强化特征提取能力的航拍图像目标检测算法

Target Detection Algorithm in Aerial Images with Enhanced Feature Extraction

航拍视角下的地面交通车辆目标自主检测是智能交通系统中获取交通信息的新兴手段。近年来,随着深度学习在诸多领域应用取得显著的成功,卷积神经网络也开始应用于视频图像的目标检测中。针对航拍图像下的较小车辆目标,结合密集的拓扑结构和最优的池化策略,论文重构了深度卷积神经网络模型,重点强化网络的特征提取能力,用于提升小目标检测性能。论文提出的检测模型在NVIDIA 1080ti平台上,对航拍图像不同类型的车辆目标检测进行了实验仿真。实验结果表明,提出的检测方法对较小目标检测能力鲁棒性高,快速有效,并实现了实时检测。

Autonomous detection of ground traffic vehicles in aerial images is an emerging means of acquiring traffic informa⁃tion in intelligent transportation systems.In recent years,with the remarkable success of deep learning in many fields,convolutional neural networks have also begun to be applied to target detection in video images.Aiming at the small vehicle target in aerial imag⁃es,combined with dense topology and optimal pooling strategy,a deep convolutional neural network model is reconstructed in this paper for strengthening the ability on feature extraction of the network so as to improve the performance on small target detection.With detection model proposed,the experimental simulation of various vehicle target detection in aerial images have been done on the NVIDIA 1080ti platform.The experimental results have shown that the proposed method is robust to small target detection,fast and effective,and realizes real-time detection.