压缩感知信号位置和类型进行确定。 2. 基于显著结构表示和可靠的先验信息嵌入，实现结构信号多观测和分层合作编码。要求在常规压缩感知外，附加正交补充观测、二次和频谱分层多观测结果，可灵活计算重构，可提取感知压缩域特征。 3. 实现压缩域目标信号的准确检测。要求给出观测阵投影矢量在目标识别中的特性，用感知学习方法完成压缩域信号分析系统，并从观测数据中直接分析和定性目标类型。

The project is focused on joint measurement of structural signals and their direct pattern recognition in the compressed domain. To deal with mismatches between signals and the non-adaptive observation matrix and to classify compressed targets, multiple measurements and flexible reconstruction are carried out with combined feature vectors. Its applications include: low resolution imaging, direct analysis of compressed signals, especially those in motion detection. We expect that: 1. Realize a Chirp-matrix-based sensing and the corresponding atom-group reconstruction on scalable targets, aiming to determine structure locations and signal types. 2. With salient structure descriptions and the reliable a priori knowledge, multiple measurements with layered co-encoding are implemented. The assisting measurements include: a compensative orthogonal measurement, secondary measurements, and the spatial shifted measurement with its frequency range separated into different intervals. 3. Accurate target detection in compressed domain will be fully completed by revealing cluster performance of project vectors in an observation matrix, and by analyzing signals with sensing learning. Targets are classified in measurement domain instead of the traditional data domain.