Improvement In F1 Score Research Articles

Capsule network approach for monkeypox (CAPSMON) detection and subclassification in medical imaging system

In response to the pressing need for the detection of Monkeypox caused by the Monkeypox virus (MPXV), this study introduces the Enhanced Spatial-Awareness Capsule Network (ESACN), a Capsule Network architecture designed for the precise multi-class classification of dermatological images. Addressing the shortcomings of traditional Machine Learning and Deep Learning models, our ESACN model utilizes the dynamic routing and spatial hierarchy capabilities of CapsNets to differentiate complex patterns such as those seen in monkeypox, chickenpox, measles, and normal skin presentations. CapsNets’ inherent ability to recognize and process crucial spatial relationships within images outperforms conventional CNNs, particularly in tasks that require the distinction of visually similar classes. Our model’s superior performance, demonstrated through rigorous evaluation, exhibits significant improvements in accuracy, precision, recall, and F1 score, even with limited data. The results highlight the potential of ESACN as a reliable tool for enhancing diagnostic accuracy in medical settings. In our case study, the ESACN model was applied to a dataset comprising 659 images across four classes: 178 images of Monkeypox, 171 of Chickenpox, 80 of Measles, and 230 of Normal skin conditions. This case study underscores the model’s effectiveness in real-world applications, providing robust and accurate classification that could greatly aid in early diagnosis and treatment planning in clinical environments.

LiDAR Loop Closure Detection using Semantic Graphs with Graph Attention Networks

In this paper, we propose a novel loop closure detection algorithm that uses graph attention neural networks to encode semantic graphs to perform place recognition and then use semantic registration to estimate the 6 DoF relative pose constraint. Our place recognition algorithm has two key modules, namely, a semantic graph encoder module and a graph comparison module. The semantic graph encoder employs graph attention networks to efficiently encode spatial, semantic and geometric information from the semantic graph of the input point cloud. We then use self-attention mechanism in both node-embedding and graph-embedding steps to create distinctive graph vectors. The graph vectors of the current scan and a keyframe scan are then compared in the graph comparison module to identify a possible loop closure. Specifically, employing the difference of the two graph vectors showed a significant improvement in performance, as shown in ablation studies. Lastly, we implemented a semantic registration algorithm that takes in loop closure candidate scans and estimates the relative 6 DoF pose constraint for the LiDAR SLAM system. Extensive evaluation on public datasets shows that our model is more accurate and robust, achieving 13% improvement in maximum F1 score on the SemanticKITTI dataset, when compared to the baseline semantic graph algorithm. For the benefit of the community, we open-source the complete implementation of our proposed algorithm and custom implementation of semantic registration at https://github.com/crepuscularlight/SemanticLoopClosure.

Lightweight YOLOv8 for tongue teeth marks and fissures detection based on C2f_DCNv3

This paper propose a significantly enhanced YOLOv8 model specifically designed for detecting tongue fissures and teeth marks in Traditional Chinese Medicine (TCM) diagnostic images. By integrating the C2f_DCNv3 module, which incorporates Deformable Convolutions (DCN), replace the original C2f module, enabling the model to exhibit exceptional adaptability to intricate and irregular features, such as fine fissures and teeth marks. Furthermore, the introduction of the Squeeze-and-Excitation (SE) attention mechanism optimizes feature weighting, allowing the model to focus more accurately on key regions of the image, even in the presence of complex backgrounds. The proposed model demonstrates a significant performance improvement, achieving an average precision (mAP) of 92.77%, which marks a substantial enhancement over the original YOLOv8. Additionally, the model reduces computational cost by approximately one-third in terms of FLOPS, maintaining high accuracy while greatly decreasing the number of parameters, thus offering a more robust and resource-efficient solution. For tongue crack detection, the mAP increases to 91.34%, with notable improvements in F1 score, precision, and recall. Teeth mark detection also sees a significant boost, achieving an mAP of 94.21%. These advancements underscore the model’s outstanding performance in TCM tongue image analysis, providing a more accurate, efficient, and reliable tool for clinical diagnostic applications.

