Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering, Journal Year: 2024, Volume and Issue: 67(1), P. 27 - 42
Published: Aug. 22, 2024
High spatial resolution remote sensing images contain substantial detailed multi-scale objects. Convolutional neural networks (CNNs) are not efficient enough for detecting these objects of varying sizes. Among the multitude CNN approaches, Nested U-Net (UNet++) model shows great potential to capture more complex details by progressively enriching highresolution feature maps. However, there is room improving architecture increasing its ability detect The nested blocks used in this rely on standard convolutional layers, which limited efficacy capturing pixel information. Thus, larger receptive fields required extract Although many approaches available model, methods usually make computational efforts very heavy. Therefore, study uses dilated convolutions UNet broaden field without augmenting demand. To extent, paper performs experiments with different dilation rates convolution understand benefits employing architecture. Experiments using two image sets show that well containing both visible and multispectral wavelengths. While being able provide performance improvement, experimental results also demonstrate only optimal rate scheme proposed approach beneficial.
Language: Английский