To improve the user’s localization estimation in indoor and outdoor environment a novel radiolocalization system using deep learning dedicated to work both in indoor and outdoor environment is proposed. It is based on the radio signatures using radio signals of opportunity from LTE an WiFi networks. The measurements of channel state estimators from LTE network and from WiFi network are taken by using the developed application. The user’s position is calculated with a trained neural network system’s models. Additionally the influence of various number of measurements from LTE and WiFi networks in the input vector on the positioning accuracy was examined. From the results it can be seen that using hybrid deep learning algorithm with a radio signatures method can result in localization error 24.3 m and 1.9 m lower comparing respectively to the GPS system and standalone deep learning algorithm with a radio signatures method in indoor environment. What is more, the combination of LTE and WiFi signals measurement in an input vector results in better indoor and outdoor as well as floor classification accuracy and less positioning error comparing to the input vector consisting measurements from only LTE network or from only WiFi network.

To address the problem that a deep neural network needs a sufficient number of training samples to have a good prediction performance, this paper firstly used the Z-Map algorithm to generate a simulated profile of the milling surface and construct an optical simulation model of surface imaging to supplement the training sample size of the neural network. Then the Deep CORAL model was used to match the textures of the simulated samples and the actual samples across domains to solve the problem that the simulated samples were not in the same domain as the actual milling samples. Experimental results have shown that high texture matching could be achieved between optical simulation images and actual images, laying the foundation for expanding the actual milled workpiece images with the simulation images. The deep convolutional neural model Xception was used to predict the classification of six classes of data sets with the inclusion of simulation images, and the accuracy was improved from 86.48% to 92.79% compared with the model without the inclusion of simulation images. The proposed method solves the problem of the need for a large number of samples for deep neural networks and lays the foundation for similar methods to predict surface roughness for different machining processes.

In recent years, deep learning and especially deep neural networks (DNN) have obtained amazing performance on a variety of problems, in particular in classification or pattern recognition. Among many kinds of DNNs, the convolutional neural networks (CNN) are most commonly used. However, due to their complexity, there are many problems related but not limited to optimizing network parameters, avoiding overfitting and ensuring good generalization abilities. Therefore, a number of methods have been proposed by the researchers to deal with these problems. In this paper, we present the results of applying different, recently developed methods to improve deep neural network training and operating. We decided to focus on the most popular CNN structures, namely on VGG based neural networks: VGG16, VGG11 and proposed by us VGG8. The tests were conducted on a real and very important problem of skin cancer detection. A publicly available dataset of skin lesions was used as a benchmark. We analyzed the influence of applying: dropout, batch normalization, model ensembling, and transfer learning. Moreover, the influence of the type of activation function was checked. In order to increase the objectivity of the results, each of the tested models was trained 6 times and their results were averaged. In addition, in order to mitigate the impact of the selection of learning, test and validation sets, k-fold validation was applied.

The rapid development of grid integration of solar energy in developing countries like India has created vital concerns such as fluctuations and interruptions affecting grid operations. Improving the consistency and accuracy of solar energy forecasts can increase the reliability of the power grid. Although solar energy is available in abundance around the world, it is viewed as an unpredictable source due to uncertain fluctuations in climate conditions. Global horizontal irradiance (GHI) prediction is critical to efficiently manage and forecast the power output of solar power plants. However, developing an accurate GHI forecasting model is challenging due to the variability of weather conditions over time. This research aims to develop and compare univariate LSTM models capable of predicting GHI in a solar power plant in India over the short term. The present study introduces a deep neural network-based (DNN) hybrid model with a combination of convolutional neural network bi-directional long short-term memory (CNN BiLSTM) to predict the one minute interval GHI of a solar power plant located in the southern region of India. The model’s effectiveness was tested using data for the month of January 2023. In addition, the results of the hybrid model were compared to the long short-term memory (LSTM) and BiLSTM deep-learning (DL) models. It has been observed that the proposed hybrid model framework is more accurate compared to the LSTM and BiLSTM architectures. Finally, a GHI prediction tool was developed to understand the trend of the results.

Speech enhancement is fundamental for various real time speech applications and it is a challenging task in the case of a single channel because practically only one data channel is available. We have proposed a supervised single channel speech enhancement algorithm in this paper based on a deep neural network (DNN) and less aggressive Wiener filtering as additional DNN layer. During the training stage the network learns and predicts the magnitude spectrums of the clean and noise signals from input noisy speech acoustic features. Relative spectral transform-perceptual linear prediction (RASTA-PLP) is used in the proposed method to extract the acoustic features at the frame level. Autoregressive moving average (ARMA) filter is applied to smooth the temporal curves of extracted features. The trained network predicts the coefficients to construct a ratio mask based on mean square error (MSE) objective cost function. The less aggressive Wiener filter is placed as an additional layer on the top of a DNN to produce an enhanced magnitude spectrum. Finally, the noisy speech phase is used to reconstruct the enhanced speech. The experimental results demonstrate that the proposed DNN framework with less aggressive Wiener filtering outperforms the competing speech enhancement methods in terms of the speech quality and intelligibility.

This paper proposes a deep neural network (DNN) based method for the purpose of power-ground plane impedance modeling. A composite DNN model, which is a combination of two DNNs is used to predict the Z-parameters of power ground planes from their design parameters. The first DNN predicts the normalized Z-parameters whereas the second DNN predicts the original maximum and minimum values of the nonnormalized Z-parameters. This allows the method to retain a high accuracy when predicting responses that have large variations across designs, as is the case with the Z-parameters of the power-ground planes. We use the adaptive sampling algorithm to generate the training and validation samples for the DNNs. The adaptive sampling algorithm starts with only a few samples, then slowly generates more samples in the non-linear regions within the design parameters space. The level of non-linearity of the regions is determined by a surrogate model which is also trained using the generated samples as well. If the surrogate model has poor prediction accuracy in a region, then the adaptive sampling algorithm will generate more samples in that region. A shallow neural network is used as the surrogate model for non-linearity determination of the regions since it is faster to train and update. Once all the samples have been generated, they will be used to train and validate the composite DNN models. Finally, we present two examples, a square-shaped power ground plane and a squareshaped power ground plane with a hollow square at the center to demonstrate the robustness of the DNN composite models.