Search Results (25)

Unexpected machine failures in industrial environments lead to high maintenance costs, unplanned downtime, and safety risks. This study proposes a proactive predictive model using a hybrid of eXtreme Gradient Boosting (XGBoost) and Neural Networks (NN) to forecast machine failures. A synthetic dataset capturing recent breakdown history and time since last failure was used to simulate industrial scenarios. To address class imbalance, SMOTE and class weighting were applied, alongside a focal loss function to emphasize difficult-to-classify failures. The XGBoost model was tuned via GridSearchCV, while the NN model utilized ReLU-activated hidden layers with dropout. Evaluation using stratified 5-fold cross-validation showed that the NN achieved an F1-score of 0.7199 and a recall of 0.9545 for the minority class. XGBoost attained a higher PR AUC of 0.7126 and a more balanced precision–recall trade-off. Sample predictions demonstrated strong recall (100%) for failures, but also a high false positive rate, with most prediction probabilities clustered between 0.50–0.55. Additional benchmarking against Logistic Regression, Random Forest, and SVM further confirmed the superiority of the proposed hybrid model. Model interpretability was enhanced using SHAP and LIME, confirming that recent breakdowns and time since last failure were key predictors. While the model effectively detects failures, further improvements in feature engineering and threshold tuning are recommended to reduce false alarms and boost decision confidence. Full article

Low-altitude drones, which are unimpeded by traffic congestion or urban terrain, have become a critical asset in emergency rescue missions. To address the current lack of emergency rescue data, UAV aerial videos were collected to create an experimental dataset for action classification and localization annotation. A total of 5082 keyframes were labeled with 1–5 targets each, and 14,412 instances of data were prepared (including flight altitude and camera angles) for action classification and position annotation. To mitigate the challenges posed by high-resolution drone footage with excessive redundant information, we propose the SlowFast-Traffic (SF-T) framework, a spatio-temporal sequence-based algorithm for recognizing traffic accident rescue actions. For more efficient extraction of target–background correlation features, we introduce the Actor-Centric Relation Network (ACRN) module, which employs temporal max pooling to enhance the time-dimensional features of static backgrounds, significantly reducing redundancy-induced interference. Additionally, smaller ROI feature map outputs are adopted to boost computational speed. To tackle class imbalance in incident samples, we integrate a Class-Balanced Focal Loss (CB-Focal Loss) function, effectively resolving rare-action recognition in specific rescue scenarios. We replace the original Faster R-CNN with YOLOX-s to improve the target detection rate. On our proposed dataset, the SF-T model achieves a mean average precision (mAP) of 83.9%, which is 8.5% higher than that of the standard SlowFast architecture while maintaining a processing speed of 34.9 tasks/s. Both accuracy-related metrics and computational efficiency are substantially improved. The proposed method demonstrates strong robustness and real-time analysis capabilities for modern traffic rescue action recognition. Full article

To address the challenges of poor instance segmentation accuracy, real-time performance trade-offs, high miss rates, and imprecise edge localization in tomato grading and harvesting robots operating in complex scenarios (e.g., dense growth, occluded fruits, and dynamic viewing conditions), an accurate, efficient, and robust visual instance segmentation network is urgently needed. This paper proposes TQVGModel (Tomato Quality Visual Grading Model), a Mask RCNN-based instance segmentation network for tomato quality grading. First, TQVGModel employs a multi-branch IncepConvV2 backbone, reconstructed via ConvNeXt architecture and large-kernel convolution decomposition, to enhance instance segmentation accuracy while maintaining real-time performance. Second, the Class Balanced Focal Loss is adopted in the classification branch to prioritize sparse or challenging classes, reducing the miss rates in complex scenes. Third, an Enhanced Sobel (E-Sobel) operator integrates boundary prediction with an edge loss function, improving edge localization precision for quality assessment. Additionally, a quality grading subsystem is designed to automate tomato evaluation, supporting subsequent harvesting and growth monitoring. A high-quality benchmark dataset, Tomato-Seg, is constructed for complex-scene tomato instance segmentation. Experiments show that the TQVGModel-Tiny variant achieves an 80.05% mAP (7.04% higher than Mask R-CNN), with 33.98 M parameters (10.2 M fewer) and 53.38 ms inference speed (16.6 ms faster). These results demonstrate TQVGModel’s high accuracy, real-time capability, reduced miss rates, and precise edge localization, providing a theoretical foundation for tomato grading and harvesting in complex environments. Full article

