PeerJ Computer Science最新文献

In the era of informational ascendancy, the discourse of artistic communication has transcended the confines of conventional physical domains and geographical boundaries, extending its purview ubiquitously across the global expanse. Consequently, the predominant mode of artistic interaction has evolved towards swift and extensive engagement through virtual platforms. However, this paradigm shift has given rise to the imperative task of meticulous categorization and labeling of an extensive corpus of artistic works, demanding substantial temporal and human resources. This article introduces an innovative bimodal time series classification model (BTSCM) network for the purpose of categorizing and labeling artworks on virtual platforms. Rooted in the foundational principles of visual communication and leveraging multimedia fusion technology, the proposed model proves instrumental in discerning categories within the realm of video content. The BTSCM framework initiates the classification of video data into constituent image and sound elements, employing the conceptual framework of visual communication. Subsequently, feature extraction for both forms of information is achieved through the application of Inflated 3D ConvNet and Mel frequency cepstrum coefficient (MFCC). The synthesis of these extracted features is orchestrated through a fusion of fully convolutional network (FCN), deep Q-network (DQN), and long short-term memory (LSTM), collectively manifesting as the BTSCM network model. This amalgamated network, shaped by the union of fully convolutional network (FCN), DQN, and LSTM, adeptly conducts information processing, culminating in the realization of high-precision video classification. Experimental findings substantiate the efficacy of the BTSCM framework, as evidenced by outstanding classification results across diverse video classification datasets. The classification recognition rate on the self-established art platform exceeds 90%, surpassing benchmarks set by multiple multimodal fusion recognition networks. These commendable outcomes underscore the BTSCM framework's potential significance, providing a theoretical and methodological foundation for the prospective scrutiny and annotation of content within art creation platforms.

Feature selection is essential for enhancing the performance and reducing the complexity of speech emotion recognition models. This article evaluates various feature selection methods, including correlation-based (CB), mutual information (MI), and recursive feature elimination (RFE), against baseline approaches using three different feature sets: (1) all available features (Mel-frequency cepstral coefficients (MFCC), root mean square energy (RMS), zero crossing rate (ZCR), chromagram, spectral centroid frequency (SCF), Tonnetz, Mel spectrogram, and spectral bandwidth), totaling 170 features; (2) a five-feature subset (MFCC, RMS, ZCR, Chromagram, and Mel spectrogram), totaling 163 features; and (3) a six-feature subset (MFCC, RMS, ZCR, SCF, Tonnetz, and Mel spectrogram), totaling 157 features. Methods are compared based on precision, recall, F1-score, accuracy, and the number of features selected. Results show that using all features yields an accuracy of 61.42%, but often includes irrelevant data. MI with 120 features achieves the highest performance, with precision, recall, F1-score, and accuracy at 65%, 65%, 65%, and 64.71%, respectively. CB methods with moderate thresholds also perform well, balancing simplicity and accuracy. RFE methods improve consistently with more features, stabilizing around 120 features.

The domain of Human Activity Recognition (HAR) has undergone a remarkable evolution, driven by advancements in sensor technology, artificial intelligence (AI), and machine learning algorithms. The aim of this article consists of taking as a basis the previously obtained results to implement other techniques to analyze the same dataset and improve the results previously obtained in the different studies, such as neural networks with different configurations, random forest, support vector machine, CN2 rule inducer, Naive Bayes, and AdaBoost. The methodology consists of data collection from smartphone sensors, data cleaning and normalization, feature extraction techniques, and the implementation of various machine learning models. The study analyzed machine learning models for recognizing human activities using data from smartphone sensors. The results showed that the neural network and random forest models were highly effective across multiple metrics. The models achieved an area under the curve (AUC) of 98.42%, a classification accuracy of 90.14%, an F1-score of 90.13%, a precision of 90.18%, and a recall of 90.14%. With significantly reduced computational cost, our approach outperforms earlier models using the same dataset and achieves results comparable to those of contemporary deep learning-based approaches. Unlike prior studies, our work utilizes non-normalized data and integrates magnetometer signals to enhance performance, all while employing lightweight models within a reproducible visual workflow. This approach is novel, efficient, and deployable on mobile devices in real-time. This approach makes it an ideal fit for real-time mobile applications.

