DeiT-OEL: Data Efficient Image Transformer with Ontology-Based Evidence Linking for Intelligent Crime Scene Reconstruction

Intelligent crime scene reconstruction is critical in forensic informatics to provide an accurate analysis that supports criminal investigations and legal proceedings. This task becomes more challenging because of the complexity of visual crime evidence, including varying lighting conditions, occlusions, and scene clutter. However, traditional models struggle to maintain accuracy and effectively link evidence at complex crime scenes. To overcome this problem, this paper proposes a Data-efficient Image Transformer with Ontology-based Evidence Linking (DeiT- OEL) method for crime scene image classification. To improve evidence linking and reasoning, ontology-based evidence linking using a geographic query language for RDF data (GeoSPARQL) is integrated to connect the spatial and semantic relationships between crime scene elements. The proposed DeiT-OEL method is evaluated using the UCF Crime and Chicago Crime datasets, where preprocessing includes image normalization, resizing, and augmentation to improve the data quality. The experimental results show that the proposed DeiT-OEL method achieves a classification accuracy of 99.99%, significantly outperforming existing traditional models.