Application of human-computer interaction technology integrating biomimetic vision system in animation design with a biomechanical perspective

The combination of Human-Computer Interaction (HCI) technology with biomimetic vision systems has transformational potential in animation design, particularly by incorporating biomechanical principles to create immersive and interactive experiences. Traditional animation approaches frequently lack sensitivity to real-time human motions, which can restrict engagement and realism. This study addresses this constraint by creating a framework that uses Virtual Reality (VR) and Augmented Reality (AR) to generate dynamic settings that include a variety of human activities, informed by biomechanical analysis. A biomimetic vision system is used to record these motions with wearable sensors, allowing for precise monitoring of user activity while considering biomechanical factors such as joint angles, force distribution, and movement patterns. The recorded data is preprocessed using Z-score normalization methods and extracted using Principal Component Analysis (PCA). This study proposed an Egyptian Vulture optimized Adjustable Long Short-Term Memory Network (EVO-ALSTM) technique for motion classification, specifically tailored to recognize biomechanical characteristics of human movements. Results demonstrate a significant improvement in precision (93%), F1-score (91%), accuracy (95%), and recall (90%) for the motion recognition system, highlighting the effectiveness of biomechanical insights in enhancing animation design. The findings indicate that integrating real-time biomechanical data into the animation process leads to more engaging and realistic user experiences. This study not only advances the subject of HCI but also provides the framework for future investigations into sophisticated animation technologies that use biomimetic and biomechanical systems.

Adaptive Tip Selection for DAG-Shard-Based Federated Learning with High Concurrency and Fairness

To cope with the challenges posed by high-concurrency training tasks involving large models and big data, Directed Acyclic Graph (DAG) and shard were proposed as alternatives to blockchain-based federated learning, aiming to enhance training concurrency. However, there is insufficient research on the specific consensus designs and the effects of varying shard sizes on federated learning. In this paper, we combine DAG and shard by designing three tip selection consensus algorithms and propose an adaptive algorithm to improve training performance. Additionally, we achieve concurrent control over the scale of the directed acyclic graph’s structure through shard and algorithm adjustments. Finally, we validate the fairness of our model with an incentive mechanism and its robustness under different real-world conditions and demonstrate DAG-Shard-based Federated Learning (DSFL)’s advantages in high concurrency and fairness while adjusting the DAG size through concurrency control. In concurrency, DSFL improves accuracy by 8.19–12.21% and F1 score by 7.27–11.73% compared to DAG-FL. Compared to Blockchain-FL, DSFL shows an accuracy gain of 7.82–11.86% and an F1 score improvement of 8.89–13.27%. Additionally, DSFL outperforms DAG-FL and Chains-FL on both balanced and imbalanced datasets.

Detecting a wide range of epitranscriptomic modifications using a nanopore-sequencing-based computational approach with 1D score-clustering.

To date, over 40 epigenetic and 300 epitranscriptomic modifications have been identified. However, current short-read sequencing-based experimental methods can detect <10% of these modifications. Integrating long-read sequencing technologies with advanced computational approaches, including statistical analysis and machine learning, offers a promising new frontier to address this challenge. While supervised machine learning methods have achieved some success, their usefulness is restricted to a limited number of well-characterized modifications. Here, we introduce Modena, an innovative unsupervised learning approach utilizing long-read nanopore sequencing capable of detecting a broad range of modifications. Modena outperformed other methods in five out of six benchmark datasets, in some cases by a wide margin, while being equally competitive with the second best method on one dataset. Uniquely, Modena also demonstrates consistent accuracy on a DNA dataset, distinguishing it from other approaches. A key feature of Modena is its use of 'dynamic thresholding', an approach based on 1D score-clustering. This methodology differs substantially from the traditional statistics-based 'hard-thresholds.' We show that this approach is not limited to Modena but has broader applicability. Specifically, when combined with two existing algorithms, 'dynamic thresholding' significantly enhances their performance, resulting in up to a threefold improvement in F1-scores.

