Research on Information Extraction and Planning of Urban Public Space Based on Deep Learning

Traditional methods for gathering information on urban public spaces are challenging and consume significant human and material resources. With the continuous advancement of remote sensing technology, higher resolution remote sensing images provide a reliable data source for extracting surface feature information. However, faced with the vast amount of high-resolution remote sensing image data, researchers encounter difficulties in practical applications due to the considerable workload, low extraction precision, and high complexity. In recent years, the rapid development of deep learning technologies has offered new avenues for related research in the field of computer vision. This paper focuses on the extraction of urban public space information using satellite imagery as the primary data source. The backbone network of DeepLabv3+ was modified by replacing Xception71 with ResNet101, and a channel attention mechanism was introduced to further optimize the model. Ultimately, the average Intersection over Union (IoU) of the DeepLabv3+ model with ResNet101 as the backbone increased by approximately 2.7 %. The trained DeepLabv3+ network was then utilized to segment and extract information regarding the built-up areas in City A. Statistics revealed that the proportions of roads, playgrounds, squares, vegetation, and building gaps within the public space in City A are <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$6.29 \%, 1.57 \%, 1.80 \%, 20.55 \%$</tex>, and 18.25 %, respectively, with the overall public space accounting for 48.46%.