In the process of intelligentization of security checks, research on the recognition of whether pedestrians are carrying objects is essential. Aiming at the problem that pedestrian detection neglects to detect their carry-on together and leads to misdetection and omission in complex scenes due to multi-scale and occlusion, a method for simultaneous recognition of pedestrians and carry-on in security checks is proposed. An easy-to-deploy lightweight deep learning network model PCD-MLNet is proposed for detecting pedestrians and carrying object targets. The improved HGNetV2 is utilized as the backbone network of the model to extract input features. A dilated multi-branch residual module is proposed, which combines the large kernel convolution operation to enhance the hierarchical and detailed awareness of the pedestrian and the carrying objects feature extraction. The EIoU is a localization regression loss function for the detection network to improve the model's ability to localize the target. In the created pedestrian-carrier dataset experiments, the PCD-MLNet performs well, with a detection accuracy of 72.3%. After pruning the redundant channels, the number of parameters in the final model decreases by 58.6% compared to the baseline model, and the video inference speed increases by 35%. The simulated security check experiments on the simulation platform obtaine good results.