The interactions between target features are more intricate in high-resolution remote sensing images, which encompass more feature in- formation such as texture, structure, geometry, and other geometric elements. These characteristics impede the process by which stan- dard convolutional neural networks execute feature classification on remote-sensing images and achieve optimal results. We suggested DMAU-Net, an attention-based multiscale maxpooling dense net- work based on U-Net for ground object categorization, as a solutionto this problem. The network is built with an inbuilt max-pooling module that uses dense connections in the encoder section to im- prove the feature map’s quality and, consequently, the network’s ca- pacity to extract features. Similarly, we present the Efficient Chan- nel Attention (ECA) module in the decoding process, which can am-plify the useful elements and stifle the superfluous ones.
Article Details
Unique Paper ID: 165091
Publication Volume & Issue: Volume 11, Issue 1
Page(s): 65 - 70
Article Preview & Download
Share This Article
Join our RMS
Conference Alert
NCSEM 2024
National Conference on Sustainable Engineering and Management - 2024