Title: The feature extraction and fusion algorithm for multi-source data based on deep belief network

Authors: Xinjie Qian; Guixiang Hu; Yuqin Dai

Addresses: College of Digital Economy, Yibin Industry Polytechnic College, Yibin, 644000, China ' School of Economics and Trade Management, Yibin Vocational and Technical College, Yibin, 644000, China ' School of Electronic Information and Artificial Intelligence, Yibin Vocational and Technical College, Yibin, 644000, China

Abstract: This paper proposes DGACO-Net, a new model combining deep belief network (DBN), graph convolution network (GCN), and ant colony optimisation (ACO) to address the challenges of feature extraction and spatial relationship modelling in multi-source remote sensing data for land cover classification. DBN is used to extract advanced features, GCN captures spatial topological relationships, and ACO optimises hyperparameters to enhance model accuracy. Experimental results on the UC Merced Land Use and WHU-RS19 datasets demonstrate significant improvements in classification performance, with accuracies of 95% and 94%, respectively, outperforming benchmark models like SVM, random forest, and CNN. Ablation studies and feature visualisation validate the synergy of DBN, GCN, and ACO. DGACO-Net shows great potential for remote sensing image analysis and land resource management, offering an innovative solution for multi-source data classification.

Keywords: deep learning; multi-source data fusion; DBN; deep belief network; feature extraction; image classification; machine learning.

DOI: 10.1504/IJDS.2025.149862

International Journal of Data Science, 2025 Vol.10 No.2, pp.136 - 155

Received: 26 Dec 2024
Accepted: 25 Jun 2025
Published online: 14 Nov 2025 *

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