| 摘 要: 针对主流视觉模型边缘部署适配难的核心问题,本研究提出轻量化算法优化、模块化架构以及边缘硬件适配的协同方案。以 YOLO-FastestV1.1 为骨干,结合知识蒸馏实现模型轻量化,创新设计 Sobol-PRM 路径规划算法,采用模块化架构解耦核心模块。实验表明,优化后模型体积压缩至 0.54MB,推理延迟对比优化前降低 72%,实现在边缘板卡端实现高效部署；多车场景路径规划成功率 94.1%,避障与路口识别性能优异。本研究的车群避障方案设计为快递车群资源在受限场景中的可靠部署提供量化依据,为边缘智能物流系统工程化落地提供关键技术支撑。关键词：智慧交通; 知识蒸馏; 路径规划; 边缘设备 |
| 关键词: 智慧交通 知识蒸馏 路径规划 边缘设备 |
|
中图分类号:
文献标识码:
|
|
| Research on Edge Deployment and Efficient Delivery Methods for Express Delivery Vehicle Fleets Based on Lightweight Models and Modular Architecture |
|
nichunyan, caizhi, xiangqipeng, xiaojian
|
Nanjing University of Posts and Telecommunications
|
| Abstract: Addressing the core challenge of adapting mainstream vision models for edge deployment, this study proposes a collaborative solution integrating lightweight algorithmic optimization, modular architecture, and edge hardware adaptation. Utilizing YOLO-FastestV1.1 as the backbone, model lightweighting is achieved through knowledge distillation, while an innovative Sobol-PRM path planning algorithm is designed alongside a modular architecture to decouple core modules. Experimental results show that the optimized model size is reduced to 0.54MB, with inference latency decreased by 72% compared to the pre-optimization baseline, enabling efficient deployment on edge devices. In multi-vehicle scenarios, the path planning success rate reaches 94.1%, with excellent obstacle avoidance and intersection recognition performance. The proposed obstacle avoidance framework for vehicle clusters provides a quantitative basis for reliable deployment of express delivery fleets in constrained environments and offers critical technical support for the engineering implementation of edge-intelligent logistics systems. |
| Keywords: Smart Transportation Knowledge Distillation Path Planning Edge Devices |