With the rapid development of the furniture industry, automatic classification of furniture images has become an important research area. However, this task faces several challenges, including complex image backgrounds, diverse furniture types, and varying forms. To address these issues, we propose a novel furniture image classification method, MobileNetNAK, based on the MobileNetV3 network. First, the method integrates a non-local attention module to capture non-local dependencies within images, significantly enhancing the model's ability to extract key information. Second, the Adamax optimizer is employed to train the model. By adaptively adjusting the learning rate, it accelerates convergence and reduces the risk of overfitting. Third, the Kolmogorov-Arnold networks method is incorporated to decompose complex convolution operations into multiple simpler ones, thereby improving computational efficiency and feature extraction capabilities. Experimental results demonstrate that MobileNetNAK significantly improves classification performance in furniture image tasks. On Dataset 1, the model achieves improvements of 6.7%, 6.6%, 6.6%, and 6.6% in accuracy, precision, recall, and F1-score, respectively, compared to the baseline. On Dataset 2, the corresponding improvements are 2.7%, 2.4%, 2.7%, and 2.9%. Additionally, the model maintains a high inference speed of 147.80 fps, balancing performance with computational efficiency. These results highlight the strong adaptability and deployment potential of MobileNetNAK in multi-category and fine-grained furniture image classification tasks, offering a novel and effective solution for this domain.

In today's digital world, cryptocurrencies like Bitcoin can secure transactions without banks. However, the rise of quantum computing poses significant threats to their security, as traditional cryptographic methods may be easily compromised. In addition, the existing algorithms face difficulties like slow transaction speeds, interoperability issues between different cryptocurrencies, and privacy concerns. Hence, Quantum Crypto Guard for Secure Transactions (QCG-ST), a novel blockchain framework, is introduced, offering enhanced security and efficiency for cryptocurrency transactions. The QCG-ST employs lattice-based cryptography to provide robust protection against quantum threats and incorporates a new consensus mechanism to increase the transaction speed and reduce energy consumption. The QCG-ST system uses lattice-based encryption that is based on the Ring Learning With Errors (Ring-LWE) issue to protect itself from quantum assaults. It uses sharding, a Proof-of-Stake (PoS) consensus method, and a threshold signature scheme (TSS) to make the system more scalable and use less energy. Zero-knowledge proofs (ZKPs) are used to check transactions without giving out private information. We offer a cross-chain atomic swap protocol that uses hashed time-lock contracts to make sure that it works on all platforms. Blockchain transaction data utilized in testing originated from the Bitcoin Historical Dataset available on Kaggle, and quantum resistance has been assessed using the Qiskit Aer simulator. It evaluated the framework's performance to that of traditional methods like Payment Channel-Lightning Network (PC-LN), Variational Quantum Eigensolver (VQE), and Cross-Chain Transaction with Hyperledger (CCT-H). Results show that QCG-ST does far better than traditional systems in terms of transaction success rate (up to 98.5%), speed, energy efficiency, latency, and throughput, especially when tested in a quantum-simulated environment. This study completes in an essential vacuum in blockchain technology by suggesting a strong, quantum-resistant, privacy-protecting architecture that can handle the problems that could arise up in decentralized digital banking in the future.

This study proposes a decision-making model based on deep reinforcement learning (DRL) for agricultural financial transactions, addressing core challenges such as significant data noise, strong time-series dependence, and limited strategy adaptability. We developed a multifactor dynamic denoising framework by integrating the Grubbs test for outlier detection and the median absolute deviation (MAD) method for noise handling. This framework categorizes agricultural financial indicators into six feature types, significantly enhancing robustness against data noise and improving model reliability. Furthermore, an long short-term memory (LSTM)-enhanced DRL architecture is employed, incorporating a sliding window mechanism to capture market timing features. This framework constructs a transaction cost-based reward function. It establishes an intelligent trading decision model based on the LSTM algorithm and the data query language (DQL). Experimental results demonstrate an annualized return of 45.12% and a 35% reduction in maximum retracement for Deere & Company and BAYN.DE. The Sharpe ratio reaches 1.51, reflecting a 62% improvement over the benchmark model. The results validate the robustness of the proposed decision-making model in the face of price fluctuations and policy interventions. This model addresses critical bottlenecks in the application of DRL in agricultural finance, facilitating the transition of agricultural economic management from empirical judgment to data-driven approaches. Through three key innovations-data denoising, time-series modeling, and domain adaptation-it provides a vital decision-support tool for advancing smart agriculture.