Zero-Shot Learning Algorithms for Object Recognition in Medical and Navigation Applications

One category-invariant methodology for object identification is zero-shot learning (ZSL), which uses semantic embeddings to categorize unseen categories. The ZSL method is indispensable in fields where data is scarce, including medical diagnosis and navigation. This paper proposes an Improved ZSL (I-ZSL) framework to increase the object recognition accuracy and generalization for the medical and navigation applications. The proposed framework is a hybrid architecture that uses Variational Autoencoders (VAEs) for robust feature generation and Transformer-based embeddings for semantic alignment. A domain-adaptive classifier, trained through contrastive learning, bridges the gap between the seen and unseen classes. The classifier has identified the framework with minimal training data in medical diagnostics for rare disease diagnosis. The proposed I-ZSL framework achieved a 20% improvement in F1-score over state-of-the-art models. In navigation, it demonstrated 25% better performance in novel landmark recognition under dynamic environmental conditions. These results show the framework's efficiency in addressing domain-specific challenges. This work presents ZSL with great potential to further object recognition in applications with significant impacts.

Enhanced brain tumor diagnosis using combined deep learning models and weight selection technique.

Brain tumor classification is a critical task in medical imaging, as accurate diagnosis directly influences treatment planning and patient outcomes. Traditional methods often fall short in achieving the required precision due to the complex and heterogeneous nature of brain tumors. In this study, we propose an innovative approach to brain tumor multi-classification by leveraging an ensemble learning method that combines advanced deep learning models with an optimal weighting strategy. Our methodology integrates Vision Transformers (ViT) and EfficientNet-V2 models, both renowned for their powerful feature extraction capabilities in medical imaging. This model enhances the feature extraction step by capturing both global and local features, thanks to the combination of different deep learning models with the ViT model. These models are then combined using a weighted ensemble approach, where each model's prediction is assigned a weight. To optimize these weights, we employ a genetic algorithm, which iteratively selects the best weight combinations to maximize classification accuracy. We trained and validated our ensemble model using a well-curated dataset comprising labeled brain MRI images. The model's performance was benchmarked against standalone ViT and EfficientNet-V2 models, as well as other traditional classifiers. The ensemble approach achieved a notable improvement in classification accuracy, precision, recall, and F1-score compared to individual models. Specifically, our model attained an accuracy rate of 95%, significantly outperforming existing methods. This study underscores the potential of combining advanced deep learning models with a genetic algorithm-optimized weighting strategy to tackle complex medical classification tasks. The enhanced diagnostic precision offered by our ensemble model can lead to better-informed clinical decisions, ultimately improving patient outcomes. Furthermore, our approach can be generalized to other medical imaging classification problems, paving the way for broader applications of AI in healthcare. This advancement in brain tumor classification contributes valuable insights to the field of medical AI, supporting the ongoing efforts to integrate advanced computational tools in clinical practice.

Advancing emotion recognition in social media: A novel integration of heterogeneous neural networks with fine-tuned language models

Social media platforms have emerged as crucial sources for emotion analysis, but the issue of non-compliance in labeling by fine-tuned large language models (LLMs) can significantly impact the accuracy of emotion classification. This study addresses this challenge by introducing a novel compliance-driven training set that systematically harmonizes label discrepancies across multiple LLMs, thereby enhancing classification accuracy by over 5% on the non-compliance set. Integrating this compliance set with a Heterogeneous Neural Network (HNN) architecture, we propose a robust framework for emotion classification. Our approach is validated on three diverse datasets, GoEmotion, Friends, and TEC, demonstrating substantial improvements in accuracy, F1 score, and recall over baseline models. These results confirm the effectiveness of our compliance-driven strategy and establish a new benchmark for emotion recognition in social media content. The proposed framework offers a versatile and scalable solution applicable across various languages and platforms, ensuring broad utility in advanced emotion classification tasks.

Slovak morphological tokenizer using the Byte-Pair Encoding algorithm.

