相比传统图像目标检测算法,基于大数据和深度学习的检测算法无须人工设计特征,且检测性能更稳健。在防空应用背景下,自建了空中目标静态和视频图像数据集用于训练和测试,改进了基于深度学习的目标检测框架Faster R-CNN,将其专用于空中目标检测。结合空中目标检测任务的特点和需求,提出膨胀积累、区域放大、局部标注、自适应阈值、时空上下文等改进策略,弥补了Faster R-CNN对弱小目标和被遮挡目标不敏感的缺陷,提高了检测速度和精度。实验表明,改进后的Faster R-CNN在应对弱小目标、多目标、杂乱背景、光照变化、模糊、大面积遮挡等检测难度较大的情况时,均能获得很好的效果。数据集上测试结果的平局准确率均值较改进之前提高了16.7%,检测速度提高了3倍。

Compared with the traditional detectors, the detectors based on large data and deep learning do not require manually designed features and are more robust. Under the background of air defense, we build the images and videos dataset of aerial target for training and test, improve the deep learning-based detector Faster R-CNN, and specialize it in aerial target detection. Aiming at the peculiarities and requirements of aerial target detection, we propose the strategies such as accumulation of dilation, regional amplification, local tagging, adaptive threshold and spatio-temporal context to make up the shortage of Faster R-CNN that small weak or occluded targets can not be detected and improve the detection speed and accuracy. Experimental results show that the improved Faster R-CNN performs well under circumstances such as small weak or multiple targets, clutter, illumination changes, blur and large-area occlusion. Compared to the original Faster R-CNN, the mean average precision is improved by 16.7% on the built dataset, and the speed is 3 times faster.