5:15 PM - 7:15 PM
[U04-P07] DEVELOPMENT OF AN AUTOMATED MULTIPLE OBJECT DETECTION (AMOD) TECHNIQUE FOR SCANNED TOPOGRAPHIC MAPS USING DEEP LEARNING
Keywords:Automated Multiple Object Detection, Convolutional Neural Network, Deep Learning, Geospatial Artificial Intelligence, Topographic Maps
Recently, object detection in image analysis has seen extensive development, particularly for satellite imagery and scanned topographic maps. However, there is a noticeable gap in research attention when it comes to topographic maps. Topographic maps present unique challenges due to the abundance of multiple object requirements, the less attention for development of the technique specifically for topographic map as the dataset to ascertain its applicability, efficiency, and accuracy. This study addresses these gaps by focusing on the development of Deep-Learning Based Automated Multiple Object Detection (AMOD) technique of scanned topographic map. The objective of the study is to evaluate the developed automated multiple object detection technique for scanned topographic maps in terms of applicability, accuracy, and efficiency. The system undergoes evaluation through system testing, which evaluates its acceptance, usability, and scalability. The results highlight that the proposed technique shows better achievement for multiple object detection on a scanned topographic map. This study utilizes YOLOv8m-seg as the primary deep learning model for dataset training, achieving consistently high accuracy in detecting objects within scanned topographic maps, with a mean average precision (mAP) exceeding 90%. Throughout the successful achievement of this study, it will enhance the preservation activities of heritage documents, especially topographic map in Malaysian library institutions as well as worldwide. Areas for improvement are identified, emphasizing the importance of iterative development and user feedback integration for enhancing system functionality. The deep learning technique proves effective in detecting objects in a topographic map. The study's impact lies in its contribution to an integrated framework, and the proposed technique offers an applicable, efficient, and accurate for multiple object detection in topographic maps. Through highlighting the complexities and nuances of multiple object detection technique in this context, the study's novelty contributes to the advancement of techniques in Geospatial Artificial Intelligence (GeoAI) and digital cartography.