Journal of Engineering, Project, and Production Management, 2026, 16(5), 2025-357
Rural Landscape Image Processing: Improved DeepLab v3+ Segmentation and K-means Color Quantification
Director, Department of Design, Zhongge Art College, Guangdong Ocean University, Zhanjiang, 524088, China, E-mail: qifang2018@outlook.com
Project Management
Received December 15, 2025; revised February 10, 2026; accepted May 26, 2026
Available online June 17, 2026
Abstract: To improve the accuracy and effectiveness of digital optimization of rural landscapes, a method combining improved Deep Laboratory v3+ (DeepLab v3+) and optimized K-Means Clustering (KMC) quantification is proposed. In the landscape image segmentation process, the Deep Laboratory v3+ model is improved by introducing cross-stripe pooling, a convolutional attention mechanism, and a residual feature fusion module. The experiment verified that the average segmentation accuracy of the model reached 99.1%, the average Dice coefficient was 0.906, the average intersection to union ratio was 0.881, and the average segmentation speed was increased to 0.22 seconds per image. All indicators were better than the comparison model. The ablation experiment showed that the residual feature fusion module could improve the intersection to union ratio by 1.7%, with the most significant improvement in segmentation performance. In the color quantization optimization stage, the node index method, elbow rule, and average error vector optimization K-means algorithm were used. Validation results showed that the weighted average peak signal-to-noise ratios for natural, semi-natural, and cultural rural landscapes were 32.4 dB, 34.5 dB, and 36.6 dB, respectively. The average quantization speed was 1.6 seconds per image, and the average quantization error was only 0.781, outperforming other comparison methods. Through precise segmentation of landscape elements and efficient color quantification, the study has enriched color levels and enhanced aesthetic value in rural landscapes. This provides operational technical support for digital design, ecological livability planning, and other management decisions in rural landscapes.
Keywords: Rural landscape, DeepLab v3+, k-means, cross stripe pooling, residual feature fusion.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Qi, F. (2026). Rural Landscape Image Processing: Improved DeepLab v3+ Segmentation and K-means Color Quantification. Journal of Engineering, Project, and Production Management, 16(5), 2025-357.
