Domain-level transformation scenario classification based on joint optimization

In order to solve the problem of insufficient accuracy of image classification in the on-site monitoring of power system, this paper proposes an algorithm based on the joint optimization of non-negative matrix factorization (NMF) and convolutional neural network (CNN). The method is to carry out initial feature extraction and dimensionality reduction by NMF, and then input the processed features into CNN for deep learning and classification, and then estimate the model performance by loss function, and continuously optimize the CNN parameters and the base matrix of NMF by using feedback mechanism. The experimental results show that the algorithm achieves a classification accuracy of up to 93.2% in the scenarios of overhead operation, leakage operation, high temperature operation and staff body abnormality, which significantly improves the classification performance and reliability, and provides a highly efficient and reliable solution for the on-site safety monitoring of the power system.