Improved YOLOv8-seg model for dangerous goods detection in x-ray security images

Detecting and segmenting multi-scale objects in complex backgrounds of X-ray security images is a significant area of research. Traditional methods for processing X-ray images have limitations in both efficiency and accuracy. To address these challenges, this paper presents an improved YOLOv8-seg model incorporating the Spatial Pyramid Dilated Convolution (SPD-Conv) module to enhance feature extraction capabilities, particularly for detecting and segmenting small objects in intricate backgrounds. By performing object detection and segmentation concurrently, the model significantly improves segmentation accuracy while maintaining detection precision, demonstrating enhanced robustness in handling overlapping and occluded scenarios. Experiments conducted on the PIDray dataset reveal that the proposed method achieves higher accuracy and efficiency in segmentation tasks, particularly excelling in the segmentation of small objects within complex scenes.