Scholarlink

<p>Object detection in snowy weather poses significant challenges due to degraded image quality and obscured features. This paper introduces a unified detection framework that integrates image restoration tasks with object detection to mitigate weather interference. The proposed model consists of an Adaptive Multi-Level Feature Restoration Branch for noise suppression, a Large Kernel Refinement Module for cross-task integration, and a Cross-Scale Feature Integration Module for feature aggregation. A large-scale snowy weather image dataset, srSnow, is constructed to support training and testing. Experimental results demonstrate that our method achieves mean Average Precision improvements of 3.80, 2.10, and 1.86 percentage points on the rSnow, RSOD, and Snowy-weather real-world datasets, respectively, validating its effectiveness in snowy weather object detection. The relevant code is available at <a href="https://github.com/zhang12345688/SnowNet">https://github.com/zhang12345688/SnowNet</a>.</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

Enhanced object detection in snowy scenes via adaptive restoration-assisted learning and large kernel refinement

  • 期刊论文

  • 发表日期：

    2025年12月6日

查看全文
  • Jing Wang,
  • Beilei Zhang,
  • Shuobo Wang,
  • Zhanqiang Huo

摘要

Object detection in snowy weather poses significant challenges due to degraded image quality and obscured features. This paper introduces a unified detection framework that integrates image restoration tasks with object detection to mitigate weather interference. The proposed model consists of an Adaptive Multi-Level Feature Restoration Branch for noise suppression, a Large Kernel Refinement Module for cross-task integration, and a Cross-Scale Feature Integration Module for feature aggregation. A large-scale snowy weather image dataset, srSnow, is constructed to support training and testing. Experimental results demonstrate that our method achieves mean Average Precision improvements of 3.80, 2.10, and 1.86 percentage points on the rSnow, RSOD, and Snowy-weather real-world datasets, respectively, validating its effectiveness in snowy weather object detection. The relevant code is available at https://github.com/zhang12345688/SnowNet.