Intelligent Transportation Systems (ITS) demand reliable vehicle identification and overload detection to enable proactive traffic management, enhancing infrastructure durability and road safety. Prior studies have employed unimodal approaches that primarily rely on visual imagery/video or sensor data, which often struggle in real-world environments due to challenges like occlusions, variable lighting, and sensor noise. To address such challenges, we propose VehicleSense, an adaptive, real-time, and multi-modal learning framework that fuses visual and inertial data for robust vehicle identification and overload detection. The framework integrates image-based (ImageSense) and sensor time-series-based (SensorSense) modalities using a trainable neural meta-classifier. We further introduce a late-fusion strategy to integrate predictions from both modalities, enabling dynamic weighting and conflict resolution to improve robustness. This fusion approach enhances classification reliability, ensuring adaptability across diverse environmental conditions and operational scenarios. Comprehensive evaluations on the collected real-world multi-modal dataset demonstrate that VehicleSense significantly outperforms unimodal baselines, achieving substantial gains. In addition, the framework is optimized for edge deployment, enabling scalable, real-time vehicle intelligence in resource-limited settings.

Adversarial robustness in Unsupervised Domain Adaptation (UDA) remains a significant challenge due to noisy pseudo labels and inherent distributional shifts between the clean source and adversarially perturbed target domains. Existing approaches often fail to achieve an optimal trade-off between robustness and accuracy, as pseudo-labels generated by domain-adapted models tend to introduce classification errors under adversarial attacks. In this work, we propose SFT+RL, a two-stage robust UDA framework that integrates Supervised Fine Tuning (SFT) and Reinforcement-inspired Learning (RL) on top of CLIP’s pre-trained visual encoder. In the SFT stage, we adversarially fine-tune a linear classifier using PGD-based perturbations over the labelled source domain while partially unfreezing CLIP’s projection layer. It allows adaptation to adversarial noise while preserving CLIP’s rich semantic priors. We introduce a confidence-guided pseudo-labeling strategy in the RL stage to annotate unlabeled target samples progressively. Pseudo labels are filtered using a decaying confidence threshold to balance quality and coverage, and the model is trained on a composite dataset formed by combining clean source samples with high-confidence target samples. Adversarial training is applied to mixed batches of clean and adversarial examples to enhance cross-domain robustness. Comprehensive evaluations on three benchmark datasets OfficeHome [18], PACS [7], and VisDA [13] demonstrate the effectiveness of our approach. Notably, SFT+RL achieves average improvements of 10.2% in clean accuracy and 15.8% in adversarial robustness across all three datasets, outperforming existing state-of-the-art methods.