%0 Journal Article
%T 基于改进SqueezeNet的油气管道泄漏检测算法<br>Oil and Gas Pipeline Leak Detection Technology Based on Improved SqueezeNet
%A 王健
%A 徐迎斌
%A 黄传富
%A 何婷婷
%A 柏俊杰
%J Journal of Image and Signal Processing
%P 271-280
%@ 2325-6745
%D 2024
%I Hans Publishing
%R 10.12677/jisp.2024.133023
%X 我国石油和天然气的运输方式主要是依靠管道进行运输，根据浴盆曲线，目前我国现存的运输管道处于事故高发期，近些年来全国各地油气管道泄漏事故频发。近些年，如何实时高效检测油气管道泄漏成为了研究热点。本文探索一种改进的基于SqueezeNet神经网络的轻量化模型检测算法，通过将SqueezeNet网络中的Fire模块中的扩展层更换为Ghost模块，再引入SE注意力机制形成G-S-SqueezeNet网络。在实验中对管道泄漏的音频数据集进行预处理并转化为梅尔倒谱进行训练，并进行验证。结果表明在模型缩小的情况下，其鲁棒性和准确性明显改善，准确率相比未改进的网络提升了1.58%，模型大小降低了4 MB，检测速度提升了1.1 s。<br />
The transportation mode of oil and natural gas in our country mainly relies on pipelines for transportation. According to the bathtub curve, the existing transportation pipelines in our country are in the period of high occurrence of accidents. In recent years, oil and gas pipeline leakage accidents occur frequently across the country. In recent years, how to detect oil and gas pipeline leakage in real time and efficiently has become a research hotspot. In this paper, we explore an improved lightweight model detection algorithm based on the SqueezeNet neural network, by replacing the extension layer in the Fire module in the SqueezeNet network with Ghost module, and then introducing the SE attention mechanism to form a G-S-SqueezeNet network. In the experiment, the audio data set of pipeline leakage is preprocessed and transformed into MEL cepstrum for training and verification. The results show that when the model is reduced, its robustness and accuracy are significantly improved, the accuracy is increased by 1.58%, the model size is reduced by 4 MB, and the detection speed is increased by 1.1 seconds compared with the unimproved network.
%K 油气管道泄漏，SqueezeNet，Ghost，SE注意力机制<br>Oil and Gas Pipelines
%K SqueezeNet
%K Ghost
%K SE Attention Mechanisms
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=91229