Citation

Suvon, Injamul Haque (2025) Satellite image-based business site recommendation via multi-visual feature fusion and classification. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

Satellite imagery holds immense potential across various applications in disaster response, business recommendation, agricultural decision-making and urban planning owing to its accessibility and ability to capture intricate details of the Earth’s surface over time. However, traditional analytical methods face challenges when extracting valuable information from such imagery. Deep learning has emerged as an effective solution by enabling direct, automated feature extraction, which facilitates the differentiation of different class categories. This study leverages deep learning techniques for the analysis and fusion of visual features from satellite imagery to determine suitable business types for specific locations in Malaysia. By scrutinizing surrounding urban structures and land characteristics, this study captures the visual context of business environments. Multiple deep learning models, employing transfer learning and supervised classification, are used to analyse satellite and map images, extracting deep features alongside hand-crafted features and Scale Invariant Feature Transform (SIFT) features from road network images to represent distinctive visual patterns associated with different business types for classification-based recommendation. Additionally, road network analysis incorporates hand-crafted statistical and visual features to complement deep visual representations, demonstrating that combined feature sets as more reliable than relying solely on individual feature sets for business suitability assessments. This study analyses 12,500 satellite images across five classes, generating four feature sets with dimensions of 128 and 512. The 512dimensional concatenated feature set achieved an accuracy of 0.57 using ResNet50 and 0.61 with ERNet. Furthermore, the inclusion of analytical structured data resulted in additional performance improvements, highlighting the importance of multi-modal feature fusion and contextual information.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: Q325.73 .S88 2025
Uncontrolled Keywords:	Deep learning (Machine learning)
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	14 Apr 2026 01:53
Last Modified:	14 Apr 2026 01:53
URII:	http://shdl.mmu.edu.my/id/eprint/15701

Downloads

Edit (login required)