|
| 面向无人机平台的轻量化林火热红外图像分割方法 |
| A Lightweight Thermal Infrared Forest Fire Segmentation Method for UAVs |
| 投稿时间：2026-01-21 修订日期：2026-03-27 |
| 查看/发表评论 下载PDF阅读器 |
| DOI： |
| 基金项目: |
|
| 中文关键词: 林火检测 轻量化网络 YOLOv11-Seg 实例分割 热红外成像 |
| 英文关键词:forest fire detection lightweight network YOLOv11-Seg instance segmentation thermal infrared imaging. |
| 摘要点击次数: 145 |
| 全文下载次数: 0 |
| 中文摘要: |
| 针对无人机热红外林火检测中模型复杂、算力受限且实时性要求高的问题，提出一种面向无人机平台的轻量化林火热红外图像分割算法LFYOLO。该算法在YOLOv11-Seg基础上，引入特征互补映射模块(FCM)、基于动态上采样的多尺度特征融合网络(DyNeck)以及高效分割头(SegmentEfficient)，在显著降低模型参数量和计算复杂度的同时保持较高的检测与分割精度。在自建热红外林火数据集上的实验结果表明，LFYOLO的mAP@0.5达到84.0%，Mask mAP@0.5达到81.6%，参数量仅为0.83M，相较YOLOv11-Seg模型减少71%。实验结果表明，LFYOLO在低算力条件下能够实现稳定的检测与分割性能，满足无人机边缘计算平台的实时监测需求。该研究为无人机热红外火灾检测提供了一种高效、可部署的解决方案，也为热红外图像目标检测与分割技术的发展提供了新的思路。 |
| 英文摘要: |
| To address the challenges of complex models, limited computational resources, and high real-time requirements in UAV-based thermal infrared forest fire detection, a lightweight thermal infrared image segmentation algorithm for UAV platforms, named LFYOLO, is proposed. Based on YOLOv11-Seg, this algorithm introduces a Feature Complementary Mapping (FCM) module, a multi-scale feature fusion network based on dynamic upsampling (DyNeck), and an efficient segmentation head (SegmentEfficient). These improvements significantly reduce the model's parameter count and computational complexity while maintaining high detection and segmentation accuracy. Experimental results on a custom thermal infrared forest fire dataset show that LFYOLO achieves an mAP@0.5 of 84.0% and a Mask mAP@0.5 of 81.6%. The parameter count is only 0.83M, representing a 71% reduction compared to the baseline YOLOv11-Seg model. Furthermore, evaluations demonstrate that LFYOLO delivers stable detection and segmentation performance under low-compute conditions, satisfying the real-time monitoring needs of UAV edge computing platforms. This research provides an efficient, deployable solution for UAV thermal infrared fire detection and offers new perspectives for the development of thermal infrared image object detection and segmentation technologies. |
| 关闭 |
|
|
|
