A spectrum denoising method based on low-rank recovery dictionary learning with its application to chemical oxygen demand detection

Abstract

In practical spectral detection, the presence of noise degrades the detection performance. In this paper, we propose a new spectrum denoising method combines sparse representation and low-rank learning called low-rank recovery dictionary learning (LRRDL). Firstly, based on the homogeneity of the spectrum, we employ dictionary learning to extract the sparse features in the spectrum. Secondly, the superposition of spectrum is exploited to embed the low-rank property of clean dictionary into dictionary learning to address the problem that dictionary atoms can be contaminated by high levels of noise. Then, we use LRRDL to built a low-rank denoising model. Finally, in the denoising stage, the dictionary model will be used to perform a sparse reconstruction of the noisy spectrum to complete the denoising work. Experiments are conducted on three datasets and the proposed method is compared with several advanced methods. The result shows our method achieves better denoising performance than other methods.