Computational Intelligence最新文献

In the Internet of Medical Things (IoMT), medical images play a vital role in healthcare systems by enabling accurate diagnosis, effective treatment planning, and continuous monitoring of various diseases. Deep learning models have strong potential to facilitate early detection by recognizing patterns in medical images. Although deep learning systems are more advanced, ambiguity continues to act as a major obstacle to progress. Ambiguity typically arises in areas such as accuracy, interpretability, and generalization. The challenge lies in balancing high performance with transparent and understandable decision-making processes. This uncertainty can hinder trust computation and practical deployment, especially in critical applications. Moreover, these models face challenges with data handling and often lack robustness, which affects their reliability. To address the above-mentioned issues, a novel multi-disease detection framework is proposed using a deep learning approach. The input medical images are gathered from online datasets using the IoMT-enabled devices. To detect different kinds of diseases, an adaptive vision transformer-based multi-scale residual dense network (AViT-MRDNet) is developed. This model processes the input images to classify four types of diseases including skin lesions, brain tumors, lung tumors, and breast cancer. Here, the hybrid bonobo with rat swarm optimizer (HB-RSO) is developed to optimize the parameters. Finally, the proposed framework produces the disease detection results. The overall performance of the implemented framework is verified by using various validation metrics such as precision, accuracy, false discovery rate, sensitivity, F1-score, false positive rate, specificity, Mathew correlation coefficient, false negative rate, and negative predictive value to showcase the effectiveness in detecting the multi-disease. It attains an accuracy of 96.022, sensitivity of 95.164, specificity of 96.880, and precision of 96.825. Thus, these findings prove that the introduced framework has enhanced the system performance. The hybrid optimization algorithm in the proposed approach offers a robust solution for doctors and healthcare professionals to predict multiple diseases in patients. Therefore, it shows that the developed model is superior and delivers more reliable performance to other frameworks. The result confirmed that the proposed model can accurately and precisely detect the multi-disease and help the patients for early diagnosis. It leverages the strength of the vision transformer and the residual densenet to enhance the system performance.

Online reviews are essential for users to make purchasing decisions over any products or films. Reviews found on e-commerce websites often closely align with ratings and enhance the learning experience for potential buyers. The customer review information is of huge interest to companies as well as customers. The information from customer feedback is not only valuable for consumers but also provides valuable feedback to companies. However, since customer reviews are usually presented as unstructured free text, extracting meaningful ratings from user opinions can be quite challenging. To overcome this issue, an effectual system is modeled for sentiment classification and review rating prediction utilizing reviews named fuzzy-based hierarchical attention gated recurrent network (Fuzzy-HAGRN). At first, the input reviews are fed to Bidirectional Transformers for Language Understanding (BERT) tokenization, and after that, feature extraction is executed for the extraction of several essential features. Then, Fuzzy-HAGRN is utilized for sentiment classification and review rate prediction. Here, the Fuzzy-HAGRN method is developed by integrating the fuzzy concept, hierarchical attention network (HAN) as well as gated recurrent network (GRU). Finally, rating prediction is performed using the same Fuzzy-HAGRN. Fuzzy-HAGRN has obtained a precision of 92.20%, recall of 91.80%, and F1-score of 91.50%.

De-duplication is critically important for cloud computing since it permits the detection of repeated data within the cloud system under fewer resources and expenses. De-duplication removes unnecessary data from the cloud centers, which helps to identify the appropriate owner of cloud material. Each piece of data saved in the cloud is owned by a large number of cloud users, even though it contains just a single copy of the data. The dynamic nature of the cloud resources is not handled by the prior deduplication models and the existing models require more computing power for accurately determining the presence of duplicate files in the cloud system. In addition, the prior models split the files into chunks for determining the similar files in the cloud system which affects the quality of the data. To conquer these difficulties, an adaptive deep learning-based data deduplication model is developed using an optimization algorithm. The main innovation of the proposed research is to rapidly detect duplicate records in the cloud data and also provide high-level security while maintaining the operational efficiency of the cloud system. The proposed model acts as an efficient attack resistance system and it also ensures the data availability of the cloud system more rapidly. This data deduplication implies in detecting and examining the patterns inside records of information to precisely notice and eliminate repeated identical information. Hence, the data connected to the input pattern is given to the Serial Cascaded Autoencoder with Gated Recurrent Unit (SCA-GRU) for the deduplication process. After deduplication, the unnecessary data are removed for the precise consumption of resources to store exclusive data. To maintain the security of data, the optimal key-based data sanitization process is performed, in which the key is optimally generated with the aid of a Mutated Fitness-Based Krill Herd Optimization Algorithm (MF-KHO). This encoded data is then safely kept in the cloud, which protects the data from illegal access and possible defense breaches. The outcome of the suggested approach is validated with the previous data deduplication system to show the efficiency of the developed model. The experimental results showed that the recommended deduplication approach reaches an accuracy of 95.37%. Through efficient data deduplication, the storage requirement of the data is greatly reduced, which facilitates cost reduction and resource optimization within the cloud system and also the storage capacity utilization of the cloud system is greatly improved.

