飞行器光学成像末制导快速数据处理与智能目标识别方法

In the future complex battlefield environment, the optical-imaging-terminal-guidance (visible and infrared) aircrafts, which adopt the cooperative combat pattern, should solve the following difficulties: the high-efficient compression, the long-distance transmission, and the high-quality reconstruction of the imaging data under interference; the fast restoration of the degraded images gathered in the rigorous imaging environment; and the real-time intelligent target recognition under the complex background conditions. The current methods corresponding to the problems above present the following flaws: the compression ratio and the robustness to the data loss of the image compression/reconstruction method are low; the result quality of the image restoration method is not high due to the low capability of prior knowledge mining; and the target recognition is either low intelligent or not fast enough. Accordingly, this project researches the image compression/reconstruction method based on compressed sensing, which raises the compression ratio, the robustness to the data loss, and the quality of the reconstruction; the fast image restoration method based on regularization, which raises the quality of the restoration of the degraded image; the intelligent target recognition method based on compressing deep learning, which raises the intelligence level of the target recognition; and the fast optimizing solution method of nonlinear/nonsmooth objective functions based on operator splitting, which raises the real-time quality of the data processing and the target recognition. The new research fruits provide solid theoretic foundations and crucial technical preparations for the optical-imaging-terminal-guidance of aircrafts.

未来复杂战场环境下，协同作战模式下的光学成像末制导飞行器需克服以下难题：干扰条件下的成像数据高效压缩、远距离传输与高质量重建问题；严酷成像条件下的快速退化图像复原问题；复杂背景条件下的实时智能目标识别问题。应对以上问题的现有方法存在以下不足：图像压缩/重建压缩比低、抗数据丢失能力差；图像复原挖掘图像先验信息不足，复原质量有待提高；目标识别或智能化程度不高，或难以实时实现。针对此，研究基于压缩感知的图像压缩/重建方法，以提高飞行器图像压缩、传输和重建环节的压缩比、抗数据丢失能力和重建质量；研究基于正则化的快速图像复原方法，以提高退化成像数据的复原品质；研究基于压缩深度学习的智能目标识别方法，以提高飞行器智能目标检测识别的能力；研究基于并行算子分裂的非线性非光滑目标函数快速优化求解方法，以提高飞行器图像数据处理和目标识别的实时性。研究成果可为飞行器光学成像末制导提供坚实的理论基础和技术储备。

未来复杂战场环境下，协同作战模式下的光学成像末制导飞行器需克服以下难题：干扰条件下的成像数据高效压缩与高质量重建问题；严酷成像条件下的快速退化图像复原问题；复杂背景条件下的实时智能目标识别问题。以上述问题为研究背景，该项目对光学成像制导中的快速图像数据处理和智能目标识别问题进行了深入研究。具体地，研究了基于压缩感知的图像压缩和重建，构造了约束模型适应能力强的快速临近分裂方法，相关方法适用于压缩感知图像的快速复原重建；研究了退化图像复原问题，提出了基于正则化和学习特征的图像去雾方法和基于特征稀疏表达先验和距离加权表达先验的图像去模糊和匹配一体化方法；研究了基于深度学习的智能目标识别问题，提出了复杂背景下的小目标检测网络和半监督学习网络，并在此基础上研究了网络的压缩方法；研究了非线性、非光滑目标函数的优化问题，提出了基于并行算子分裂理论，利于并行计算的非线性、非光滑目标函数优化方法。研究成果对于提升复杂背景下图像数据处理和目标识别的实时性具有重要意义，可为飞行器光学成像末制导提供坚实的理论基础和技术储备。