Open AccessAn Island Remote Sensing Image Segmentation Algorithm Based on A Fusion Network with Attention Mechanism
Author(s) -
Tianyuan Chen,
Hongfei Wang,
Hao Liu,
Peng Wu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1693/1/012179
Subject(s) - segmentation , computer science , artificial intelligence , focus (optics) , feature (linguistics) , convolutional neural network , image (mathematics) , pattern recognition (psychology) , image segmentation , interference (communication) , net (polyhedron) , aerial image , deep learning , image fusion , layer (electronics) , remote sensing , channel (broadcasting) , geography , mathematics , telecommunications , linguistics , philosophy , physics , geometry , organic chemistry , chemistry , optics
With the increasing importance of islands in many fields, it has become the focus of research to obtain information from island remote sensing images efficiently by using image semantic segmentation algorithm. In recent years, deep learning methods based on convolutional neural network have been widely used in image segmentation. However, in view of the problems that remote sensing images contain richer ratio information and complex background interference, we propose an island remote sensing image segmentation algorithm based on a fusion network with attention mechanism, called AFU-Net. The network is built on the basis of FC_U-Net [1] . An attention mechanism is added to pre-weight the shallow features before deep-shallow layer feature fusion, in order to enhance the response capability of the target features, suppress the background interference and improve the segmentation accuracy of the network. The testing and comparative experiments on NWPU-RESISC45 dataset show that the quantitative metrics and visual effects of AFU-Net are greatly improved compared to FC_U-Net, and are also superior to other three state-of-the-art methods, U-Net, FCN, SegNet, which indicates the effectiveness of our method.