Media of University of Babylon | Ph.D. Dissertation at the College of Information Technology Explores Land Cover Classification of Satellite SAR Imagery Using Deep Learning Techniques

Media of University of Babylon - كلية الطب

Ph.D. Dissertation at the College of Information Technology Explores Land Cover Classification of Satellite SAR Imagery Using Deep Learning Techniques

by : Duhaa Fadill Abbas

Date : 11/9/2025 50:14:06 PM

Veiw : 68

Ph.D. Dissertation at the College of Information Technology Explores Land Cover Classification of Satellite SAR Imagery Using Deep Learning Techniques

Duhaa Fadill Abbas
The Department of Software at the College of Information Technology, University of Babylon, witnessed on Thursday, September 11, 2025, the public defense of a Ph.D. dissertation presented by researcher Safaa Hadi Khudair Abis, entitled “Land Cover Classification of Synthetic Aperture Radar (SAR) Satellite Images Using Machine Learning.” The dissertation was supervised by Prof. Dr. Hidhab Khalid Abis and held in the College Conference Hall.

The research addressed the vital topic of land cover classification, which constitutes a fundamental aspect of remote sensing. Accurate classification provides critical data that supports diverse applications, including environmental monitoring, climate change assessment, urban planning, and natural disaster management. The dissertation emphasized that achieving high precision in classification is essential to maximize the benefits of these applications, particularly with the rapid advancement of deep learning techniques that have emerged as transformative tools in recent years.

The study aimed to develop an integrated model based on Sentinel-1 (SAR) and Sentinel-2 (Optical) satellite imagery. A novel strategy was proposed to enhance and expand training samples through the application of zonal statistics, which significantly increased the number of usable training samples from 300 to more than 11,400 pure pixels. This methodological innovation contributed to improving the accuracy and robustness of the classification process.

The dissertation introduced three advanced models. The first relied on spatial data using the U-Net architecture and achieved an accuracy rate of 97%. The second model, based on temporal data, employed the Long Short-Term Memory (LSTM) approach, processing seasonal time-series data across four periods, and likewise attained an accuracy of 97%. The third model integrated both spatial and temporal features through an attention-based fusion mechanism, which enabled the selection and combination of the most informative features. This hybrid spatio-temporal model demonstrated superior performance, achieving an overall accuracy of 99.52% and a Kappa coefficient of 98.97%.

The results of the dissertation highlighted that the proposed model represents a significant advancement in the field of land cover classification, offering a high degree of precision and reliability. Such an approach provides a promising foundation for supporting scientific research and practical applications in remote sensing, while also contributing to future projects in environmental planning, natural resource management, and sustainable development.

ت	اسماء اعضاء لجنة المناقشة	اللقب العلمي	الاختصاص الدقيق	مكان العمل	المنصب
1	د. مهدي نصيف جاسم	استاذ	تنقيب بيانات	جامعة تكنولوجيا المعلومات و الاتصالات	رئيساً
2	د. وفاء محمد سعيد	استاذ	ذكاء اصطناعي معالجة صورية	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
3	د. احمد خلفة عبيد	استاذ مساعد	ذكاء اصطناعي و انظمة معلومات	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
4	د. احمد حبيب سعيد	استاذ مساعد	نظم معلومات و شبكات	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
5	د.صفا سعد عباس	استاذ مساعد	وسائط متعددة و امنية المعلومات	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
6	د. هضاب حالد عبيس	استاذ	ذكاء اصطناعي	جامعة بابل	عضوا و مشرفا