导航

ACTA AERONAUTICAET ASTRONAUTICA SINICA ›› 2022, Vol. 43 ›› Issue (5): 325270.doi: 10.7527/S1000-6893.2021.25270

• Electronics and Electrical Engineering and Control • Previous Articles     Next Articles

Adaptive aerial object detection based on multi-scale deep learning

LIU Fang, HAN Xiao   

  1. Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
  • Received:2021-01-14 Revised:2021-02-08 Published:2021-03-01
  • Supported by:
    National Natural Science Foundation of China (61171119)

Abstract: Unmanned Aerial Vehicles (UAV) have been widely used in various fields, and object detection has become one of the key technologies in the field of UAV vision. Considering the problems of complex scenes, variable scales and too many small targets in UAV images, an adaptive aerial object detection algorithm is proposed based on multi-scale convolution. Firstly, the multi-scale feature extraction network is constructed, and the multi-scale convolution method is introduced. Different types of convolution are used to check different sizes of targets to expand the receptive field effectively. Secondly, the convolution module is designed according to the advantages of attention mechanism, and the feature weight is adaptively optimized to obtain more representational features. Finally, a prediction network is constructed based on multi-scale feature fusion. According to the characteristics of small targets, multi-level feature maps are selected to fuse into high-resolution feature maps, and object classification and boundary box regression are carried out on a single scale. Experimental results show that the proposed algorithm improves the object detection accuracy of UAV aerial images, and has good robustness.

Key words: Unmanned Aerial Vehicles (UAV), object detection, multi-scale convolution, attention mechanism, feature fusion

CLC Number: