复杂环境下的目标检测识别算法及其硬件加速研究

This proposal takes the typical background of many different types of UAV systems attacking the sea, air and ground targets and integrated ISR as the background of demand. In order to solve the problem of target detection and recognition in complex environment, this proposal uses massive image data to study the data mining based on big data and deep learning Technology with high accuracy, robustness of the target detection and identification methods. Based on the research of target detection and recognition algorithm, the hardware acceleration method of target detection and recognition algorithm based on depth learning is further studied to improve the speed of target detection and recognition in complex environment and meet the practical application of computing real-time requirements. Finally, using the research results of target detection and recognition algorithm and deep learning hardware accelerator, the design and implementation of prototype verification system such as real-time target detection and recognition under complex background conditions and massive target search and trace reproduction are carried out. It is expected that the research results of this proposal can be applied to unmanned reconnaissance aircraft at high-altitude long-haul flights and to reconnaissance and reconnaissance unmanned aerial vehicles to perform reconnaissance and surveillance tasks to rapidly detect and identify weakly hidden targets in complex environments. For the small bee drone to perform integrated ISR tasks to achieve low cost, low power consumption, high real-time, high-precision target detection and recognition. In addition, the research results can be extended to other military applications such as perception and recognition.

本项目以未来多种不同类型无人机系统攻击海上、空中、地面目标及综合ISR等典型任务为需求背景，针对复杂环境下目标检测识别难题，利用海量图像数据，研究基于大数据和深度学习技术的具有高精度、强鲁棒性的目标检测识别方法。在目标检测识别算法研究的基础上，进一步研究基于深度学习的目标检测识别算法的硬件加速方法，提升复杂环境下目标检测与识别速度，满足实际应用对计算实时性的需求。最后，利用目标检测识别算法和深度学习硬件加速器的研究成果，开展复杂背景条件下的实时目标检测识别、海量目标以图搜图与轨迹重现等原型验证系统的设计与实现。预期本项目的研究成果可应用于高空长航时无人侦察机、察打一体无人机执行侦察监视任务，快速实现复杂环境下弱隐目标的检测与识别。面向小型蜂群无人机执行综合ISR任务，实现低成本、低功耗、高实时、高精度的目标检测与识别。此外，研究成果可推广应用于感知、识别等其他军事应用领域。

本课题以未来多种不同类型无人机系统攻击海上、空中、地面目标及综合ISR等典型任务为需求背景，针对复杂环境下目标检测识别难题，利用海量图像数据，研究基于大数据和深度学习技术的具有高精度、强鲁棒性的目标检测识别方法。在目标检测识别算法研究的基础上，进一步研究基于深度学习的目标检测识别算法的硬件加速方法，提升复杂环境下目标检测与识别速度，满足实际应用对计算实时性的需求。最后，利用目标检测识别算法和深度学习硬件加速器的研究成果，开展复杂背景条件下的实时目标检测识别原型验证系统的设计与实现。预期本课题的研究成果可应用于高空长航时无人侦察机、察打一体无人机执行侦察监视任务，快速实现复杂环境下弱隐目标的检测与识别。面向小型蜂群无人机执行综合ISR任务，实现低成本、低功耗、高实时、高精度的目标检测与识别。此外，研究成果可推广应用于感知、识别等其他军事应用领域。