School of Cyber Security and Computer, Hebei University, Baoding, Hebei, China.
Information Technology and Electrical Engineering, University of Oulu, Oulu, Finland.
Math Biosci Eng. 2023 Jan;20(2):2793-2814. doi: 10.3934/mbe.2023131. Epub 2022 Dec 1.
With the explosive growth of edge computing, huge amounts of data are being generated in billions of edge devices. It is really difficult to balance detection efficiency and detection accuracy at the same time for object detection on multiple edge devices. However, there are few studies to investigate and improve the collaboration between cloud computing and edge computing considering realistic challenges, such as limited computation capacities, network congestion and long latency. To tackle these challenges, we propose a new multi-model license plate detection hybrid methodology with the tradeoff between efficiency and accuracy to process the tasks of license plate detection at the edge nodes and the cloud server. We also design a new probability-based offloading initialization algorithm that not only obtains reasonable initial solutions but also facilitates the accuracy of license plate detection. In addition, we introduce an adaptive offloading framework by gravitational genetic searching algorithm (GGSA), which can comprehensively consider influential factors such as license plate detection time, queuing time, energy consumption, image quality, and accuracy. GGSA is helpful for Quality-of-Service (QoS) enhancement. Extensive experiments show that our proposed GGSA offloading framework exhibits good performance in collaborative edge and cloud computing of license plate detection compared with other methods. It demonstrate that when compared with traditional all tasks are executed on the cloud server (AC), the offloading effect of GGSA can be improved by 50.31%. Besides, the offloading framework has strong portability when making real-time offloading decisions.
随着边缘计算的爆炸式增长，数十亿个边缘设备中产生了大量的数据。在多个边缘设备上进行对象检测，同时平衡检测效率和检测精度确实非常困难。然而，考虑到现实挑战，如有限的计算能力、网络拥塞和长延迟，很少有研究来调查和改进云计算和边缘计算之间的协作。为了解决这些挑战，我们提出了一种新的多模型车牌检测混合方法，在边缘节点和云服务器上处理车牌检测任务时在效率和准确性之间进行权衡。我们还设计了一种新的基于概率的卸载初始化算法，该算法不仅可以获得合理的初始解，而且有助于提高车牌检测的准确性。此外，我们通过引力遗传搜索算法（GGSA）引入了自适应卸载框架，该框架可以综合考虑车牌检测时间、排队时间、能耗、图像质量和准确性等影响因素。GGSA 有助于提高服务质量（QoS）。广泛的实验表明，与其他方法相比，我们提出的 GGSA 卸载框架在边缘和云协同的车牌检测方面表现出良好的性能。与传统的所有任务都在云服务器上执行（AC）相比，GGSA 的卸载效果可以提高 50.31%。此外，该卸载框架在进行实时卸载决策时具有很强的可移植性。
