针对高速飞行器光学（可见光、红外）成像末制导中目标识别面临的成像严酷、实时性强等难点，开展图像校正、特征提取、目标识别的理论、方法与技术研究，提出基于内容的气动光学模糊图像校正、点扩展函数非中心对称条件下的模糊不变量的感知与构造、时-空-谱上下文指导的目标识别等方法。在此基础上，提出图像校正与目标识别一体化的处理框架，通过目标识别对图像校正的反馈，将校正计算快速“聚焦”到对目标识别有价值的区域，降低图像校正与目标识别的整体计算量，在方法和算法的层面提高目标识别的实时性。设计异构多处理器实时系统体系结构，研制验证处理平台，在实现技术层面提高目标识别的实时性。突破高速飞行条件下光学成像目标末制导面临的“看得清、看得准、看得快”的关键技术难点，为高速飞行器光学成像末制导提供理论和技术基础支撑。

Aiming at the key problems in optical imaging terminal guidance of high-speed aircraft, such as poor imaging conditions and strict real-time requirements, this project focuses on the studies upon the theory, method and technology of image restoration, feature extraction, object recognition, and proposes content-based aero-optical blurry image restoration method, blur invariants extraction method in the case of non-symmetric point spread funtions, and spatial-temporal-spectral context-based object recognition method. On this basis, we propose an integration processing framework by combining image restoration and object recognition together in an iterative manner. Through the feedback from object recognition, the rough restoration is restrained to the interesting regions related to objects, thus the objects recognition can be more accurate and efficient. Moreover, we plan to study the heterogeneous multi-processor architecture and design the prototype platform for real-time implementation. The project will solve the key problems of “seeing clearly, accurately and fast” in high speed condition, and provide new methods and technologies for optical imaging terminal guidance of high-speed aircraft.