Dangerous Objects Detection and Segmentation in X-Ray Images of Passenger Goods Using YOLOV8

Security check using X-ray machine is often found at public places, especially airports, to mitigate criminal activities or even unwanted accidents. However, there exists the possibility of overlooked dangerous objects, especially during busy hours. Moreover, obscured dangerous objects becomes another challenge for the security personnels. This study used YOLOV8 model to detect and segment dangerous objects in X- ray images using the PIDray dataset. Therefore, this study extensively reviews the performance of YOLOV8 model by comparing multiple combinations of its hyperparameters to achieve the most optimal version of YOLOV8 model. There were 2 model hyperparameters that were tuned, which are epochs and batch size. The findings in this study showed that every model is able to give promising and competitive mAP results with little mAP score differences between each model. All failures to detect and segment dangerous objects are caused by two main factors, including small-sized and obscured dangerous objects. Through detailed evaluations, yolov8s-seg model with 50 epochs and 16 batch size is considered to be the most optimal version as it has the fastest training time yet able to provide competitive results compared to the other models. Future development of this study will include adding more variances of dangerous objects in X-ray images to increase model generalization ability.