ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume V-2-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-2-2020, 909–916, 2020
https://doi.org/10.5194/isprs-annals-V-2-2020-909-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-2-2020, 909–916, 2020
https://doi.org/10.5194/isprs-annals-V-2-2020-909-2020

  03 Aug 2020

03 Aug 2020

RESOLUTION-AWARE NETWORK WITH ATTENTION MECHANISMS FOR REMOTE SENSING OBJECT DETECTION

Z. Tian, W. Wang, B. Tian, R. Zhan, and J. Zhang Z. Tian et al.
  • College of Electronic Science and Technology, National University of Defense Technology, Changsha, China

Keywords: Remote Sensing, Object Detection, Convolutional Neural Network, Backbone Network, Attention Mechanism

Abstract. Nowadays, deep-learning-based object detection methods are more and more broadly applied to the interpretation of optical remote sensing image. Although these methods can obtain promising results in general conditions, the designed networks usually ignore the characteristics of remote sensing images, such as large image resolution and uneven distribution of object location. In this paper, an effective detection method based on the convolutional neural network is proposed. First, in order to make the designed network more suitable for the image resolution, EfficientNet is incorporated into the detection framework as the backbone network. EfficientNet employs the compound scaling method to adjust the depth and width of the network, thereby meeting the needs of different resolutions of input images. Then, the attention mechanism is introduced into the proposed method to improve the extracted feature maps. The attention mechanism makes the network more focused on the object areas while reducing the influence of the background areas, so as to reduce the influence of uneven distribution. Comprehensive evaluations on a public object detection dataset demonstrate the effectiveness of the proposed method.