Application of Nonnegative Matrix Factorization in Intelligent Patrol Inspection of Substations

Non negative matrix factorization (NMF) is a very popular data dimensionality reduction technique in recent years, which can reduce dimensionality while preserving the corresponding spatial relationships and internal structures of the original data. However, the feature variables extracted by traditional NMF algorithms are often not sparse. Therefore, a new sparse constrained non negative matrix factorization (SCNMF) algorithm is proposed. Firstly, based on the alternating constrained least squares method for solving the NMF problem, sparsity constraints are applied to the coefficient matrix H to obtain a more sparse representation of the data. Subsequently, a non negative double singular value decomposition (NDSVD) method was proposed to initialize the SCNMF algorithm to improve its efficiency. Finally, the proposed algorithm is applied to the image processing of a transformer in a certain substation. The experimental results show that the convergence and computational accuracy of the SCNMF algorithm are significantly better than traditional NMF algorithms, and the effectiveness of image processing is also better than the NMF method.