This paper presents a comprehensive analysis of a computer vision-based threat detection system designed for X-ray baggage screening. The system utilizes advanced image processing methods, morphological analysis, and multi-criteria classification algorithms to achieve accurate identification of prohibited items while minimizing false alarms from safe objects such as electronics. The work addresses the task of automatic detection and classification of prohibited items in X-ray baggage images using transformer architectures. An innovative model is proposed that combines multi-scale processing and attention mechanisms, ensuring high accuracy in object segmentation and classification under complex conditions such as occlusion and noise. A comparative analysis demonstrates the superiority of transformers over traditional convolutional networks.