In order to improve the detection performance of urban road vehicles from the perspective of unmanned aerial vehicle（UAV） aerial photography

an aerial photography urban road vehicle detection network is improved and built based on single shot detector（SSD） deep learning network framework.On the one hand

the prior knowledge that most vehicles are small targets from the perspective of UAV aerial photography is fully explored

and the default candidate box parameter information suitable for aerial photography vehicle targets is obtained by using K-means++ clustering algorithm.On the other hand

in order to ensure the effective transfer and accurate extraction of small-scale target features

the structure of the benchmark SSD network is improved

and a learnable low-pass filter layer with anti-aliasing effect is designed in the skeleton of the feature extraction network

and the large-size feature map for small-scale vehicle target detection regression is retained.Experiments show that the detection precision of the proposed method is improved by 4.3 percentage points compared with the benchmark SSD network in real-time

which can significantly improve the detection effect of small-scale vehicle targets and improve the overall detection precision of urban road vehicle targets from the perspective of aerial photography.