A deep residual neural network for semiconductor defect classification in imbalanced scanning electron microscope datasets

This paper proposes a method to evaluate the degradation stages of stability of the screw with data fusion technology and deep residual neural network. Firstly, the data provided by multi-sensors are fused, and then the time domain images of the signals are input into the deep residual neural network for training and testing. The effectiveness of the proposed method is verified using data sets collected from the degradation test bench of ball screw. The data sets contain massive samples involving 3 degradation stages under 9 working conditions obtained by 3 accelerometers. From the results and comparison, it can be found that the deep residual neural network can automatically extract the features from the original data layer by layer and achieve a better evaluation rate when it is used to evaluate the performance stage of the screw, and the multi-sensor data fusion vibration data can better reflect the degradation stages of the screw's performance. It is showed that this method enhances the intelligence of the evaluation compared with the traditional method.

Data-driven rolling-bearing fault diagnosis methods are mostly based on deep-learning models, and their multilayer nonlinear mapping capability can improve the accuracy of intelligent fault diagnosis. However, problems such as gradient disappearance occur as the number of network layers increases. Moreover, directly taking the raw vibration signals of rolling bearings as the network input results in incomplete feature extraction. In order to efficiently represent the state characteristics of vibration signals in image form and improve the feature learning capability of the network, this paper proposes fault diagnosis model MTF-ResNet based on a Markov transition field and deep residual network. First, the data of raw vibration signals are augmented by using a sliding window. Then, vibration signal samples are converted into two-dimensional images by MTF, which retains the time dependence and frequency structure of time-series signals, and a deep residual neural network is established to perform feature extraction, and identify the severity and location of the bearing faults through image classification. Lastly, experiments were conducted on a bearing dataset to verify the effectiveness and superiority of the MTF-ResNet model. Features learned by the model are visualized by t-SNE, and experimental results indicate that MTF-ResNet showed better average accuracy compared with several widely used diagnostic methods.

Contact-map of a protein sequence dictates the global topology of structural fold. Accurate prediction of the contact-map is thus essential to protein 3D structure prediction, which is particularly useful for the protein sequences that do not have close homology templates in the Protein Data Bank. We developed a new method, ResPRE, to predict residue-level protein contacts using inverse covariance matrix (or precision matrix) of multiple sequence alignments (MSAs) through deep residual convolutional neural network training. The approach was tested on a set of 158 non-homologous proteins collected from the CASP experiments and achieved an average accuracy of 50.6% in the top-L long-range contact prediction with L being the sequence length, which is 11.7% higher than the best of other state-of-the-art approaches ranging from coevolution coupling analysis to deep neural network training. Detailed data analyses show that the major advantage of ResPRE lies at the utilization of precision matrix that helps rule out transitional noises of contact-maps compared with the previously used covariance matrix. Meanwhile, the residual network with parallel shortcut layer connections increases the learning ability of deep neural network training. It was also found that appropriate collection of MSAs can further improve the accuracy of final contact-map predictions. The standalone package and online server of ResPRE are made freely available, which should bring important impact on protein structure and function modeling studies in particular for the distant- and non-homology protein targets. https://zhanglab.ccmb.med.umich.edu/ResPRE and https://github.com/leeyang/ResPRE. Supplementary data are available at Bioinformatics online.

Image acquisition is an essential step in image processing. When the image acquisition is done the image that is generated is subjected to Impulse noise, Gaussian noise etc. We have performed the image denoising on images inflicted with impulse noise. Image denoising is an essential step in all types of image processing. Traditional techniques reduce the noise in the image but it also reduces the quality of the image. Traditional filters like gaussian filter, median filter is analyzed which work in the spatial domain and filters working in the frequency domain are also considered like Butterworth filters, Weiner filter. A Deep residual Neural Network filter is proposed which is compared with the Fuzzy Neural Network denoiser. Their performance is compared on the metrics PSNR and SSIM. The Fuzzy Neural Network system improves the SSIM significantly compared to a deep residual neural network and a comparison is made with traditional image denoising methods. We also compare the performance of the deep residual neural network, Fuzzy Neural Network system and Median denoising algorithm on impulse noise has been compared. The performance of deep neural networks depends on the total number of examples used and the performance can be improved if we have more image pairs.

