Abstract: To address the challenges of insufficient detection accuracy for prohibited items in security inspection X-ray images, the susceptibility to deep network information loss, and the constraints of resource-limited deployment devices, this paper proposes an improved YOLO11 network model for accurate detection and segmentation of prohibited items. First, a DWR (Dilated Wide Residual) module is introduced to enhance the C3k2 structure, expanding the receptive field and strengthening the extraction capability of high-level features. Second, a Dy-SSFF (Dynamic Skip-Spatial Feature Fusion) module is proposed to deeply fuse the high-level semantic information extracted by the DWR module with fine-grained detail features of prohibited items, generating multi-scale fused feature maps. This suppresses the loss of critical detail information and improves the model's perception of multi-scale objects. Finally, a lightweight shared convolutional normalization strategy is adopted to improve the segmentation head, enhancing sensitivity to multi-scale features while reducing the model's parameter count and computational complexity. Experimental results on the Xray-battery dataset demonstrate that, compared to the original YOLO11n-seg model, the proposed model improves mAP@0.5 and mAP@0.5:0.95 by 1.5% and 3.5% for bounding boxes, and by 2.1% and 2.8% for masks, respectively. Simultaneously, the parameter count is reduced by 7%, and the computational cost is reduced by 8%. The proposed model achieves high-precision detection and segmentation while being deployable on resource-constrained devices, providing effective technical support for real-world security inspection X-ray image scenarios.