This research aims to see if it can improve computer-based anomaly detection systems for contraband detection in correctional facilities using x-ray imaging. In this research, we will investigate how the performance of these systems can be improved by increasing the dataset size and increasing the height of the x-ray images. We use YOLOv5, a state-of-the-art single-stage algorithm for object detection, and apply tiling methods to increase the performance for smaller anomalies. We evaluate our system on two datasets with different amounts of images. We also evaluate our system on images with increased height. We find that increasing the dataset size significantly improves the performance of the system. Despite the fact that increasing the image height improved the performance, the results were overall inconclusive and need further research.
