Remote Sensing Image Retrieval by Multi-Scale Attention-Based CNN and Product Quantization
- Resource Type
- Conference
- Authors
- Chu, Jun; Li, Linhao; Xiao, Xiaowu
- Source
- 2021 40th Chinese Control Conference (CCC) Chinese Control Conference (CCC), 2021 40th. :8292-8297 Jul, 2021
- Subject
- Computing and Processing
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Deep learning
Visualization
Quantization (signal)
Satellites
Costs
Image retrieval
Feature extraction
Remote Sensing (RS)
Image Retrieval
Product Quantization
Multi-Scale Attention
- Language
- ISSN
- 1934-1768
With the development of information technology, all kinds of image information are expanding. It has been a hot issue in computer vision to quickly retrieval interested images from data sets. The complexity of remote sensing images brings new challenges to the retrieval process. This paper presents a new method for remote sensing image retrieval. In our proposed multi-scale attention-based convolutional neural network with improved product quantization method (APQ), we first use deep neural network with visual attention mechanism to extract feature representation of remote sensing images, and then we use an improved product quantization method to reduce the dimension of the features for the purpose of reducing the retrieval computation cost. Experiments on two remote sensing datasets Satellite Remote Sensing and NWPU show that our APQ method can outperform some state-of-the-art remote sensing image retrieval methods.