Unstructured text plays a crucial role in crime scene investigation, as it contains vital information about events, suspects, witnesses, and other relevant details. However, extracting meaningful information from unstructured police reports remains a challenging task. Despite the significance of extracting information from crime scene text, this area has received limited research attention. Few studies have adequately addressed the challenges specific to crime scene reports, resulting in a lack of comprehensive solutions. This research focuses on the importance of addressing this challenge by utilizing advanced deep learning techniques to generate scene graphs, enabling a structured representation of crime scene information. We propose a BERT-based NER (named entity recognition) applied on a custom dataset tailored to crime scene-related entities and relationships, facilitating more accurate and contextually informed information extraction. The generated scene graphs serve as powerful tools for crime scene investigation, enabling investigators to visualize complex relationships, uncover hidden connections, and gain a comprehensive understanding of the crime scene dynamics.