Title: Urban and rural landscape design analysis technology based on image semantic segmentation deep learning and remote sensing image processing

Authors: Xuen Hou; Qiuyue Shan

Addresses: Department of Environmental Art, Hebei University of Environmental Engineering, Qinhuangdao, 066102, China ' Department of Environmental Art, Hebei University of Environmental Engineering, Qinhuangdao, 066102, China

Abstract: This study addresses the rising demand for high-quality urban and rural environments by proposing a deep learning-based landscape analysis method. This study proposes a landscape analysis method based on deep learning to address the growing demand for high-quality urban and rural environments. This method first uses Deeplab-v3+as the model foundation. Then, two-dimensional decomposition is used for segmentation to reduce the number of convolution parameters, and depth wise separable convolution operations are optimised to improve the accuracy of the training model. Finally, select two representative deep and shallow layers for feature fusion operations at different levels. The results showed that compared with X-DeepLab-v3+ and the original Deeplab-v3+, the optimised model achieved up to 58.3% higher intersection-over-union and 47.6% higher pixel accuracy. Additionally, it demonstrated strong alignment with real-world compactness and connectivity values, reaching 99% and 98% respectively. The model enables precise analysis of landscape fragmentation and ecological connectivity, offering valuable insights for environmental protection and landscape design.

Keywords: image semantic segmentation; ISS; Deeplab-v3+; deep learning; landscape analysis; remote sensing image processing.

DOI: 10.1504/IJEE.2025.151270

International Journal of Environmental Engineering, 2025 Vol.13 No.4, pp.375 - 390

Received: 14 Apr 2025
Accepted: 20 Aug 2025
Published online: 20 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article