Razaque Abdul, Bektemyssova Gulnara, Yoo Joon, Alotaibi Aziz, Ali Mohsin, Amsaad Fathi, Amanzholova Saule, Alshammari Majid
School of Computing, Gachon University, Seongnam-si 13120, Republic of Korea.
Department of Computer Engineering, International Information Technology University, Almaty 050000, Kazakhstan.
Sensors (Basel). 2023 Jul 20;23(14):6554. doi: 10.3390/s23146554.
Deep learning algorithms have a wide range of applications, including cancer diagnosis, face and speech recognition, object recognition, etc. It is critical to protect these models since any changes to them can result in serious losses in a variety of ways. This article proposes the consortium blockchain-enabled conventional neural network (CBCNN), a four-layered paradigm for detecting malicious vehicles. Layer-1 is a convolutional neural network-enabled Internet-of-Things (IoT) model for the vehicle; Layer-2 is a spatial pyramid polling layer for the vehicle; Layer-3 is a fully connected layer for the vehicle; and Layer-4 is a consortium blockchain for the vehicle. The first three layers accurately identify the vehicles, while the final layer prevents any malicious attempts. The primary goal of the four-layered paradigm is to successfully identify malicious vehicles and mitigate the potential risks they pose using multi-label classification. Furthermore, the proposed CBCNN approach is employed to ensure tamper-proof protection against a parameter manipulation attack. The consortium blockchain employs a proof-of-luck mechanism, allowing vehicles to save energy while delivering accurate information about the vehicle's nature to the "vehicle management system." C++ coding is employed to implement the approach, and the ns-3.34 platform is used for simulation. The ns3-ai module is specifically utilized to detect anomalies in the Internet of Vehicles (IoVs). Finally, a comparative analysis is conducted between the proposed CBCNN approach and state-of-the-art methods. The results confirm that the proposed CBCNN approach outperforms competing methods in terms of malicious label detection, average accuracy, loss ratio, and cost reduction.
深度学习算法有广泛的应用，包括癌症诊断、面部和语音识别、目标识别等。保护这些模型至关重要，因为对它们的任何更改都可能以各种方式导致严重损失。本文提出了基于联盟区块链的传统神经网络（CBCNN），这是一种用于检测恶意车辆的四层范式。第一层是用于车辆的启用卷积神经网络的物联网（IoT）模型；第二层是用于车辆的空间金字塔池化层；第三层是用于车辆的全连接层；第四层是用于车辆的联盟区块链。前三层准确识别车辆，而最后一层防止任何恶意企图。这个四层范式的主要目标是使用多标签分类成功识别恶意车辆并减轻它们带来的潜在风险。此外，所提出的CBCNN方法用于确保针对参数操纵攻击的防篡改保护。联盟区块链采用运气证明机制，使车辆在向“车辆管理系统”提供有关车辆性质的准确信息的同时节省能源。采用C++编码来实现该方法，并使用ns-3.34平台进行仿真。ns3-ai模块专门用于检测车联网（IoV）中的异常。最后，对所提出的CBCNN方法与现有方法进行了比较分析。结果证实，所提出的CBCNN方法在恶意标签检测、平均准确率、损失率和成本降低方面优于竞争方法。
