基于RT-DETR的轻量化铁路侵限异物检测模型研究
Lightweight Railway Intrusion Obstacle Detection Model Based on RT-DETR
针对铁路侵限异物检测模型参数量大、计算复杂而难以嵌入资源受限的铁路边侧设备,导致实时预警能力不足的问题,提出一种基于RT-DETR的轻量化模型。该模型从特征提取、特征融合和计算压缩3个层面进行优化:在特征提取阶段引入空间状态建模机制,通过线性复杂度与硬件感知设计增强关键特征的捕捉能力;在特征融合阶段采用分组混洗卷积替换标准卷积,降低参数量与计算负担;在计算压缩阶段引入哈尔小波下采样模块,对特征图进行高效压缩,显著减少解码器的计算开销。实验结果表明,所提模型在仅占用25.995 MB参数和71.411 G FLOPs计算量的条件下,在多项复杂度指标上均优于现有先进模型,同时保持高精度检测能力,满足了铁路侵限检测高精度、低延迟的核心业务需求。
The railway intrusion obstacle detection models, due to their large parameter size and high computational complexity, are difficult to deploy on resource-constrained railway trackside equipment, resulting in insufficient real-time warning capability. In response to this problem, a lightweight model based on RT-DETR was proposed. The optimization of this model was carried out from three aspects: feature extraction, feature fusion, and computational compression. In the feature extraction stage, a state-space modeling mechanism was introduced to enhance the ability to capture key features through linear complexity and hardware-aware design. In the feature fusion stage, grouped shuffle convolution was adopted to replace standard convolution, reducing parameter count and computational burden. In the computational compression stage, a Haar wavelet downsampling module was incorporated to efficiently compress feature maps, significantly lowering the computational overhead of the decoder. Experimental results demonstrate that the proposed model, with only 25.995 MB parameters and 71.411 G FLOPs, outperforms the existing state-of-the-art models on multiple complexity metrics while maintaining high detection accuracy, thus satisfying the core business requirements of high-precision and low-latency railway intrusion detection.
| [1] |
|
| [2] |
杨 文,胡 昊,李凌志, |
| [3] |
栗文韬,徐成伟,张万鹏, |
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
| [22] |
|
| [23] |
|
| [24] |
|
| [25] |
|
| [26] |
|
| [27] |
|
| [28] |
|
| [29] |
|
| [30] |
|
/
| 〈 |
|
〉 |