Scholarlink

<p>Vehicle type recognition is crucial for intelligent transportation systems. However, accurate and effective recognition is compromised by challenges such as varying target scales, target overlap, and interference from complex weather conditions. To address the aforementioned challenges, we propose an adaptive feature-aware network with wavelet frequency learning for vehicle recognition. First, we develop a Wavelet Balance Optimization Module (WBOM) to mitigate information interference caused by various weather conditions through frequency-domain complementarity. Next, to enhance the recognition of multi-scale vehicle types while balancing resource consumption, we devise an Aggregation Feature Pyramid Network (AF-FPN) in the feature fusion stage to enable cross-level information complementation. Meanwhile, these aggregated features are fed into the designed Adaptive Feature-Aware Module (AFAM). Specifically, the AFAM amplifies the weight of key information, which enhances the adaptability of the model to diverse features. Finally, we introduce the SPPELAN structure, which not only decreases the computational complexity of the model but also enhances its capability to capture target cues across different scales. Extensive experimental results on the UA-DETRAC-1 dataset (3.0% mAP improvement), the BDD100k dataset (3.3% mAP improvement), and the BIT Vehicle dataset (1.3% mAP improvement) demonstrate that the proposed method achieves significant performance gains compared with the original model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive feature-aware network with wavelet frequency learning for vehicle recognition

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Yingzhao Xue,
  • Yijin Ning,
  • Luxia Yang,
  • Hongrui Zhang,
  • Yongjie Ma

摘要

Vehicle type recognition is crucial for intelligent transportation systems. However, accurate and effective recognition is compromised by challenges such as varying target scales, target overlap, and interference from complex weather conditions. To address the aforementioned challenges, we propose an adaptive feature-aware network with wavelet frequency learning for vehicle recognition. First, we develop a Wavelet Balance Optimization Module (WBOM) to mitigate information interference caused by various weather conditions through frequency-domain complementarity. Next, to enhance the recognition of multi-scale vehicle types while balancing resource consumption, we devise an Aggregation Feature Pyramid Network (AF-FPN) in the feature fusion stage to enable cross-level information complementation. Meanwhile, these aggregated features are fed into the designed Adaptive Feature-Aware Module (AFAM). Specifically, the AFAM amplifies the weight of key information, which enhances the adaptability of the model to diverse features. Finally, we introduce the SPPELAN structure, which not only decreases the computational complexity of the model but also enhances its capability to capture target cues across different scales. Extensive experimental results on the UA-DETRAC-1 dataset (3.0% mAP improvement), the BDD100k dataset (3.3% mAP improvement), and the BIT Vehicle dataset (1.3% mAP improvement) demonstrate that the proposed method achieves significant performance gains compared with the original model.