Fault detection of Substation Power Transformer by Non-contact measurement is important for the safety of machines, instruments, and human beings. To make non-contact measurement as convenient as possible, it is desirable that efficient algorithms based on AE (acoustic emission) discrimination are developed. This paper presents a system for quick and effective fault detection of substation power transformer, based on AE signals collected by non-contact single microphones. In the experiment, collected data were preprocessed in multiple ways and three machine learning algorithms were designed based on classifiers (Convolutional Neural Network (CNN), support vector machine (SVM), and k-nearest neighbors (KNN) algorithm) trained and tested by a tenfold cross-validation technique. After comparison among the designed classifiers, the results show the two-dimensional principal component analysis (2DPCA) preprocess combined with SVM achieved the best comprehensive effectiveness.

Ocean sound speed is an essential foundation for marine scientific research and marine engineering applications. In this article, a model based on a comprehensive optimal back propagation artificial neural network model is developed. The Levenberg–Marquardt algorithm is used to optimize the model, and the momentum term, normalization, and early termination method were used to predict the high precision marine sound speed profile. The sound speed profile was described by five indicators: date, time, latitude, longitude, and depth. The model used data from the CTD observation dataset of scientific investigation over the South China Sea (2009–2012) (108°–120° E, 6°–8° N), which includes comprehensive scientific investigation data from four voyages. The feasibility of modeling the sound speed field in the South China Sea is investigated. The proposed model uses the momentum term, normalization, and early termination in a traditional BP artificial neural network structure and mitigates issues with overtraining and difficulty when determining the BP neural network parameters. With the LM algorithm, a fast-modeling method for the sound field effectively achieves the precision requirement for sound speed prediction. Through the prediction and verification of the data from 2009 to 2012, the newly proposed optimized BP network model is shown to dramatically reduce the training time and improve precision compared to the traditional network model. Results showed that the root mean squared error decreased from 1.7903 m/s to 0.95732 m/s, and the training time decreased from 612.43 s to 4.231 s. Finally, the sound ray tracing simulations confirm that the model meets the accuracy requirements of acoustic sounding and verify the model’s feasibility for the real-time prediction of the vertical sound speed in saltwater bodies.

In deep learning-based local stereo matching methods, larger image patches usually bring better stereo matching accuracy. However, it is unrealistic to increase the size of the image patch size without restriction. Arbitrarily extending the patch size will change the local stereo matching method into the global stereo matching method, and the matching accuracy will be saturated. We simplified the existing Siamese convolutional network by reducing the number of network parameters and propose an efficient CNN based structure, namely adaptive deconvolution-based disparity matching net (ADSM net) by adding deconvolution layers to learn how to enlarge the size of input feature map for the following convolution layers. Experimental results on the KITTI2012 and 2015 datasets demonstrate that the proposed method can achieve a good trade-off between accuracy and complexity.