Media of University of Babylon | Seminar for Topic Approval of PhD Student Nidaa Kareem Shalaan – Software Department / College of Information

Media of University of Babylon - كلية الطب

Seminar for Topic Approval of PhD Student Nidaa Kareem Shalaan – Software Department / College of Information

by : Duhaa Fadill Abbas

Date : 18/5/2025 31:06:08 PM

Veiw : 50

Seminar for Topic Approval of PhD Student Nidaa Kareem Shalaan – Software Department / College of Information Technology

Duhaa Fadill Abbas
As part of the academic activities for postgraduate students, the Software Department at the College of Information Technology – University of Babylon organized a seminar for PhD candidate Nidaa Kareem Shalaan to present and gain approval for her dissertation proposal titled:

"Classification and Reconstruction of GIS Image Components based on Geometric Feature and Deep Learning"

The seminar was held under the supervision of Prof. Dr. Tawfiq Abdul Khaliq Abbas, and in the presence of the scientific committee composed of:

- Prof. Dr. Nabeel Hashim Kaghd – Chair
- Prof. Dr. Nidaa Abdulmuhsin Abbas – Member
- Prof. Dr. Iman Saleh Sakban – Member
- Asst. Prof. Dr. Mai Abdul Muneim – Member
- Asst. Prof. Dr. Ahmed Habeeb Saeed – Member

The seminar addressed the crucial role of precise classification and reconstruction of Geographic Information System (GIS) image components in enabling a wide range of geospatial applications. The proposed research introduces an advanced framework that leverages geometric feature extraction and deep learning to improve the interpretation and generation of GIS imagery.

The methodology begins with the development of a Holistically-Nested Edge Detection (HED) system to extract fine-grained, multi-scale edge features, effectively capturing the geometric boundaries of GIS components such as roads, buildings, and natural terrains. Subsequently, the study proposes a novel algorithm to construct a mesh of irregular components based on the segmented image. These irregular mesh structures align more accurately with object boundaries, leading to more precise segmentations—particularly in edge or surface-based detection methods.

Following segmentation, the components are classified using an adaptive point-based network that also eliminates irrelevant components. The classifier is designed to learn high-level semantic representations from spatially distributed, irregular point features, thus enabling accurate classification of heterogeneous image elements.

This research aims to contribute significantly to the field of geospatial image analysis by providing a robust and intelligent system for processing and understanding complex GIS data.