漏水检测信号的稀疏表示研究
Sparse Representation of Water Leakage Detection Signals
为解决压缩感知在漏水检测系统中面临的信号重构效果不理想的问题,通过对比多种漏水信号稀疏表示方法,寻找最适合漏水检测信号的稀疏表示。具体方法包括设计伯努利随机观测矩阵,并采用不同类型的稀疏基对漏水信号进行稀疏表示和重构。在实验中,选用离散余弦变换(DCT)和db4小波变换作为固定稀疏基,K-SVD算法和自编码器(autoencoder)方法作为学习型字典的稀疏表示,并结合正交匹配追踪算法(OMP)、压缩采样匹配追踪算法(CoSaMP)和梯度下降法(GD)进行实验分析。实验结果表明,在小数据量条件下,K-SVD相比DCT和db4小波变换在均方误差(MSE)、峰值信噪比(PSNR)和余弦相似度(cosine similarity)三个指标上均有所提升。而在大数据量的条件下,自编码器的稀疏表示方法表现出了更好的重构性能,其MSE值最低可达到0.000 67,Cosine Similarity值最高可达到0.999 4,且在信号分布密度和波形拟合方面优于K-SVD,展现出更高的重构效率和鲁棒性。
This study aimed to address the unsatisfactory signal reconstruction effect of compressed sensing in water leakage detection systems. Multiple sparse representation methods of water leakage signals were compared to identify the most suitable sparse representation for water leakage detection signals. The methodology involved designing a Bernoulli random observation matrix and applying different types of sparse bases for sparse representation and reconstruction of water leakage signals. In experiments, discrete cosine transform (DCT) and db4 wavelet transform were selected as fixed sparse bases, while the K-singular value decomposition (K-SVD) algorithm and the autoencoder method were employed for sparse representation of learning-type dictionary. Additionally, the orthogonal matching pursuit (OMP) algorithm, the compressive sampling matching pursuit (CoSaMP) algorithm, and the gradient descent (GD) algorithm were also used for analysis. The results indicated that in the case of small data sets, K-SVD outperformed DCT and db4 wavelet transform in terms of mean squared error (MSE), peak signal-to-noise ratio (PSNR), and cosine similarity. For large data sets, the autoencoder-based sparse representation method demonstrated better reconstruction performance, achieving the lowest MSE of 0.000 67 and the highest cosine similarity of 0.999 4. In addition, it was superior to K-SVD in regard to signal distribution density and waveform fitting, showing higher reconstruction efficiency and robustness.
稀疏表示 / 固定稀疏基 / 学习型字典 / 自编码器 / 信号重构
sparse representation / fixed sparse basis / learning-type dictionary / autoencoder / signal reconstruction
| [1] |
胡云峰. 基于压缩感知的信号重构方法研究[D]. 吉林: 东北电力大学, 2019. |
| [2] |
赵琦, 郭改枝. 基于MATLAB的压缩感知重构漏水信息改进算法研究[J]. 内蒙古师范大学学报(自然科学汉文版), 2022, 51(3): 319-324. |
| [3] |
|
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
颜上取, 汤昊, 刘备, |
| [8] |
李家强, 胡张燕, 姚昌华, |
| [9] |
|
| [10] |
|
| [11] |
翟岳. 基于深度学习的漏水信号检测研究[D]. 呼和浩特: 内蒙古师范大学, 2023. |
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
翟岳, 郭改枝. 基于经验模态分解和小波降噪的漏水信号滤波方法[J]. 内蒙古师范大学学报(自然科学汉文版), 2023, 52(3): 269-275. |
| [16] |
王海安, 郭改枝, 张鹏举. 基于ATmega128的连续信号压缩感知实现方法[J]. 内蒙古大学学报(自然科学版), 2023, 54(2): 206-212. |
内蒙古自治区科技计划资助项目“远程控制地下供水管道漏水监测系统的实现”(2020GG0165)
中央引导地方科技发展资助项目“基于‘人工智能+’管道漏水检测系统的实现”(2024ZY0144)
/
| 〈 |
|
〉 |