| 摘 要: 针对红外气体泄漏检测中非刚性形变建模计算冗余与弱显著性特征判别力不足的问题,本文提出一种轻量化时空建模与注意力增强网络ISF-CoTRCNN。该方法通过一种基于方向感知的稀疏建模机制,分别设计了定向稀疏时空建模模块与静态-动态上下文协同增强模块,在降低约80%计算量的同时增强了对气体微弱特征的判别能力。在GOD-Video数据集上的实验表明,所提方法以更低的参数量和计算量,取得了更高的检测精度,为气体泄漏实时检测提供了一种高精度、高效率的解决方案。 |
| 关键词: 红外气体检测 气体泄漏检测 时序建模 注意力机制 红外气体数据集(GOD) VSF-RCNN |
|
中图分类号:
文献标识码:
|
|
| A Lightweight Spatiotemporal Modeling and Attention?Enhanced Network for Infrared Gas Leak Detection |
|
lousenjie1, zhuyaodong2, niewenxiao1
|
1.School of Information Science and Engineering(School of Cyber Science and Technology), Zhejiang Sci-Tech University;2.School of Mechanical Engineering, Jiaxing University
|
| Abstract: To address the computational redundancy in modeling non?rigid gas deformation and the weak feature discriminability in infrared videos, this paper proposes a lightweight spatiotemporal and attention?enhanced network, ISF?CoTRCNN. Its core innovation lies in a direction?aware sparse collaborative modeling mechanism, implemented through an ISF module for efficient directional spatiotemporal sampling and a CoT?Gas module for static?dynamic context enhancement with temporal smoothing. Experiments on the GOD?Video dataset and newly added data show that ISF?CoTRCNN achieves higher detection accuracy with significantly lower complexity, offering an accurate and efficient solution for real?time gas leak detection. |
| Keywords: infrared gas detection gas leak detection Temporal Modeling attention mechanism GOD infrared gas dataset VSF-RCNN |