This study introduces a new approach to text tokenization, SlovaK Morphological Tokenizer (SKMT), which integrates the morphology of the Slovak language into the training process using the Byte-Pair Encoding (BPE) algorithm. Unlike conventional tokenizers, SKMT focuses on preserving the integrity of word roots in individual tokens, crucial for maintaining lexical meaning. The methodology involves segmenting and extracting word roots from morphological dictionaries and databases, followed by corpus preprocessing and training SKMT alongside a traditional BPE tokenizer. Comparative evaluation against existing tokenizers demonstrates SKMT's outstanding ability to maintain root integrity, achieving 99.7% root integrity compared to SlovakBERT (90.5%) and a pureBPE tokenizer (93.1%). Further validation involved fine-tuning models on a sentiment classification NLP task, where models trained with SKMT achieved an F1-score improvement of 3.5% over those trained with conventional BPE tokenization, followed by a focus on the Semantic Textual Similarity (STS) task. These findings suggest that training language models on the SKMT tokenizer significantly enhances model performance and quality.

Automatic identification of the endangered hawksbill sea turtle behavior using deep learning and cross-species transfer learning.

The accelerometer, an onboard sensor, enables remote monitoring of animal posture and movement, allowing researchers to deduce behaviors. Despite the automated analysis capabilities provided by deep learning, data scarcity remains a challenge in ecology. We explored transfer learning to classify behaviors from acceleration data of critically endangered hawksbill sea turtles (Eretmochelys imbricata). Transfer learning reuses a model trained on one task from a large dataset to solve a related task. We applied this method using a model trained on green turtles (Chelonia mydas) and adapted it to identify hawksbill behaviors such as swimming, resting and feeding. We also compared this with a model trained on human activity data. The results showed an 8% and 4% F1-score improvement with transfer learning from green turtle and human datasets, respectively. Transfer learning allows researchers to adapt existing models to their study species, leveraging deep learning and expanding the use of accelerometers for wildlife monitoring.

Comparative Performance Analysis of Ensemble Models for Breast Cancer Classification

Breast cancer remains a prevalent global health issue, accounting for approximately 2.3 million new cases and 670,000 deaths worldwide in 2022. Early detection and accurate diagnosis are crucial to improving patient outcomes, as delayed identification can lead to severe complications. Advances in machine learning (ML) have facilitated improvements in cancer diagnosis, with various algorithms enhancing predictive accuracy. This study proposes a novel ensemble model for breast cancer classification, utilizing 31 features from the University of Wisconsin Breast Cancer dataset. We applied six algorithms—K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression, Random Forest, Gradient Boosting, and XGBoost—and combined them with ensemble techniques, specifically Hard Voting, to develop a high-accuracy model. The model was evaluated on classification performance metrics, achieving improvements in accuracy, precision, recall, and F1 score. Results indicate that the proposed ensemble model outperforms individual classifiers and other ensembles, showing potential as a reliable tool for early breast cancer detection.

Federated Learning Using Multi-Modal Sensors with Heterogeneous Privacy Sensitivity Levels

Data from multi-modal sensors, such as Red-Green-Blue (RGB) cameras, thermal cameras, microphones, and mmWave radars, have gradually been adopted in various classification problems for better accuracy. Some sensors, like RGB cameras and microphones, however, capture privacy-invasive data, which are less likely to be used in centralized learning. Although the Federated Learning (FL) paradigm frees clients from sharing their sensor data, doing so results in reduced classification accuracy and increased training time. In this article, we introduce a novel Heterogeneous Privacy Federated Learning (HPFL) paradigm to better capitalize on the less privacy-invasive sensor data, such as thermal images and mmWave point clouds, by uploading them to the server for closing the performance gap between FL and centralized learning. HPFL not only allows clients to keep the more privacy-invasive sensor data private, such as RGB images and human voices, but also gives each client total freedom to define the levels of their privacy concern on individual sensor modalities. For example, more sensitive users may prefer to keep their thermal images private, while others do not mind sharing these images. We carry out extensive experiments to evaluate the HPFL paradigm using two representative classification problems: semantic segmentation and emotion recognition. Several key findings demonstrate the merits of HPFL: (i) compared to FedAvg, it improves foreground accuracy by 18.20% in semantic segmentation and boosts the F1-score by 4.20% in emotion recognition, (ii) with heterogeneous privacy concern levels, it achieves an even larger F1-score improvement of 6.17–16.05% in emotion recognition, and (iii) it also outperforms the state-of-the-art FL approaches by 12.04–17.70% in foreground accuracy and 2.54–4.10% in F1-score.

