CNN Network Research Articles

A Data Filling Methodology for Time Series Based on CNN and (Bi)LSTM Neural Networks

A Data Filling Methodology for Time Series Based on CNN and (Bi)LSTM Neural Networks

Boosting CNN Network Performance for Face Recognition in an Authentication System

Boosting CNN Network Performance for Face Recognition in an Authentication System

The Practice and exploration of deep learning algorithm in the creation and realization of intangible cultural heritage animation

Abstract The rapid development of modern technology and civilization has made the survival of non-heritage culture more and more serious, and the protection and inheritance of intangible cultural heritage is a heavy task and a long way to go. This paper takes the feasibility of non-heritage animation creation as an entry point, analyzes the ideological mechanism in the process of non-heritage animation creation, and explores the economic realization brought by deep learning technology assisting non-heritage animation creation. For the lens scene switching in the process of non-heritage animation creation, this paper utilizes the CNN network for the initial positioning of the lens boundary. It establishes the tangent detection model of a non-heritage animation lens by combining it with the 3D-CNN network. To understand the diversity of non-heritage animation creation styles, this paper establishes a model for style migration of non-heritage cultural images based on the VGG-Net network and conducts experimental investigations. The results show that when the hyperparameter value of the model is set to λ = 3, γ = 1.2, the model retains only 5.19% of the candidate boundary frames, and the accuracy of the detection of the non-heritage animation creation tangent shots is 0.939. The total loss value in the process of style migration fluctuates around 0.005, and the subjective evaluation score of the images generated by the style migration network is 4.82. The deep learning algorithm that promotes the creation of non-heritage animation can expand the content and characteristics of non-heritage animation creation and can also realize the economic realization of non-heritage animation.

Dynamic prediction of steam consumption in beer production process based on Attention Mechanism CNN-BiLSTM

During the normal production process of the brewery, the steam consumption can be accurately predicted, and the boiler steam output can be planned to achieve a balance between steam production and use. In order to improve the accuracy of steam consumption prediction, this paper proposes a dynamic prediction method of steam consumption based on attention mechanism-convolution-bidirectional long short-term memory neural network (CNN-BiLSTM). In this paper, the real-time monitoring data of the brewery energy management system is selected as the experimental data for analysis, and the CNN-BiLSTM network model based on attention mechanism is compared with the results of CNN net-work, LSTM network, BiLSTM network and CNN-BiLSTM network prediction. The experimental results show that the mean absolute error, root mean square error and R-Square of the model are 0.0329, 0.0449 and 97.5%, which are better than the other four models, and can predict the steam consumption of brewery more accurately.

A Study on the Cultivation of Multicultural Literacy in College English Teaching Based on the Construction of Cognitive Maps

Abstract Using the CNN network model and attention mechanism, this paper completes the construction of the cognitive graph pattern layer based on Protégé programming, crawls the relevant data from the dimensions of language ability, learning ability, thinking, and cultural character of multicultural literacy in English teaching, extracts the entity-relationship of multicultural literacy in English teaching, stores it in the Neo4j corpus, and ultimately completes the construction of the cognitive graph for the cultivation of multicultural literacy in English teaching. Following simulated experiments to evaluate the algorithm’s performance, a case study was carried out to foster English multicultural literacy from the standpoint of the cognitive graph. The information demonstrates that the convolution technique and an attention mechanism work well together to increase relationship extraction efficiency. Further evidence that the cultivation of multicultural literacy in English teaching focuses on language ability, learning ability, thinking character, and cultural character in order to quickly improve students’ English multicultural literacy comes from the keywords with high word frequency in English multicultural literacy: “language ability (388, 0.899)”, “learning ability (185, 0.449)”, “thinking character” (114, 0.38), and “cultural character” (105, 0.335).

Boosting CNN network performance for face recognition in an authentication system

Boosting CNN network performance for face recognition in an authentication system

Effectiveness of Pre-Trained CNN Networks for Detecting Abnormal Activities in Online Exams

Effectiveness of Pre-Trained CNN Networks for Detecting Abnormal Activities in Online Exams

Incorporating Kernels into Convolutional Neural Networks for Enhanced Feature Extraction

The simulation was used to evaluate the method of kernel K , Neural network NN, Convolution Neural network CNN bys using (MINIST) data set. The accuracy of the method was tested and compared with the convolutional neural network as well as with the kernel function for the same input data (training and testing). The results of simulation showed that there is a high accuracy of the method, and at the same time there is a decreasing loss over the epochs, which indicates the. We note high smooth by method for recognize among features.