Accurate sales forecasting is vital for balancing demand and supply and enhancing profitability in the retail sector. Deep learning (DL) models have shown promise in this area; however, most either handle temporal or spatial patterns in isolation. Moreover, many studies rely on synthetic datasets or omit critical contextual variables, reducing real-world accuracy. This study proposes a hybrid convolutional neural network (CNN)-long short-term memory (LSTM) model for retail sales forecasting using real-world data enhanced with environmental and demographic variables in term of holidays, salary days, protests, and weather conditions. CNNs capture spatial patterns, while LSTMs model temporal dependencies, making the hybrid architecture well-suited for multivariate forecasting tasks. Our model demonstrates a significant improvement in predictive performance, achieving a mean absolute percentage error (MAPE) of 4.16%, outperforming traditional and standalone neural models. By incorporating external factors, the proposed approach enables more reliable forecasting and supports informed decision-making in retail operations.

Early diagnosis of Parkinson's disease (PD) is challenging due to subtle initial symptoms. This study introduces an advanced machine learning framework that leverages particle swarm optimization (PSO) to improve PD detection through vocal biomarker analysis. Our novel approach unifies the optimization of both acoustic feature selection and classifier hyperparameter tuning within a single computational architecture. We systematically evaluated PSO-enhanced predictive models for PD detection using two comprehensive clinical datasets. Dataset 1 includes 1,195 patient records with 24 clinical features, and Dataset 2 comprises 2,105 patient records with 33 multidimensional features spanning demographic, lifestyle, medical history, and clinical assessment variables. For Dataset 1, the PSO model achieved 96.7% testing accuracy, an absolute improvement of 2.6% over the best-performing traditional classifier (Bagging classifier at 94.1%), while maintaining exceptional sensitivity (99.0%) and specificity (94.6%). Results were even more significant for Dataset 2, where the PSO model reached 98.9% final accuracy, a 3.9% improvement over the LGBM classifier (95.0%), with near-perfect discriminative capability (AUC = 0.999). These performance gains were achieved with reasonable computational overhead, averaging 250.93 s training time for Dataset 2, suggesting the practical viability of PSO optimization for clinical prediction tasks. Our findings underscore the potential of intelligent optimization techniques in developing practical decision support systems for early neurodegenerative disease detection, with significant implications for clinical practice.

In recent years, influencer marketing has gained increasing popularity, with many influencers embedding product information into their content (e.g., videos and articles). When fans encounter these messages, they may make unplanned purchases, resulting in impulse buying behavior, a long-standing issue in marketing research. This study aims to explore the factors that lead to such behavior. Using the Stimulus-Organism-Response (S-O-R) model as a framework, the study investigates how interactions between individuals and influencer content (Stimuli) trigger psychological changes in consumers, namely positive affect, flow state, and emotional attachment (Organism), which in turn lead to impulse buying behavior (Response). The study surveyed fans who had previously purchased products recommended by influencers, collecting 404 valid responses. The findings reveal that: (1) Consumers' psychological changes (positive affect, flow state, and emotional attachment) significantly and positively influence impulse buying behavior. (2) Scarcity, discounted price, review quality, and observational learning also have significant positive effects on impulse buying. (3) Social presence and sense of belonging significantly enhance flow state. (4) Entertainment and informativeness significantly enhance emotional attachment.

The practice of diagnosing breast cancer retains its scope for improvement in medical imaging, where every correct and timely diagnosis enhances the survival rate of patients. This article presents an integrated approach utilizing patch-wise breast image segmentation, hybrid deep feature extraction, followed by progressive cyclical convolutional neural networks (P-CycCNN), and firebug swarm optimization (FSO) to enhance breast cancer detection. This method first incorporates image segmentation by patches to break down the mammography images into smaller patches, which are easier to focus on and allow for the extraction of more features to boost detection rates. Hybrid feature extraction combines convolutional neural network (CNN) features extracted from pre-trained models with handcrafted features that describe texture and shape, thereby enabling the model to grasp the nuances of both coarse and fine images comprehensively. The progressive cyclical CNN strategy incorporates cyclical, re-adjusted learning rates and a progressive training schedule to accelerate and enhance the model's convergence. FSO is introduced to adjust the hyperparameters of the CNN topology, including the learning rate and regularisation parameters, thereby enhancing training and feature-fusion processes. Evaluated on the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) dataset, the proposed model achieved 98% test accuracy, 95% precision, 97.2% recall, 96% F1-score, and an AUC of 0.95, outperforming baseline CNN models by 4%-6% across key metrics. This approach holds great potential for enhancing detection systems in clinics, allowing earlier and more accurate detection of malignant lesions.