Visual impairment has a drastic impact on the psychological and cognitive well-being of individuals. Recent progress in advanced assistive technologies (AATs) has emerged as an essential tool to mitigate the adverse impact of blindness and enhance the quality of life of visually impaired persons (VIPs). Like generic object identification, the VIPs face difficulties in identifying their garments. Such a limitation severely impacts their identity as they cannot select dresses according to their preferences for different contexts and occasions. To this end, in this paper, we present a proof-of-concept (POC) implementation of a mobile application for real-time fabric pattern classification to assist VIPs in selecting the cloth with fabric patterns of their choice. The proposed framework uses a robust and compute-efficient convolutional neural network (CNN) named FabricNet to classify four types of fabric patterns (lattice, printed, solid, and stripe). The designed FabricNet model uses efficient feature enhancement (EFE), efficient feature refinement (EFR), and enhanced feature fusion (EFF) blocks to extract discriminative texture features from the fabric pattern images. We evaluated the performance of the proposed FabricNet on a recent open-source fabric image dataset. Compared to the existing baseline, the FabricNet model, with 1.64 M parameters, 6.25 MB memory storage size, and 4.09 GFLOPs, attained state-of-the-art classification accuracy with reduced model storage size and competitive computation time. Finally, we optimized the trained model and integrated it with a mobile application developed using the Android Studio software development kit (SDK) for real-time classification of fabric images. The designed mobile application on an Android mobile phone uses its back camera to capture garment images and provides predicted fabric type using audio feedback. Thus, it can assist VIPs in selecting clothing with fabric patterns of their choice in real time.

Diverse musical styles are crucial ways for human beings to represent their emotions and interact with each other, whereas the essentials of musical signals are a time-lagged nonlinear dynamical system and their nonlinearity is difficult to analyze by conventional approaches. In this paper, the music is firstly framed depending on the subsections of its structure, then the Lyapunov exponent and the correlation dimension of the music signal are computationally analyzed, which reveals that the internal construction of the music signal is sophisticated with weak chaotic features. By retrieving the local characteristics of the music signal and extrapolating its holistic characteristics, the nonlinearity of the signal rendered by diverse musical styles also has a distinguishable difference. It is observed from the experiments that the maximum Lyapunov exponent of music characterized as “happy” and “relaxing” reaches 0.23, while the range of fluctuations in the correlation dimensions spans from 3.2 to 5.7. Furthermore, a discrepancy of 4.1 is noted in the correlation dimensions of music classified as “loud” and “uplifting,” indicative of the intricate nature of music signals' internal structures and the attenuation of chaotic characteristics. The M5 model exhibits an accuracy of 91.26% for classical music, representing a 2.9% enhancement over conventional methodologies. According to the aforementioned chaotic analysis, the originally designed nonlinearity extracting pattern for diverse music signals in the musical recognizing system demonstrates excellent performance.

There has been an increasing interest in using GNNs to build recommender systems as they enable the representation of complex relationships between users and items through knowledge graph embeddings. However, most of the knowledge-graph-based systems focus only on ratings or reviews to build relationships. This prevents a comprehensive understanding of structural and positional information within graph data as well as user preferences that can change in time, as well. In order to address these issues, this paper aims to propose an advanced end-to-end Graph Neural Network architecture that significantly enhances recommendation system capabilities through the integration of state-of-the-art embedding techniques, knowledge graph frameworks, and transfer learning strategies. Incorporating positional encoding and topological feature extraction, the proposed model captures intricate user–item relationships and offers a robust representation that surpasses current approaches. A pretrained encoder facilitates knowledge transfer, effectively bridging domain gaps and amplifying prediction accuracy. Comprehensive evaluations against established baseline models reveal that our architecture has demonstrated enhanced accuracy, precision, and overall robustness. These results highlight the efficacy of combining knowledge graphs, sophisticated embedding strategies, and cross-domain transfer learning in building next-generation recommender systems, providing valuable insights for future advancements in the field.

In this study, the main objective is to develop an algorithm capable of identifying and delineating tumor regions in breast ultrasound (BUS) and mammographic images. The technique employs two advanced deep learning architectures, namely U-Net and pretrained SAM, for tumor segmentation. The U-Net model is specifically designed for medical image segmentation and leverages its deep convolutional neural network framework to extract meaningful features from input images. On the other hand, the pretrained SAM architecture incorporates a mechanism to capture spatial dependencies and generate segmentation results. Evaluation is conducted on a diverse dataset containing annotated tumor regions in BUS and mammographic images, covering both benign and malignant tumors. This dataset enables a comprehensive assessment of the algorithm's performance across different tumor types. Results demonstrate that the U-Net model outperforms the pretrained SAM architecture in accurately identifying and segmenting tumor regions in both BUS and mammographic images. The U-Net exhibits superior performance in challenging cases involving irregular shapes, indistinct boundaries, and high tumor heterogeneity. In contrast, the pretrained SAM architecture exhibits limitations in accurately identifying tumor areas, particularly for malignant tumors and objects with weak boundaries or complex shapes. These findings highlight the importance of selecting appropriate deep learning architectures tailored for medical image segmentation. The U-Net model showcases its potential as a robust and accurate tool for tumor detection, while the pretrained SAM architecture suggests the need for further improvements to enhance segmentation performance.

Brain tumor segmentation is critical for diagnosis, treatment planning, and evaluation. However, existing methods such as U-Net, FCN, and Mask R-CNN often struggle with capturing fine-grained tumor boundaries, handling complex tumor heterogeneity, and maintaining high sensitivity across different tumor subregions. To overcome these challenges, this study proposes an Attention-Driven GAN-UNet framework that integrates U-Net with Generative Adversarial Networks (GANs) and a Channel-Spatial Attention Module (CSAM). This innovative approach enhances segmentation accuracy and focus mapping by directing the network's attention to clinically relevant regions. Trained on the BraTS 2020 dataset, our method surpasses traditional techniques, achieving a Dice Similarity Coefficient (DSC) of 0.99. The proposed framework visualizes intricate tumor morphologies, reduces false positives, and offers robust computational efficiency, making AttnGAN-UNet a promising tool for clinical brain tumor segmentation and analysis.