Automatic modulation classification (AMC) is a signal processing technology used to identify the modulation type of unknown signals without prior information such as modulation parameters for drone communications. In recent years, deep learning (DL) has been widely used in AMC methods due to its powerful feature extraction ability. The significant performance of DL-based AMC methods is highly dependent on large amount of data. However, with the increasingly complex signal environment and the emergence of new signals, several recognition tasks have difficulty obtaining sufficient high-quality signals. To address this problem, we propose an AMC method based on a deep residual neural network with masked modeling (DRMM). Specifically, masked modeling is adopted to improve the performance of a deep neural network with limited signal samples. Both complex-valued and real-valued residual neural networks (ResNet) play an important role in extracting signal features for identification. Several typical experiments are conducted to evaluate our proposed DRMM-based AMC method on the RadioML 2016.10A dataset and a simulated dataset, and comparison experiments with existing AMC methods are also conducted. The simulation results illustrate that our proposed DRMM-based AMC method achieves better performance in the case of limited signal samples with low signal-to-noise ratio (SNR) than other existing methods.

The neuron classification problem is significant for understanding structure-function relationships in computational neuroscience. Advances in recent years have accelerated the speed of data collection, resulting in a large amount of data on the geometric, morphological, physiological, and molecular characteristics of neurons. These data encourage researchers to strive for automated neuron classification through powerful machine learning techniques. This paper extracts a statistical dataset of 43 geometrical features obtained from 116 human neurons, and proposes a neuronal morphology classification approach based on deep residual neural networks with feature scaling. The approach is applied to classify 18 types of human neurons and compares the accuracy of different number of residual block. Then, we also compare the accuracy between the proposed approach and other mainstream machine learning approaches, the classification accuracy of our approach is 100% in the training set and the testing set accuracy is 76.96%. The experimental results show that the deep residual neural network model has better classification accuracy for human neurons.

This paper proposes a novel fault detection and classification method via deep residual convolutional neural network (DRCNN). The DRCNN captures the deep process features represented by convolutional layers from local to global. Unlike traditional methods, this feature representation can extract the deep fault information and learn the latent fault patterns. Besides, a data preprocessing approach is also proposed to transform the shape of original data into the shape available for convolutional neural network. Finally, experiments based on the data set of Tennessee Eastman process (TEP), a chemical industrial process benchmark, show that the proposed method achieves superior fault detection and better classification performance compared with the state‐of‐the‐art methods.

This paper makes a research for the speed control of wind turbine system under multiple operating regimes, which also studied the sleep residual recurrent neural network method in this work. We aim at designing deep residual recurrent neural network robust controllers, which guarantee the existence of the multiple regime system poles in some predefined zone and wind speed precise tracking. Moreover, the feedback gains which guarantee desired speed tracking performance are obtained by solving the Lyapunov stability functions. The results are applied to a directly driven wind energy conversion experiment systems and the numerical experiment, comparing with the existing results, shows the satisfactory performance of the proposed method.

The state-of-art models for speech synthesis and voice conversion are capable of generating synthetic speech that is perceptually indistinguishable from bonafide human speech. These methods represent a threat to the automatic speaker verification (ASV) systems. Additionally, replay attacks where the attacker uses a speaker to replay a previously recorded genuine human speech are also possible. We present our solution for the ASVSpoof2019 competition, which aims to develop countermeasure systems that distinguish between spoofing attacks and genuine speeches. Our model is inspired by the success of residual convolutional networks in many classification tasks. We build three variants of a residual convolutional neural network that accept different feature representations (MFCC, Log-magnitude STFT, and CQCC) of input. We compare the performance achieved by our model variants and the competition baseline models. In the logical access scenario, the fusion of our models has zero t-DCF cost and zero equal error rate (EER), as evaluated on the development set. On the evaluation set, our model fusion improves the t-DCF and EER by 25% compared to the baseline algorithms. Against physical access replay attacks, our model fusion improves the baseline algorithms t-DCF and EER scores by 71% and 75% on the evaluation set, respectively.

