Kamarudin, Puteri Nur Farzanah Faghira (2025) Milk quality analysis with multimodal data fusion: combining image and numerical features. Masters thesis, Universiti Teknikal Malaysia Melaka.
![[img]](http://eprints.utem.edu.my/style/images/fileicons/text.png) |
Text (24 pages)
Milk quality analysis with multimodal data fusion - combining image and numerical features (24 pages).pdf - Submitted Version Download (941kB) |
|
Text (Full Text)
Milk quality analysis with multimodal data fusion - combining image and numerical features.pdf - Submitted Version Restricted to Registered users only Download (2MB) |
Abstract
Ensuring the quality of milk is a critical challenge in the food industry, with significant implications for consumer safety, economic efficiency, and supply chain reliability. Traditional methods of milk quality assessment, such as chemical analysis and sensory evaluations, while accurate, are labor-intensive, costly, and unsuitable for real-time or large scale applications. The growing adoption of artificial intelligence (AI) has introduced advanced methods for automating these processes, yet most existing AI-based approaches rely on single-modality data, such as either visual features or numerical measurements. These methods are often limited in their ability to capture the multidimensional nature of milk quality indicators, resulting in reduced prediction accuracy and robustness. This study seeks to address these limitations by developing a multimodal deep learning model that combines image and numerical data for classifying milk quality into three categories which are good, spoiling, and spoiled. This study employs intermediate fusion and late fusion techniques to combine the outputs of pre-trained models for each modality. This study also highlights the potential of multimodal deep learning in addressing the complex interplay of physical and chemical factors influencing milk quality. Results show that the intermediate fusion technique achieved an accuracy of 98.87% while late fusion technique using concatenation with proposed layers achieved an accuracy of 99.77% This proves that the multimodal framework outperforms single-modality approaches in terms of accuracy, scalability, and generalizability across diverse milk storage and spoilage conditions. By incorporating complementary data sources, the proposed framework achieves a holistic and automated approach to milk quality assessment, suitable for industrial-scale applications. Additionally, the study contributes to the advancement of fusion strategies in multimodal AI, demonstrating the efficacy of combining heterogeneous data for improved decision making. While the findings are promising, the research also identifies several areas for further investigation. Future work could explore the integration of additional modalities, such as odor sensors or spectral data, to further enhance classification performance. Extending the framework to other perishable goods could also validate its applicability in broader food quality assessment contexts. This study not only bridges an important gap in the literature but also sets a foundation for scalable, efficient, and robust AI-driven solutions in food safety and quality control.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Classification, Intermediate fusion, Late fusion, Milk quality, Multimodal deep learning |
| Subjects: | T Technology T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Faculty Of Electronics And Computer Technology And Engineering |
| Depositing User: | Norhairol Khalid |
| Date Deposited: | 21 Jan 2026 07:13 |
| Last Modified: | 21 Jan 2026 07:13 |
| URI: | http://eprints.utem.edu.my/id/eprint/29437 |
| Statistic Details: | View Download Statistic |
Actions (login required)
 |
View Item |