Research on the method of predicting the trend of criminal activities based on time series analysis from the perspective of criminal procedure law

Abstract Criminal activity has always been a problem that seriously affects people’s sense of security and well-being in life, and it is also the focus of public security departments in terms of security prevention and control. This study aims to use scientific methods to analyze the time series of existing historical data related to criminal activities and to advance inference or determination of the future trend of criminal activities. In this paper, after mining the spatiotemporal pattern of criminal activities using the centered moving average method, the spatiotemporal cycle characteristics and time distribution characteristics of criminal activities are analyzed. Based on the conclusions drawn, an improved LSTNet algorithm is proposed by combining deep learning spatio-temporal network, convolutional neural network CNN, recurrent neural network RNN, autoregressive AR, and specially designed Skip-RNN, and a GRU-gated recurrent unit structure is used to realize the optimization of the prediction accuracy of the criminal activities, to solve the problems related to the spatio-temporal dependencies, and to establish an improved ST- Res Net crime prediction model. Finally, the prediction performance of the model is examined. The RMSE, MSE, and MAPE values of this paper’s prediction model are 0.88, 5.12, and 5.12%, respectively, which are better than other models. The experiments demonstrate that the prediction model presented in this paper can enhance the accuracy of criminal activity prediction to a certain extent and can be employed in the prevention and management of criminal activities.

Processing and recognition of EMG signals through CNN networks for the control of electric vehicles

The increase in autonomous driving technologies, as well as biometrics using biosignals from vehicle drivers, provide information that can be used for the development of personalized biosecurity and driving systems for each user. Currently, studies are being carried out on the extraction and classification of driver characteristics with great precision, to generate intelligent systems that are auxiliary and that help to safeguard the integrity of people while driving vehicles. This work presents the recognition of 5 hand gestures to control the driving actions of an electric vehicle using the EMG signals from the MYOTM bracelet, these signals have also been used to detect users and thus allow the use only of the people registered in the application. To perform gesture recognition, a convolutional neural network was trained and implemented for the classification of actions. Finally, a cross-validation was carried out to validate the reliability of the proposed system, obtaining 99.2% accuracy during the classification.

COVID-19 classification based on a deep learning and machine learning fusion technique using chest CT images

Coronavirus disease (COVID-19), impacted by SARS-CoV-2, is one of the greatest challenges of the twenty-first century. COVID-19 broke out in the world over the last 2 years and has caused many injuries and killed persons. Computer-aided diagnosis has become a necessary tool to prevent the spreading of this virus. Detecting COVID-19 at an early stage is essential to reduce the mortality risk of patients. Researchers seek to find rapid solutions based on techniques of Machine Learning and Deep Learning. In this paper, we introduced a hybrid model for COVID-19 detection based on machine learning and deep learning models. We used 10 different deep CNN network models to extract features from CT images. We extract features from different layers in each network and find the optimum layer that gives the best-extracted features for each CNN network. Then, for classifying these features, we used five different classifiers based on machine learning. The dataset consists of 2481 CT images divided into COVID-19 and non-COVID-19 categories. Three folds are extracted with a different size between testing and training. Through experiments, we define the best layer for all used CNN networks, the best network, and the best-used classifier. The measured performance shows the superiority of the proposed system over the literature with a highest accuracy of 99.39%. Our models are tested with the three folds that gained maximum average accuracy. The result is 98.69%.

The effect of frame size on speech emotion recognition

Speech emotion recognition task, as well as most audio recognition machine learning tasks, uses the so-called framing. This is the process of dividing the original audio signal into frames of a certain size, each of which is processed separately. This article presents a comparison of the effect of frame size on the emotion recognition result using a CNN network as an example. For the experiments, the CREMA-D dataset was used with the augmentations using noise adding, time stretching, and pitch shifting. We managed to achieve a recognition accuracy of 98.8% using dynamic frame size.

