Deep learning-based analysis of multimedia contents in forensic evidence investigation process

Today, with the development of technology, digital environments have become an indispensable part of human life. This situation causes the amount of data to be examined in forensics evidence investigations to increase day by day. Especially forensic examination of multimedia data is more difficult as it cannot be easily searched due to its structure. This situation has led to the need for automatic evidence extraction systems in order not to delay forensic evidence examinations and to ensure that the process proceeds without errors. In recent years, successful applications have been developed in various areas such as image and audio classification, object detection, speech recognition, and speaker identification on multimedia data using deep learning methods. The utilization of deep learning methods for automated evidence extraction in forensic investigation processes will contribute to the forensic examination process and alleviate the workload of forensic examiners to a significant extent. Within the scope of this thesis, models to automatically analyze multimedia data in the forensic evidence examination process are proposed. The proposed models offer solutions such as automatic object detection in image and video data and accelerating the processing of video data, automatic detection of obscenity and child abuse in image and video data, and detection of keywords in audio data. The proposed object detection system achieves 85 mAP on a newly created dataset of 19000 images. The proposed obscene content and child sexual abuse detection model achieves an accuracy of 99% on the NPDI benchmark dataset. The keyword detection model achieves an accuracy of 77% on the newly created Turkish speech dataset. The proposed models have shown that successful support systems can be developed in the forensic evidence examination process with experimental results. The studies conducted within the scope of this thesis have the potential to guide future studies in the field of digital forensics for the development of the evidence examination process.