Emotions are extensive psycho-physiological processes associated with various physical changes within and outside the body. In addition to being vital for human–machine interfaces, they are crucial for human–human interaction. However, when people unintentionally or purposefully hide their feelings, these indicators may be unhelpful since they cannot convey underlying emotions. Since emotions are typically uncontrollable, research has shown that examining brain activity and physiological signals can result in more accurate emotion detection. The effectiveness and challenges in identifying emotions from various media, utilizing brain activity and other physiological signals, are explored and discussed in this research. The purpose of the suggested multimodal emotion recognition experiment using the ASCERTAIN dataset is to evaluate our method’s effectiveness. The suggested model detects emotions by using the spectrogram of the EEG, ECG, and GSR signals as input. Our experimental findings show that our proposed CNNBiL. (1-dConvolutionalNeuralNetwork + Bidirectional-LongShortTermMemory) The approach has a recognition accuracy of 82.8
The identification of network signals is one of the important factors that affect the communication processing and control effect, but anomalies and errors often occur in applications. With the development of artificial intelligence technology, this paper automatically identifies computer network communication based on a machine learning model. We use the support vector machine model for simulation and solve the problem of overlearning through parameter optimization. Using the feature parameters obtained, the support vector machine and radial basis function neural network model are used to classify the modulated signals. The experimental results show that the performance of the support vector machine is better than that of the radial basis function neural network. The network communication signal has good performance under high SNR. However, at low SNR, the former still has a high accuracy, while the latter has a serious performance degradation. Therefore, the support vector machine model is considered to be more robust when applied to computer network communication signals.
In facial recognition, occlusion has an impact on the accuracy of facial detection. OpenCV and multi-task convolutional neural network are used to analyze the impact of occlusion on facial recognition. Three networks of the multi-task convolutional neural network, namely P-Net, R-Net, and O-Net are cascaded for training. Face detection network is trained using the Wider Face dataset, and key point detection network is trained using the CNN_ FacePoint dataset. The results showed that each network achieved recognition accuracy of 91.63, 95.68, and 99.26
In order to solve the problem of Spanish phonetic translation of label terms in new energy equipment, this study chooses the strategy of artificial intelligence algorithm modeling to construct an optimization model for Spanish phonetic translation of label terms in new energy equipment. Based on the existing Spanish phonetic translation system, this study constructs a Spanish phonetic translation model of label terms that combines the main vowel and the sequence of Spanish phonetic elements with the help of some preliminary knowledge of artificial intelligence algorithms. In this way, the Spanish phonetic translation model of label terms in this study was updated. After testing, this study found that through the addition of artificial intelligence algorithms, the recognition ability of the Spanish pronunciation optimization model for label terms in new energy equipment did not decline significantly, but performed well in Spanish pronunciation translation, so the artificial intelligence algorithm has a positive impact on the Spanish pronunciation optimization of label terms in new energy equipment.
With the continuous development of information technology, under the existing teaching environment and the degree of informationization, relying only on teachers to pay attention to each student’s learning situation is not able to accurately analyze the students' status. To solve this issue, the study suggests a hierarchical multilabel classification technique based on clustering relationships. Using the integration principle, the clustered clustering tree is built as a random forest classification model after the labels in the sample set are clustered using the K-mean clustering method. Moreover, the random forest classification model is improved to address the problem of reduced classification effect caused by the increased randomness of the decision tree. The outcomes indicated that the accuracy of the multilabel classification algorithm with clustered relationships was 95.6
As 5G technology advances, particularly with the adoption of 3GPP Rel-16/Rel-17 standards, there is a growing need for terminal equipment in the field of power communication that has low power consumption, cheap cost, and high reliability. In order to meet the demands of medium- and low-speed IoT application scenarios, 5G RedCap (reduced capability) technology is a new type of 5G terminal defined in the Rel-17 version. It is especially suitable for industrial IoT, wearable devices, and other scenarios. By reducing terminal design complexity and system cost, 5G RedCap technology meets the needs of medium- and low-speed IoT application scenarios. In this context, this paper thoroughly studies the application of 5G RedCap technology based on 5G RedCap technology in electric power communication services, and explores its ability to support electric power business scenarios. This article first examines the communication needs of the power business and how well they align with 5G network slicing. It next suggests a 5G network slicing coupling model that is tailored to various power business scenarios. Second, the complexity reduction technology, power saving technology, and its identification method of 5G RedCap terminals are examined, and their power consumption optimization effect is evaluated, based on the current scenario of 1800 MHz spectrum allotment. In conclusion, this paper presents the particular needs for the integration of the 5G RedCap and the 4G electric power network along with the future access requirements of large-scale electric power terminals. It also designs multimode communication terminal equipment to enhance the security and efficiency of network access for the electric power system’s end equipment. The study’s findings demonstrate that terminal equipment’s energy economy and communication performance may be significantly enhanced by implementing 5G RedCap technology in electric power communication.
