随着无人机等低空平台在侦察领域的不断扩展以及对性能要求的不断提高，各应用场景对目标检测精度和速度也提出了越来越高的要求。传统的目标成像方法难以满足图像质量需求，人工识别目标的方法也无法应对战场环境的快速变化。结合深度学习和偏振高光谱成像技术的发展，通过模拟偏振高光谱低空目标检测平台，提出基于 Faster R-CNN的地面**目标检测方法。采用区域建议网络模块进行模型训练，而在目标检测阶段通过对特征图进行兴趣区域池化操作得到建议特征图，最后利用建议特征图完成目标类别判定。实验选取 3种典型的**车辆缩比模型，通过偏振高光谱相机在室内外模拟环境中获取目标在不同场景条件的图像数据，以及某型无人机在低空条件下的地面车辆目标数据进行实验验证。实验表明，该方法在有效完成地面目标的检测时，能够达到理想的检测精度和速度。

The use of unmanned aerial vehicles (UAV) for reconnaissance requires continuous improvements in performance, as each new type of scene observed can place more stringent requirements for the accuracy and speed of object detection. Traditional object-imaging methods have difficulty meeting such requirements, and artificial object recognition is not suited for rapidly changing battle field environments. Leveraging the concomitant development of deep learning and hyperspectral polarization imaging, ground object detection based on fast R-CNNs is proposed that’s imulates a polarized hyperspectral low-altitude object detection platform. We describe a region proposal network module for training models. In its object detection phase, this approach generates a feature map by pooling feature regions into the map, which is then used to complete the object categorization decision. Three typical scaled models of military vehicles were selected to test the technique experimentally. With a polarization hyperspectral camera, object images in different scene conditions were acquired in simulated indoor and outdoor environments, and ground vehicles were successfully observed by a low-altitude UAV. The experimental results show that the proposed method achieves the ideal detection accuracy and speed when the ground object is effectively detected.