A deep learning model for depression detection based on MFCC and CNN generated spectrogram features

Depression is one of the leading forms of mental health issues encountered by individuals of diverse age groups today worldwide. Like any other mental health concerns, depression too poses diagnostic challenges for medical practitioners and clinical experts, given obvious social reservations and lack of awareness and acceptance in the society. Since long researchers have been looking for methods to identify symptoms of depression among individuals from their speech and responses, by utilizing automation systems and computers. In this paper, we propose an audio based depression detection method, which relies on neural networks for audio spectrogram based feature extraction as well as classification between speech/response patterns of depressed vs. non-depressed persons. We adopt a multi-modal approach in our work, by combining Mel-Frequency Cepstral Coefficients (MFCC) features, as well as Spectrogram features extracted from an audio file, by a novel CNN network. Our CNN model demonstrates optimized residual blocks and the “glorot uniform” kernel initializer. The proposed method’s performance is assessed in both multi-modal and multi-feature trials. We show our results on standard benchmark datasets DAIC-WOZ and MODMA, which provide repositories of questionnaire and patient responses, relevant in identification of depressive symptoms. We have also tested our model on standard emotion recognition audio dataset, RAVDESS. The proposed model achieves detection accuracy of over 90% in DAIC-WOZ and MODMA, and over 85% in RAVDESS, which is proven to surpass the present state-of-the-art.

RMPCT-Net: a multi-channel parallel CNN and transformer network model applied to HAR using FMCW radar

RMPCT-Net: a multi-channel parallel CNN and transformer network model applied to HAR using FMCW radar

Multi-Modal Ensemble Deep Learning in Head and Neck Cancer HPV Sub-Typing.

Oropharyngeal Squamous Cell Carcinoma (OPSCC) is one of the common forms of heterogeneity in head and neck cancer. Infection with human papillomavirus (HPV) has been identified as a major risk factor for OPSCC. Therefore, differentiating the HPV-positive and negative cases in OPSCC patients is an essential diagnostic factor influencing future treatment decisions. In this study, we investigated the accuracy of a deep learning-based method for image interpretation and automatically detected the HPV status of OPSCC in routinely acquired Computed Tomography (CT) and Positron Emission Tomography (PET) images. We introduce a 3D CNN-based multi-modal feature fusion architecture for HPV status prediction in primary tumor lesions. The architecture is composed of an ensemble of CNN networks and merges image features in a softmax classification layer. The pipeline separately learns the intensity, contrast variation, shape, texture heterogeneity, and metabolic assessment from CT and PET tumor volume regions and fuses those multi-modal features for final HPV status classification. The precision, recall, and AUC scores of the proposed method are computed, and the results are compared with other existing models. The experimental results demonstrate that the multi-modal ensemble model with soft voting outperformed single-modality PET/CT, with an AUC of 0.76 and F1 score of 0.746 on publicly available TCGA and MAASTRO datasets. In the MAASTRO dataset, our model achieved an AUC score of 0.74 over primary tumor volumes of interest (VOIs). In the future, more extensive cohort validation may suffice for better diagnostic accuracy and provide preliminary assessment before the biopsy.

A model for remaining useful life prediction of rolling bearings based on the IBA-FELM algorithm

Rolling bearings are essential supporting components for most rotating machinery and are commonly placed at great risk of sudden failure. Accurate prediction of the remaining service life of rolling bearings is essential for ensuring reliable operation and establishing an effective maintenance strategy. Focusing on the extreme learning machine (ELM) methodology, an innovative predictive model with error feedback neuron integration is established to eliminate the deficiency in model generalization capability. To further improve the predictive accuracy, an improved bat algorithm (IBA) is introduced into the FELM model, in which the Levy flight and frequency influence factor are embedded into the traditional BA algorithm to enhance the parameter searching ability. Inverse hyperbolic function-based statistical indicators are proposed and verified by comparing with the classical RMS curve of full-life data, whose cosine similarity and correlation coefficient both exceed 0.95. Two sets of accelerated life experiments were selected to validate the effectiveness of the proposed IBA-FELM model. The results show that the integrated model can obtain high prediction accuracy and satisfactorily fit the real-life data. The maximal prediction error can be reduced from 1.57 to 0.0401 for experimental Case 1, and from 0.7375 to 0.1492 for Case 2. Compared with the other machine learning models, such as SVR, CNN, and LSTM networks, the IBA-FELM model also presents stronger optimization ability, higher generalization performance, and operation stability.

