Enhancing baggage inspection through computer vision analysis of x-ray images

This research work explores the utility of deep learning algorithms in enhancing the accuracy of weapon detection, specifically guns, within x-ray images of travel bags. Utilizing Faster R-CNN as a baseline model, the research aims to augment detection metrics including accuracy, precision, and recall, thereby fortifying security screening procedures. A comparative study was executed between the Faster R-CNN model and a hybrid model that integrated the Segment Anything (SAM) algorithm with Faster R-CNN. Evidently, the hybrid model displayed an edge in performance with the highest accuracy rate of 86.34%, a marked increase from the 72.02% accuracy of Faster R-CNN alone. The fusion model demonstrated superior precision, signaling a decrease in false positive instances, although it faced a higher rate of false negatives, as revealed by its recall rate. This study also unearths data limitations that could potentially be inhibiting maximum model performance, given the discrepancy between available training data and the sheer volume of the comprehensive SIXray dataset. The research concludes by charting avenues for future investigation which include data augmentation, SAM model pre-training, and expansion of detection capabilities to encompass a broader array of weapons. This body of work establishes a framework for advancing security measures through the application of artificial intelligence.