As a crucial component of the photovoltaic (PV) power generation system, effective defect detection of PV cells can help to ensure the power generation efficiency and the system safety. The photo-thermal-radiation infrared thermography experimental system has been built, and the infrared images of normal and defective PV cells were collected by an IR camera. An effective convolutional neural network (CNN) based model with the residual architecture is designed here to detect and classify the defects. An offline date augmentation method is performed to overcome the insufficient image dataset in the step of dataset establishment, so as to improve the defects classification capability of the designed CNN model. The defects classification accuracy of the proposed model is obtained as 96.96% with 3-class. In order to verify the defect classification ability of the designed CNN model, other four models were adopted for comparative analysis. The results indicate that the proposed CNN model has advantages in defect classification of PV cells. The completed work can provide meaningful exploration in the industrial application of defect detection of PV cells.

The aim of this paper is to provide a deep learning based method that can solve high-dimensional Fredholm integral equations. A deep residual neural network is constructed at a fixed number of collocation points selected randomly in the integration domain. The loss function of the deep residual neural network is defined as a linear least-square problem using the integral equation at the collocation points in the training set. The training iteration is done for the same set of parameters for different training sets. The numerical experiments show that the deep learning method is efficient with a moderate generalization error at all points. And the computational cost does not suffer from “curse of dimensionality” problem.

As an enabling technology for emerging and future generations of wireless networks, sparse code multiple access (SCMA) offers major improvements in terms of spectral efficiency and massive connectivity. Although the message passing algorithm (MPA) for SCMA decoding at the receiver side can achieve near optimum performance, it entails high computational complexity. In this paper, to address this issue, we propose a novel SCMA decoder based on deep residual neural network (ResNet), wherein the decoder is trained to predict the transmit codewords. In our approach, residual blocks are employed to tackle the problems of accuracy saturation and vanishing gradients with deep learning based decoder, while batch normalization is utilized to enhance the stability and robustness of the decoder. The performance of the proposed ResNet decoder for SCMA is validated by means of simulations over AWGN and Rayleigh fading channels. The results show that besides a much reduced complexity, the proposed decoder leads to improvements in term of bit error rate (BER) over competing deep neural network (DNN) based decoders.

An electrocardiograph (ECG) is widely used in diagnosis and prediction of cardiovascular diseases (CVDs). The traditional ECG classification methods have complex signal processing phases that leads to expensive designs. This paper provides a deep learning (DL) based system that employs the convolutional neural networks (CNNs) for classification of ECG signals present in PhysioNet MIT-BIH Arrhythmia database. The proposed system implements 1-D convolutional deep residual neural network (ResNet) model that performs feature extraction by directly using the input heartbeats. We have used synthetic minority oversampling technique (SMOTE) that process class-imbalance problem in the training dataset and effectively classifies the five heartbeat types in the test dataset. The classifier’s performance is evaluated with ten-fold cross validation (CV) using accuracy, precision, sensitivity, F1-score, and kappa. We have obtained an average accuracy of 98.63%, precision of 92.86%, sensitivity of 92.41%, and specificity of 99.06%. The average F1-score and Kappa obtained were 92.63% and 95.5% respectively. The study shows that proposed ResNet performs well with deep layers compared to other 1-D CNNs.

For people in developing countries, cassava is a major source of calories and carbohydrates. However, Cassava Mosaic Disease (CMD) has become a major cause of concern among farmers in sub-Saharan Africa countries, which rely on cassava for both business and local consumption. The article proposes a novel deep residual convolution neural network (DRNN) for CMD detection in cassava leaf images. With the aid of distinct block processing, we can counterbalance the imbalanced image dataset of the cassava diseases and increase the number of images available for training and testing. Moreover, we adjust low contrast using Gamma correction and decorrelation stretching to enhance the color separation of an image with significant band-to-band correlation. Experimental results demonstrate that using a balanced dataset of images increases the accuracy of classification. The proposed DRNN model outperforms the plain convolutional neural network (PCNN) by a significant margin of 9.25% on the Cassava Disease Dataset from Kaggle.

We have developed an automatic sleep stage classification algorithm based on deep residual neural networks and raw polysomnogram signals. Briefly, the raw data is passed through 50 convolutional layers before subsequent classification into one of five sleep stages. Three model configurations were trained on 1850 polysomnogram recordings and subsequently tested on 230 independent recordings. Our best performing model yielded an accuracy of 84.1% and a Cohen's kappa of 0.746, improving on previous reported results by other groups also using only raw polysomnogram data. Most errors were made on non-REM stage 1 and 3 decisions, errors likely resulting from the definition of these stages. Further testing on independent cohorts is needed to verify performance for clinical use.