Development and testing of an image transformer for explainable autonomous driving systems

Dong, Jiqian; Chen, Sikai; Miralinaghi, Mohammad; Chen, Tiantian; Labi, Samuel

doi:10.1108/jicv-06-2022-0021

Cited by 21 publications

(3 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Third, XAI can also improve the performance of the AI model via understanding why and how the model works thus the AV's manufacturers and scientists can know exactly what to fine-tune and optimize (e.g., which sensors of the AV are more important). Therefore, XAI will explain the decisions by the AI model in a manner that can be understood by human operators (e.g., safety drivers) [17].…”

Section: Introductionmentioning

confidence: 99%

XAI-ADS: An Explainable Artificial Intelligence Framework for Enhancing Anomaly Detection in Autonomous Driving Systems

Nazat,

Li,

Abdallah

2024

IEEE Access

View full text Add to dashboard Cite

The advent of autonomous driving systems has given rise to pressing cybersecurity issues regarding the vulnerability of vehicular ad hoc networks (VANETs) to potential attacks. This critical security problem necessitates the application of artificial intelligence (AI) models for anomaly detection in VANETs of autonomous vehicles (AVs). However, the lack of explainability of such AI-based anomaly detection models presents challenges. This motivates an emerging research direction of utilizing explainable AI (XAI) techniques to elucidate the behaviors of anomaly detection models in AV networks. In this work, we propose an end-to-end XAI framework to interpret and visualize the anomaly detection classifications made by AI models securing VANETs. We evaluate the framework on two real-world autonomous driving datasets. The framework furnishes both global and local explanations for the black-box AI models using two XAI methods. Moreover, we introduce two novel feature selection techniques to identify the salient features contributing to anomaly detection, derived from the popular SHAP XAI method and the accuracy of six different black-box AI models. We compare our proposed feature selection approaches with six state-of-the-art feature selection techniques (including two wrapper-based feature selection methods), demonstrating superior performance on various evaluation metrics. To generalize the impact of our feature selection methods, we apply three independent classifiers to evaluate our proposed feature selection approaches. The novel feature selection methods effectively distill the most explanatory features, enhancing model interpretability. Finally, we assess the efficiency (how quickly the XAI models can yield explanatory findings) for each of the six black-box AI models we employed on our two datasets, identifying the most efficient model. By furnishing explanations and visualizations of anomaly detection by AI models, our XAI framework can help in enabling trust and transparency for securing vehicular networks.

show abstract

Section: Introductionmentioning

confidence: 99%

XAI-ADS: An Explainable Artificial Intelligence Framework for Enhancing Anomaly Detection in Autonomous Driving Systems

Nazat,

Li,

Abdallah

2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…To address the issue of traditional routing algorithms, DRL is engaged in this study to allow direct acquisition of optimal routing choices from the environment; this boosts the adaptability to changes and scalability potential (S. Chen et al., 2020; Ding et al., 2022; Dong et al., 2022; Ha et al., 2022; Liu et al., 2023; Peng et al., 2021), in a bid to bridge the gap in the literature regarding the use of RL for dynamic rerouting. In addition, in past literature, multiple agent RL has been explored for vehicle path determination (Arokhlo et al., 2011; Dong et al., 2023; Tang et al., 2020; J. Zhao et al., 2022).…”

