To address the problems of limited information expression, imbalance, and dynamic spatio-temporal characteristics of accident data, an accident prediction model fusing heterogeneous traffic situations was proposed. In which, the semantic enhancement was completed by the spatio-temporal state aggregation module through traffic events and weather features representing dynamic traffic situations, and the historical multi-period spatio-temporal states of four types of regions (single region, adjacent region, similar region, and global region) were aggregated; the dynamic local and global spatio-temporal characteristics of accident data were captured by the spatio-temporal relation capture module from both micro- and macro-perspectives; and the multi-region and multi-angle spatio-temporal states were further fused by the spatio-temporal data fusion module, and the accident prediction task in the next period was realized. Experimental results on five city datasets of US-Accident demonstrate that the average F1-scores of the proposed model for accident, non-accident, and weighted average samples are 85.6%, 86.4%, and 86.6% respectively, which are improved by 14.4%, 5.6%, and 9.3% in the three metrics compared to the traditional Feedforward Neural Network (FNN), indicating that the proposed model can effectively suppresses the influence of accident data imbalance on experimental results. Constructing an efficient accident prediction model helps to analyze the safety situation of road traffic, reduce the occurrence of traffic accidents and improve the traffic safety.
In the face of adversarial example attack, deep neural networks are vulnerable. These adversarial examples result in the misclassification of deep neural networks by adding human-imperceptible perturbations on the original images, which brings a security threat to deep neural networks. Therefore, before the deployment of deep neural networks, the adversarial attack is an important method to evaluate the robustness of models. However, under the black-box setting, the attack success rates of adversarial examples need to be improved, that is, the transferability of adversarial examples need to be increased. To address this issue, an adversarial example method based on image flipping transform, namely FT-MI-FGSM (Flipping Transformation Momentum Iterative Fast Gradient Sign Method), was proposed. Firstly, from the perspective of data augmentation, in each iteration of the adversarial example generation process, the original input image was flipped randomly. Then, the gradient of the transformed images was calculated. Finally, the adversarial examples were generated based on this gradient, so as to alleviate the overfitting in the process of adversarial example generation and to improve the transferability of adversarial examples. In addition, the method of attacking ensemble models was used to further enhance the transferability of adversarial examples. Extensive experiments on ImageNet dataset demonstrated the effectiveness of the proposed algorithm. Compared with I-FGSM (Iterative Fast Gradient Sign Method) and MI-FGSM (Momentum I-FGSM), the average black-box attack success rate of FT-MI-FGSM on the adversarially training networks is improved by 26.0 and 8.4 percentage points under the attacking ensemble model setting, respectively.