When shopping online, consumers can only see partial photos of products from merchants or photos of models trying on the clothes, but the quality of these photos differs significantly from the actual wear, which leads to a high return rate. This paper introduces a new algorithm for human stereoscopic sensation using convolutional neural networks. The 3D human body modeling based on an image of a male user’s clothing is a more effective method for human body sensing. Firstly, a mathematical model for predicting important body types was developed using VGGNet, which takes as input frontal and side-dressed images with chest, waist, and hip circumference to improve prediction accuracy. Secondly, by parametric modeling of the human body, the correspondence between important scales and morphological dimensions was established, effectively reconstructing the 3D morphology of the human body.
To solve the issues of low recognition efficiency and high feature extraction error rate in the current intelligent recognition models of tongue images in traditional Chinese medicine, this study uses attention mechanism to improve the convolutional neural network algorithm. Moreover, based on an improved algorithm, it constructs a new intelligent recognition model for traditional Chinese medicine tongue images to improve the accuracy of tongue image recognition. In the comparative experiment of the improved algorithm, the algorithm had the highest classification accuracy, the fastest calculation speed of 6.3bps, and the lowest space occupancy rate of only 23.1
As phishing attacks on unsuspecting online users become a high priority task and recognition of phishing URLs has become a critical concern now. Although machine learning (ML) methods have shown potential in detecting phishing URLs, deep learning (DL) algorithms offer the added benefit of automatically extracting features and handling large datasets more efficiently. The Proposed work analyzed various deep learning algorithms, including decision tree, CNN, RNN, NLP, and ANN. A goal of the suggested approach is to integrate recurrent neural networks (RNN) with decision tree techniques to attain high precision in detecting Phishing URLs. Phishing attacks pose a major risk by tricking users into disclosing confidential information on websites that appear legitimate, leading to possible data breaches and financial damage An Identifying the need for robust and accurate detection methods, our study aimed to develop an effective approach to distinguish between trustworthy and fraudulent URLs, protecting consumers from becoming targets of phishing scams. Our comparison of several deep learning algorithms found that the RNN-decision tree hybrid model determined better results in correctness and effectiveness. This model attained high recall, F1-score, accuracy, and precision. The proposed work results demonstrate its efficacy, identify phishing URLS, and explain how it can be applied to lessen the danger of phishing scams.
Addressing the issues of untimely discovery and low identification accuracy of hazards in China’s distribution networks, this paper introduces a hazard identification method for distribution networks based on generative adversarial networks (GANs). The method employs a modular design comprising three parts: hazard traveling wave acquisition, hazard location, and hazard type identification. The hazard traveling wave acquisition part uses wavelet transform to extract the waveform of the initial hazard wave and records the arrival time at the device. The hazard location part improves the dual-end traveling wave location method to determine the hazard location, enhancing location accuracy. The hazard type identification part uses affinity propagation to cluster the hazard traveling waves, obtaining classification features of the hazard traveling waves, and then identifies distribution network hazards through GANs. Finally, the method was applied to a distribution network in a certain region, achieving a hazard identification accuracy of 81.3
This article presents methods for the automated security analysis of Android applications, which can be used to search for cryptographic vulnerabilities, vulnerabilities in third-party software components, authentication, and authorization, as well as to identify the storage and transmission of sensitive information in clear text. The accuracy of searching for the given types of vulnerabilities using automated vulnerability search tools and a software prototype is analyzed.
One of the key elements in solving spam filtering problems is the text vectorization method. This article proposes a vectorization method based on matching text to pairs of intentionalities. A list of intentionality pairs is extracted and a synthetic dataset is generated from text utterances. A neural network is designed and trained to determine the degree to which each intentionality belongs to the textual expression provided as the input of the model. The developed method is tested on the problem of filtering spam messages using logistic regression and the Enron dataset and SMS dataset.
The mining algorithm in smart city blockchain systems using the Proof-of-Work consensus mechanism is studied. Well-known studies in the field of selfish mining detection are analyzed. A method for protecting a blockchain from selfish mining attacks is presented, and a selfish mining detection plugin is developed based on this method. This plugin is designed for miner software and enables the analysis of data patterns received from the mining pool. The proposed solution outperforms existing selfish mining detectors by identifying the attacking mining pool and has lower error rates.