Section: Introductionmentioning

confidence: 99%

Dynamic urban traffic rerouting with fog‐cloud reinforcement learning

Du,

Chen,

Dong

et al. 2023

Computer aided Civil Eng

Self Cite

View full text Add to dashboard Cite

Dynamic rerouting has been touted as a solution for urban traffic congestion. However, its implementation is stymied by the complexity of urban traffic. To address this, recent studies suggest the efficacy of novel technologies like fog computing and deep reinforcement learning. However, there exist significant challenges in this regard: (1) sorting massive amounts of data associated with large urban networks, (2) large action space that hinders learning efficiency, (3) impairment of rerouting efficacy due to overreliance on regional/local information, and (4) the issue of congestion shifting. To overcome these challenges, this paper presents a novel two‐step approach that integrates graph attention Q network (GAQ) with entropy‐balanced k shortest path (EBkSP) via fog‐cloud information framework to carry out vehicle rerouting in dynamic traffic environments typical of large cities. The first challenge is addressed by GAQ that uses attention mechanism to evaluate traffic and assign region indices based on the relative importance of information. The second challenge is addressed using fog nodes that reduce the action space, and the third is addressed using fog computing combined with cloud computing to facilitate the computation of global optimal information in a centralized planning decentralized execution pattern. The fourth challenge is addressed using EBkSP, which identifies vehicle optimal routes based on route popularity and vehicle priority. After conducting a numerical experiment involving two other cutting‐edge vehicle rerouting models as baselines, it was established that the proposed model delivers results that are 14%–54% superior from perspectives of the mean travel speed and congestion. The superior efficacy of the proposed learning‐based (compared to rule‐based) models is pronounced when rerouting ratios are low. As the rerouting ratio increases, both the learning‐based and rule‐based model outcomes exhibit reduced likelihoods of severe congestion. However, the learning‐based model consistently outperforms the rule‐based model across all scenarios. The proposed model can effectively reroute vehicles under different situations of rerouting ratio and total count of vehicles and can help improve traffic mobility and reduce congestion in urban areas. It can be implemented easily by urban road agencies.

show abstract

“…The design of a lightweight network structure, compared with convolutional neural networks, is more suitable for real-time detection tasks. Therefore, the obtained segmented samples of different categories are fed into the smoky vehicle automatic detection model built on the lightweight MobileNetv3 network [14][15][16] for recognition and classification. In the task of automatic smoky vehicle detection, not only accurate identification of smoky vehicles is required, but also the network inference speed needs to be improved, especially when dealing with a large amount of surveillance video data.…”

Section: Introductionmentioning

confidence: 99%

Automatic Detection Method for Black Smoke Vehicles Considering Motion Shadows

Wang,

Chen,

2023

Sensors

View full text Add to dashboard Cite

Various statistical data indicate that mobile source pollutants have become a significant contributor to atmospheric environmental pollution, with vehicle tailpipe emissions being the primary contributor to these mobile source pollutants. The motion shadow generated by motor vehicles bears a visual resemblance to emitted black smoke, making this study primarily focused on the interference of motion shadows in the detection of black smoke vehicles. Initially, the YOLOv5s model is used to locate moving objects, including motor vehicles, motion shadows, and black smoke emissions. The extracted images of these moving objects are then processed using simple linear iterative clustering to obtain superpixel images of the three categories for model training. Finally, these superpixel images are fed into a lightweight MobileNetv3 network to build a black smoke vehicle detection model for recognition and classification. This study breaks away from the traditional approach of “detection first, then removal” to overcome shadow interference and instead employs a “segmentation-classification” approach, ingeniously addressing the coexistence of motion shadows and black smoke emissions. Experimental results show that the Y-MobileNetv3 model, which takes motion shadows into account, achieves an accuracy rate of 95.17%, a 4.73% improvement compared with the N-MobileNetv3 model (which does not consider motion shadows). Moreover, the average single-image inference time is only 7.3 ms. The superpixel segmentation algorithm effectively clusters similar pixels, facilitating the detection of trace amounts of black smoke emissions from motor vehicles. The Y-MobileNetv3 model not only improves the accuracy of black smoke vehicle recognition but also meets the real-time detection requirements.

show abstract

Development and testing of an image transformer for explainable autonomous driving systems

Cited by 21 publications

References 42 publications

XAI-ADS: An Explainable Artificial Intelligence Framework for Enhancing Anomaly Detection in Autonomous Driving Systems

XAI-ADS: An Explainable Artificial Intelligence Framework for Enhancing Anomaly Detection in Autonomous Driving Systems

Dynamic urban traffic rerouting with fog‐cloud reinforcement learning

Automatic Detection Method for Black Smoke Vehicles Considering Motion Shadows

Contact Info

Product

Resources

About