Adaptive Multi-Scale Bayesian Framework for MFL Inspection of Steel Wire Ropes

Magnetic flux leakage (MFL) technology is widely used in steel wire rope (SWR) inspection for non-destructive testing. However, accurate defect characterization requires advanced signal processing techniques to handle complex noise conditions and varying defect types. This paper presents a novel adaptive multi-scale Bayesian framework for MFL signal analysis in SWR inspection. Our approach integrates discrete wavelet transform with adaptive thresholding and multi-scale feature fusion, enabling simultaneous detection of minute defects and large-area corrosion. To validate our method, we implemented a four-channel MFL detection system and conducted extensive experiments on both simulated and real-world datasets. Compared with state-of-the-art methods, including long short-term memory (LSTM), attention mechanisms, and isolation forests, our approach demonstrated significant improvements in precision, recall, and F1 score across various tolerance levels. The proposed method showed superior detection performance, with an average precision of 91%, recall of 89%, and an F1 score of 0.90 in high-noise conditions, surpassing existing techniques. Notably, our method showed superior performance in high-noise environments, reducing false positive rates while maintaining high detection sensitivity. While computational complexity in real-time processing remains a challenge, this study provides a robust solution for non-destructive testing of SWR, potentially improving inspection efficiency and defect localization accuracy. Future work will focus on optimizing algorithmic efficiency and exploring transfer learning techniques for enhanced adaptability across different non-destructive testing (NDT) domains. This research not only advances signal processing and anomaly detection technology but also contributes to enhancing safety and maintenance efficiency in critical infrastructure.

NewsZoom (focus on the important): Text summarization of News Articles based on named entity recognition using Spacy library

The study discusses the importance of summarization in dealing with a large amount of data available on the internet. The study used a deep-learning algorithm based on functions from the spacy library in Python to summarize news articles and evaluated the impact of named entity recognition on the summarization process. The study assessed different datasets from CNN-DailyMail and the BBC (entertainment articles) and found that the proposed method based on named entity recognition showed significant improvement in recall, precision, and F-score compared to the word frequency method. The study also observed that the articles from CNN- DailyMail were longer, with an average of 551 words and 28 sentences, compared to the BBC (entertainment articles), which had an average of 190 words and 12 sentences. The evaluation results showed that the proposed method based onnamed entity recognition performed better on the shorter articles from the BBC, indicating that the method was more effective in summarizing shorter texts. In summary, the study highlighted the importance of summarization in dealing with a large amount of data available on the internet. It showed that named entity recognition can significantly improve the effectiveness of the summarization process. The study alsoobserved that the proposed method was more effective in summarizing shorter texts Keywords: Spacy library, entity recognition, Summarization, Deep-Learning, CNN-DailyMail

Transformer-embedded 1D VGG convolutional neural network for regional landslides detection boosted by multichannel data inputs

Up-to-date studies have proved the effectiveness of Convolutional Neural Networks (CNN) in landslide detection. With the rapid development of Remote Sensing and Geographic Information System technologies, an increasing amount of spectral and non-spectral information is available for CNN modeling. It offering a comprehensive perspective for landslide detection, but also presents challenges to CNNs, especially in efficiently learning long-range feature associations. Therefore, we proposed a novel Transformer-improved VGG network (Trans-VGG). It takes spectral (RGB images) and non-spectral information (elevation, slope, and PCA components) as data inputs and integrating both local and global feature in modeling. The method is tested in two landslide cluster areas in Litang County, China. The results in site a show that the Trans-VGG model demonstrates an improvement in F1-score, ranging from 4% to 21%, compared with the conventional machine learning and CNN models. The validation result in site b further proved the validity of our proposed method.

Generative named entity recognition framework for Chinese legal domain.