A surface electromyography based hand gesture recognition framework leveraging variational mode decomposition technique and deep learning classifier

Deciphering and classifying surface electromyography (sEMG) signals is highly essential in rehabilitation robotics, myoelectric prosthetic control, sign languages and human–computer-interface. Researchers have generally focused on utilizing multi-channel sEMG signals for gesture classification. However, in case of amputees, the residual muscles are limited and the gesture classification needs to be reframed with minimum sEMG channels. Moreover, another key problem in the gesture classification system is the nonlinear and non-stationary nature of sEMG signals, leading to poor generalization ability. Hence, to address the aforementioned problems, this study puts forward a novel deep learning classifier framework which leverages the potentials of variational mode decomposition (VMD) technique and a hybrid convolutional neural network–long short term memory (CNN–LSTM) classifier model to recognize the hand gestures from single channel sEMG signals. Collecting the sEMG signals from forearm muscles of 25 intact subjects for ten functional and grasping actions, this work implements a VMD technique to identify the prominent frequency modes in the sEMG signals. From the decomposed modes of sEMG signals, the prominent intrinsic mode functions (IMFs) are extracted through spectral analysis to minimize the computation burden on the hybrid classifier model. Furthermore, as the muscle contractions result in substantial temporal dependencies, this work exploits the potentials of CNN and LSTM networks and extracts the spatiotemporal features of the sEMG signals for various hand gestures. The experimental results corroborate that the proposed classifier framework can achieve an average classification accuracy of 98.04% and provide 3% improvement in the classification accuracy compared to conventional CNN classifier.

A Method for Improving the Pronunciation Quality of Vocal Music Students Based on Big Data Technology

With the development of internet technology, big data has been used to evaluate the singing and pronunciation quality of vocal students. However, current methods have several problems such as poor information fusion efficiency, low algorithm robustness, and low recognition accuracy under low signal-to-noise ratio. To address these issues, this article proposes a new method for evaluating sound quality based on one-dimensional convolutional neural networks. It uses sound preprocessing, BP neural networks, wavelet neural networks, and one-dimensional CNNs to improve pronunciation quality. The proposed 1D CNN network is more suitable for one-dimensional sound signals and can effectively solve problems such as feature information fusion, pitch period detection, and network construction. It can evaluate singing art sound quality with minimum errors, good robustness, and strong portability. This method can be used for the evaluation and diagnosis of voice diseases, helping to improve students' professional abilities.

Recognition of edible mushrooms based on different image classification methods

The mushroom is one of the most common foods in nature, but there are many different types and many of them are deadly. The aim of this paper is to use computer vision methods to be able to provide a range of assistance to people on safari in identifying mushrooms and the future development of mushroom poisoning and non-poisoning identification. Considering the superior performance of convolutional neural networks in image recognition and classification, the advantages of multi-layer perceptrons and deep structure make it possible to achieve high recognition accuracy, three different CNN network models, GoogLeNet ResNet and SENet were chosen. They have been used to differentiate between toxic and non-poisonous mushrooms on 11,000 mushrooms. The tests showed that SENet had the greatest results, with a recognition accuracy of over 90%. With a recognition accuracy of around 65% each, GoogLeNet and ResNet placed second and third, respectively.

Integrating machine learning in electronic health passport based on WHO study and healthcare resources

Healthcare's evolution amid rapid scientific progress demands the efficient management of extensive data. This has given rise to electronic health passports, which play a crucial role in modern healthcare by providing secure access to comprehensive medical histories. This study, exemplified by diabetes, addresses the mounting global prevalence of this disease, which saw cases soar from 108 million in 1980 to 422 million by 2014, resulting in a 5 % rise in diabetes-related premature mortality from 2000 to 2016. Disease prediction based on patient attributes holds potential to optimize healthcare resource allocation. The research aim is to identify new hidden dependencies and parameters in the diagnosis of diabetes, utilizing machine learning techniques. The first stage of this involves a thorough comparison and optimization of various modeling techniques, from Decision Trees (e.g., Random Forest, Adaboost) to advanced Deep Learning models like CNN, RNN, and LSTM networks, aiming to determine the most effective implementation method. Leveraging a subset of the MIMIC-III Critical Care Database, a comprehensive de-identified health-related dataset, this research provides insights into addressing the escalating diabetes challenge. The study revealed that the Gradient Boosting Machine was the most effective for making diagnoses. In the realm of deep learning, the Recurrent Neural Network (RNN) demonstrated the best results. This work aligns with the broader context of utilizing electronic health passports to enhance healthcare efficiency and decision-making in an era characterized by abundant health data. It is essential to emphasize that this work represents the initial phase of a broader research trajectory, and its continuation will be expounded upon in forthcoming publications. Datasethttps://doi.org/10.13026/C2XW26. Dataset licensePhysioNet Credentialed Health Data Use Agreement 1.5.0.