Enhancing Vehicle Detection under Adverse Weather Conditions with Contrastive Learning

2025-09-26

AI Summary

To address poor visibility and domain shift in drone-based vehicle detection under Nordic snowy conditions, this paper proposes the sideload-CL-adaptation framework. It first performs self-supervised pretraining of a CNN feature extractor via contrastive learning on unlabeled snowy images to harness their latent discriminative information. This pretrained extractor is then sideloaded—i.e., integrated without weight updates—onto a frozen lightweight YOLOv11n backbone, and its features are fused with multi-granularity representations to enable effective knowledge transfer while preserving inference efficiency. Crucially, the detection head remains entirely unfine-tuned. Evaluated on the NVD dataset, the method achieves a 3.8-percentage-point improvement in mAP₅₀ (reaching 9.5%), demonstrating strong generalization under low-resource, high-interference conditions and validating its robustness for adverse-weather drone perception.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Large pretrained models with web data
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Aside from common challenges in remote sensing like small, sparse targets and computation cost limitations, detecting vehicles from UAV images in the Nordic regions faces strong visibility challenges and domain shifts caused by diverse levels of snow coverage. Although annotated data are expensive, unannotated data is cheaper to obtain by simply flying the drones. In this work, we proposed a sideload-CL-adaptation framework that enables the use of unannotated data to improve vehicle detection using lightweight models. Specifically, we propose to train a CNN-based representation extractor through contrastive learning on the unannotated data in the pretraining stage, and then sideload it to a frozen YOLO11n backbone in the fine-tuning stage. To find a robust sideload-CL-adaptation, we conducted extensive experiments to compare various fusion methods and granularity. Our proposed sideload-CL-adaptation model improves the detection performance by 3.8% to 9.5% in terms of mAP50 on the NVD dataset.

Problem

• Improving vehicle detection in adverse weather conditions
• Addressing domain shifts from varying snow coverage
• Enhancing detection with unannotated data using contrastive learning

Innovation

• Uses contrastive learning on unannotated UAV data
• Sideloads pretrained features to frozen YOLO backbone
• Compares fusion methods for robust adaptation