Because of their portability, electric motorcycles are usually pushed into elevators by residents and charged in the home, which has serious safety risks. Traditional manual-based methods to manage this behavior have poor monitoring effects and high costs. As for automatic management systems using artificial intelligence (AI), the deployment method matters. Cloud-based deployment methods have the disadvantages of high latency, high risk of privacy leakage, and heavy network transmission loads. In this paper, we propose a highly secure edge-intelligent electric motorcycle management system for elevators. By using edge-based deployment method, the monitor pictures are processed locally without being uploaded to the cloud, which can effectively resist network attacks and prevent residents’ private data from being leaked. To improve the system security, we fully analyze the challenges faced in the application scenarios and introduce security threat identification (STI-1H8) model to identify the security threats. In addition, we propose several data enhancement methods to improve the system recognition accuracy. Experimental results show that our system can achieve a high recall rate of 0.82. By using data enhancement and data mixing strategies, it can reduce the misjudgment rate by 0.35. Moreover, compared to cloud computing, our edge-based method can reduce the latency by 19.6%, meeting real-time requirements.

中文翻译：

---

高度安全的电梯边缘智能电动摩托车管理系统

由于其便携性，电动摩托车通常被居民推入电梯并在家里充电，这具有严重的安全隐患。传统的基于手动的方法来管理此行为具有较差的监视效果和较高的成本。对于使用人工智能（AI）的自动管理系统，部署方法很重要。基于云的部署方法具有以下缺点：高延迟，高隐私泄露风险和沉重的网络传输负载。在本文中，我们提出了一种用于电梯的高度安全的边缘智能电动摩托车管理系统。通过基于边缘的部署方法，监控图像无需上传到云端即可在本地进行处理，可以有效抵御网络攻击，防止居民的私人数据泄露。为了提高系统安全性，我们充分分析了应用场景中面临的挑战，并引入了安全威胁识别（STI-1H8）模型来识别安全威胁。另外，我们提出了几种数据增强方法来提高系统识别的准确性。实验结果表明，我们的系统可以实现0.82的高召回率。通过使用数据增强和数据混合策略，可以将误判率降低0.35。而且，与云计算相比，我们的基于边缘的方法可以将延迟降低19.6％，满足实时要求。我们提出了几种数据增强方法来提高系统识别的准确性。实验结果表明，我们的系统可以实现0.82的高召回率。通过使用数据增强和数据混合策略，可以将误判率降低0.35。而且，与云计算相比，我们的基于边缘的方法可以将延迟降低19.6％，满足实时要求。我们提出了几种数据增强方法来提高系统识别的准确性。实验结果表明，我们的系统可以实现0.82的高召回率。通过使用数据增强和数据混合策略，可以将误判率降低0.35。而且，与云计算相比，我们的基于边缘的方法可以将延迟降低19.6％，满足实时要求。