In [1], two new approaches to key pre-distribution based on an ideal additive homomorphic secret sharing scheme (SSS) are proposed. However, it has not been possible to prove their security against insider attacks in the general case. In this study, a simple method for distributing shares, corresponding to the first approach in [1], is proposed and analyzed using Shamir’s SSS. The problem considered is the distribution of 2n shares among n participants in such a way that each participant holds two shares, and every pair of participants corresponds to a (3,4)-threshold scheme, while the overall threshold can be arbitrary. It should be noted that this is the first time such a problem is solved in the theory of secret sharing. The analysis shows that the key agreement protocol based on the proposed technique of shares pre-distribution is not resistant to insider attacks. A general necessary condition for the security of a key agreement protocol in the internal adversary model is obtained.
This paper develops theoretical justifications and presents simulation results confirming the key influence of spreading bipolar sequences on the correlation properties of marker (spread) bipolar sequences. This study develops a method to predict the values of the normalized autocorrelation function of the marker sequence and also specifies the conditions for obtaining accurate predictions. This article also proves the existence of a relationship between the properties and characteristics of the cross-correlation function between the target (embedded) marker sequence and the tested sequence (distorted by noise or attacks) and the properties of the autocorrelation function of the embedded marker sequence. Finally, in the concluding part of the article, the indicators and criteria of robustness of spreading and marker bipolar sequences are formulated.
The research focuses on methods for automating security in DevOps pipelines within the DevSecOps framework, emphasizing the integration of tools, processes, and cultural shifts to enhance the security of software products. The research set the following tasks: analysis of modern DevSecOps methodologies and tools; assess the potential of using artificial intelligence and machine learning to automate information security tasks; identify the main problems and barriers to integrating DevSecOps into continuous integration and delivery (CI/CD) processes; identify promising areas for automation development in the field of security. The study uses a comparative analytical review method, including an analysis of scientific literature, industrial practices and documentation of modern DevSecOps tools, the “Shift-Left Security” and “Security as Code” approaches. Open sources, CI/CD platform documentation, and data on the use of AI in information security were used. The research identifies key principles for integrating security into DevOps: early vulnerability detection, automation of security processes, implementation of “Security as Code,” and enhanced threat monitoring. Modern DevSecOps tools are reviewed, including static and dynamic code analysis, security policy management systems, secret management solutions, and AI-powered proactive threat detection mechanisms. The study finds that automation minimizes human error, accelerates vulnerability detection and remediation processes, and ensures compliance with regulatory requirements. However, certain limitations were also identified, including the complexity of tool integration, a shortage of DevSecOps specialists, and resistance to changes within development and operations teams. Future trends indicate further advancements in AI-driven solutions and automated frameworks for security manage-ment. This research contributes to the field of information security by uncovering methods for automating DevSecOps integration into CI/CD processes and exploring the potential of AI for predictive threat analytics. It highlights key trends in security automation within modern cloud and containerized environments.
This study focuses on ensuring the security of smart voice assistants against the most pressing threats by reducing the number of illegitimate activations. An analysis of threats to smart voice assistants (SVAs) and a list of the functional capabilities of voice assistants are presented. The goal is to reduce the number of illegitimate activations of SVAs. The architecture and operating method of a proprietary security module that reduces the number of illegitimate activations are presented. The developed module is tested and the results are positive.
This paper studies the problem of reducing the attack surface from an internal attacker in heterogeneous systems for processing and storing big data by selecting the optimal data obfuscation method based on anonymization technologies. The study analyzes the terminology and systematizes data-hiding methods to reduce the attack surface in big data processing and storage systems. A formal formulation of the problem of finding the optimal data obfuscation method and an algorithm for solving it across various types of datasets are proposed, taking into account evaluation criteria specific to each class of methods. The implementation of a software prototype for supporting decision-making and selecting the optimal method for solving practical problems is described. Experimental testing and analysis of its results are carried out.
The problem of optimizing neural networks for large language models (LLMs) such as ChatGPT is discussed. One of the directions being developed for optimizing LLMs is knowledge distillation—the transfer of knowledge from a large teacher model to a smaller student model without significant loss of accuracy of the result. The existing methods of knowledge distillation have certain disadvantages: inaccurate knowledge transfer, long learning process, and error accumulation in long sequences. A combination of methods that contribute to improving the quality of knowledge distillation is considered: selective teacher intervention in the student’s learning process and low-rank adaptation. The proposed combination of knowledge distillation methods can be applied to problems with limited computational resources.
The structure and composition of container images and associated security issues, as well as scanning methods for detecting vulnerabilities in container images, are analyzed. An approach is developed that eliminates the identified shortcomings. A software prototype for automated image security scanning with dynamic monitoring support is developed and tested.
