<p>The importance of object detection in intelligent logistics applications has become increasingly prominent, but the high computational cost and limited real-time performance hinder its deployment on edge devices with constrained computational capabilities, especially in complex logistics scenarios. To tackle these challenges, we propose an innovative Dynamic Multiscale Feature Fusion Detection Network (DMFFDet). This network enhances feature extraction efficiency through a computation-efficient FasterNeMt module, significantly reducing the computational load and parameter count of the Backbone. At the same time, we design a Dynamic Multiscale Feature Fusion Module (DyFusion), which effectively extracts cross-scale feature information using dynamic sampling and a streamlined multi-layer fusion strategy. By integrating DyFusion into the Neck architecture, we construct DMFFNeck, which enhances the extraction of contextual and background feature information. Additionally, we combine the non-monotonic focusing factor from WIoUv3 with GIoU to propose a novel loss function, WGIoU, that enhances both the accuracy and robustness of bounding box localization. Extensive experiments demonstrate that DMFFDet outperforms existing methods on the Safefork dataset and further confirms its effectiveness and generalization on the KITTI autonomous driving benchmark. The proposed model achieves low computational complexity and high detection accuracy, and satisfies real-time detection requirements in complex logistics scenarios.</p>