Named entity recognition (NER) is a crucial task in natural language processing, particularly challenging in the legal domain due to the intricate and lengthy nature of legal entities. Existing methods often struggle with accurately identifying entity boundaries and types in legal texts. To address these challenges, we propose a novel sequence-to-sequence framework designed specifically for the legal domain. This framework features an entity-type-aware module that leverages contrastive learning to enhance the prediction of entity types. Additionally, we incorporate a decoder with a copy mechanism that accurately identifies complex legal entities without the need for explicit tagging schemas. Our extensive experiments on two legal datasets show that our framework significantly outperforms state-of-the-art methods, achieving notable improvements in precision, recall, and F1 score. This demonstrates the effectiveness of our approach in improving entity recognition in legal texts, offering a promising direction for future research in legal NER.

Communication-efficient federated multi-domain learning for network anomaly detection

Efficient and accurate anomaly detection in a network is of great significance for maintaining network and device security. Most anomaly detection methods assume that different anomalous network data distributions are the same or similar and ignore data privacy preservation. In this paper, a novel Federated Learning (FL) is proposed that it can quickly detect different types of anomalies in Non-Independent and Identically Distributed (Non-IID) data. First, we design a multi-domain machine learning model for multi-domain data, named Aegean, which consists of two modules: an ensemble AutoEncoder (AE) and a Generative Adversarial Network (GAN). Second, because data from different domains are non-IID, we model the anomaly detection problem as a dual problem, which can be recast as a robust optimization problem. The robust optimization problem is non-convex and therefore difficult to solve. As a remedy, we formulate and solve a dual problem by taking the Lagrangian dual function of the original problem. Experiments demonstrate that Aegean significantly outperforms the current state-of-the-art methods, with a 16% F1 score improvement over that of a One-Class Support Vector Machine (OCSVM). The designed FL significantly reduces the communication overhead of FedAvg without sacrificing anomaly detection performance.

Railway Tracks Extraction from High Resolution Unmanned Aerial Vehicle Images Using Improved NL-LinkNet Network

The accurate detection of railway tracks from unmanned aerial vehicle (UAV) images is essential for intelligent railway inspection and the development of electronic railway maps. Traditional computer vision algorithms struggle with the complexities of high-precision track extraction due to challenges such as diverse track shapes, varying angles, and complex background information in UAV images. While deep learning neural networks have shown promise in this domain, they still face limitations in precisely extracting track line edges. To address these challenges, this paper introduces an improved NL-LinkNet network, named NL-LinkNet-SSR, designed specifically for railway track detection. The proposed NL-LinkNet-SSR integrates a Sobel edge detection module and a SimAM attention module to enhance the model’s accuracy and robustness. The Sobel edge detection module effectively captures the edge information of track lines, improving the segmentation and extraction of target edges. Meanwhile, the parameter-free SimAM attention module adaptively emphasizes significant features while suppressing irrelevant information, broadening the model’s perceptual field and improving its responsiveness to target areas. Experimental results show that the NL-LinkNet-SSR significantly outperforms the original NL-LinkNet model across multiple key metrics, including a more than 0.022 increase in accuracy, over a 4% improvement in F1-score, and a more than 3.5% rise in mean Intersection over Union (mIoU). These enhancements suggest that the improved NL-LinkNet-SSR offers a more reliable solution for railway track detection, advancing the field of intelligent railway inspection.

Multi-Head Attention Refiner for Multi-View 3D Reconstruction.

Traditional 3D reconstruction models have consistently faced the challenge of balancing high recall of object edges with maintaining a high precision. In this paper, we introduce a post-processing method, the Multi-Head Attention Refiner (MA-R), designed to address this issue by integrating a multi-head attention mechanism into the U-Net style refiner module. Our method demonstrates improved capability in capturing intricate image details, leading to significant enhancements in boundary predictions and recall rates. In our experiments, the proposed approach notably improves the reconstruction performance of Pix2Vox++ when multiple images are used as the input. Specifically, with 20-view images, our method achieves an IoU score of 0.730, a 1.1% improvement over the 0.719 of Pix2Vox++, and a 2.1% improvement in F-Score, achieving 0.483 compared to 0.462 of Pix2Vox++. These results underscore the robustness of our approach in enhancing both precision and recall in 3D reconstruction tasks involving multiple views.