一种轻量级SBLL-YOLO模型通信线缆外部入侵检测技术OA
A lightweight SBLL-YOLO model for external intrusion detection of communication cables
通信线缆是通信网络传输的交通纽带,外部损坏会导致通信网络瘫痪,保障其不受外界入侵引发通信中断是网络巡检工作的主要部分.为了进一步提升通信线缆外部入侵的检测效率、提前预警入侵行为、降低网络故障风险.提出一种轻量化高精度的SBLL-YOLO模型,首先使用分割交互混合模型(Split-interactive Fusion SIF)在主干网络中将输入信息分割为通道信息及空间信息,然后通过卷积变换进行融合,避免了空间位置信息和判别语义信息特征的退化;其次在颈部网络中引入双向特征金字塔网络(BiFPN)强化多尺度特征融合与提取能力,改进了多尺度目标的自动化感知能力;再次提出一种轻量化可变型卷积头(LDCN-head)以提升多尺度目标感知能力,设计了最小点距离交并比损失函数LMDIoU,通过优化预测框与真实框的最小损失值提升边界框回归精度.在实际入侵数据集上的实验表明,相比YO-LOv5/8/11模型,对于小目标入侵检测,SBLL-YOLO以更少的参数和更低的计算需求实现了优越的检测性能,检测效果更优,可为通信网络线缆外部入侵监测提供高效可靠的技术解决方案.
Communication cables serve as the critical transmission links in communication networks.External damage can lead to network paralysis,and protecting them from external intrusions that may cause communication interruptions is a primary task in network inspection.To further improve the detection efficiency of external intrusion on communication cables,provide early warning of intrusion behaviors,and reduce the risk of network failures,this paper proposes a lightweight and high-precision SBLL-YOLO model.First,a Split-Interactive Fusion(SIF)module is introduced in the backbone network to split the input information into channel information and spatial information,which are then fused via convolutional transformation,thereby avoiding the degradation of spatial location information and discriminative semantic features.Second,a Bidirectional Feature Pyramid Network(BiFPN)is incorporated into the neck network to enhance multi-scale feature fusion and extraction capabilities,improving the model's adaptive perception of multi-scale objects.Third,a Lightweight Deformable Convolution Head(LDCN-Head)is proposed to further boost multi-scale object perception,and a loss function named LMDIoU(Minimum Point Distance based IoU)is designed to improve bounding box regression accuracy by optimizing the mini-mum loss between predicted boxes and ground-truth boxes.Experiments on a real-world intrusion dataset demonstrate that,compared with YOLOv5/8/10 models,SBLL-YOLO achieves superior detection performance for small-target intrusion detection with fewer parame-ters and lower computational requirements.It provides an efficient and reliable technical solution for external intrusion monitoring of com-munication network cables.
郤伟;高艳;邓翠艳;王云花
晋中信息学院,山西 晋中 030800晋中信息学院,山西 晋中 030800晋中信息学院,山西 晋中 030800晋中信息学院,山西 晋中 030800
信息技术与安全科学
通信网络线缆外部入侵检测分割交互混合模型双向特征金字塔网络YOLO交并比
Communication network cablesExternal intrusion detectionSIFBiFPNYOLOIoU
《通信与信息技术》 2026 (5)
35-40,6
2025年山西省高等学校教学改革创新项目"产教融合驱动下大数据专业'能力递进式'人才培养的实践教学体系改革研究"(项目编号:J20250332)
评论