Citation

Suvon, Injamul Haque and Loh, Yuen Peng and Hashim, Noramiza and Mohd Isa, Wan Noorshahida and Ting, Choo Yee and Ghauth, Khairil Imran and Bhattacharijee, Arpita and Matsah, Wan Razali (2023) Business Category Classification via Indistinctive Satellite Image Analysis Using Deep Learning. International Journal on Advanced Science, Engineering and Information Technology (IJASEIT), 13 (6). pp. 2219-2230. ISSN 2088-5334

Text 23.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Satellite image analysis has numerous useful applications in various domains. Extracting their visual information has been made easier using remote sensing and deep learning technologies that intelligently interpret clear visual cues. However, satellite information has the potential for more complex tasks, such as recommending business locations and categories based on the implicit patterns and structures of the regions of interest. Nonetheless, this task is significantly more challenging due to the absence of obvious visual cues and the highly similar appearance of each location. This study aims to analyze satellite image similarity between business class categories and investigate the capabilities of state-of-the-art deep learning models for learning non-obvious visual cues. Specifically, a satellite image dataset is constructed using business locations and annotated with the business categories for image structural similarity analysis, followed by business category classification via fine-tuning of deep learning classifiers. The models are then analyzed by visualizing the features learned to determine if they could capture hidden information for such a task. Experiments show that business locations have significantly high SSIM regardless of categories, and deep learning models only recorded a top accuracy of 60%. However, feature visualization using Grad-CAM shows that the models learn biased features and disregard highly informative details such as roads. It is concluded that typical learning models and strategies are insufficient to effectively solve this complex visual problem; thus, further research should be done to formulate solutions for such non-obvious classifications with the potential to support business recommendation applications.

Item Type:	Article
Uncontrolled Keywords:	Road network, remote sensing
Subjects:	H Social Sciences > HE Transportation and Communications > HE1-9990 Transportation and communications (General) > HE7601-8700.9 Telecommunication industry. Telegraph Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	22 Feb 2024 04:28
Last Modified:	22 Feb 2024 04:28
URII:	http://shdl.mmu.edu.my/id/eprint/12102

Downloads

Edit (login required)