Traffic accident prediction is essential for improving road safety and optimizing intelligent transportation systems. However, deep learning models often struggle with distribution shifts and class imbalance, leading to degraded performance in real-world applications. While distribution shift is a common challenge in machine learning, Transformer-based models—despite their ability to capture long-term dependencies—often lack mechanisms for dynamic adaptation during inferencing. In this paper, we propose a TTT-Enhanced Transformer that incorporates Test-Time Training (TTT), enabling the model to refine its parameters during inferencing through a self-supervised auxiliary task. To further boost performance, an Adaptive Memory Layer (AML), a Feature Pyramid Network (FPN), Class-Balanced Attention (CBA), and Focal Loss are integrated to address multi-scale, long-term, and imbalance-related challenges. Our experimental results show that our model achieved an overall accuracy of 96.86% and a severe accident recall of 95.8%, outperforming the strongest Transformer baseline by 5.65% in accuracy and 9.6% in recall. The results of our confusion matrix and ROC analyses confirm our model’s superior classification balance and discriminatory power. These findings highlight the potential of our approach in enhancing real-time adaptability and robustness under shifting data distributions and class imbalances in intelligent transportation systems. Full article

Transmission-line defect detection is crucial for grid operation. Existing methods struggle to balance defect localization and fine segmentation. Therefore, this study proposes a novel cascaded two-stage framework that first utilizes YOLOv5s for the global localization of defective regions, and then uses U-Net for the fine segmentation of candidate regions. To improve the segmentation performance, U-Net adopts a transfer learning strategy based on the VGG16 pretrained model to alleviate the impact of limited dataset size on the training effect. Meanwhile, a hybrid loss function that combines Dice Loss and Focal Loss is designed to solve the small-target and class imbalance problems. This method integrates target detection and fine segmentation, enhancing detection precision and improving the extraction of detailed damage features. Experiments on the self-constructed dataset show that the method achieves 87% mAP on YOLOv5s, 88% U-Net damage recognition precision, a mean Dice coefficient of 93.66%, and 89% mIoU, demonstrating its effectiveness in accurately detecting transmission-line defects and efficiently segmenting the damage region, providing assistance for the intelligent operation and maintenance of transmission lines. Full article

Probability calibration and decision threshold selection are fundamental aspects of risk prediction and classification, respectively. A strictly proper loss function is used in clinical risk prediction applications to encourage a model to predict calibrated class-posterior probabilities or risks. Recent studies have shown that training with focal loss can improve the discriminatory power of gradient-boosted decision trees (GBDT) for classification tasks with an imbalanced or skewed class distribution. However, the focal loss function is not a strictly proper loss function. Therefore, the output of GBDT trained using focal loss is not an accurate estimate of the true class-posterior probability. This study aims to address the issue of poor calibration of GBDT trained using focal loss in the context of clinical risk prediction applications. The methodology utilizes a closed-form transformation of the confidence scores of GBDT trained with focal loss to estimate calibrated risks. The closed-form transformation relates the focal loss minimizer and the true-class posterior probability. Algorithms based on Bayesian hyperparameter optimization are provided to choose the focal loss parameter that optimizes discriminatory power and calibration, as measured by the Brier score metric. We assess how the calibration of the confidence scores affects the selection of a decision threshold to optimize the balanced accuracy, defined as the arithmetic mean of sensitivity and specificity. The effectiveness of the proposed strategy was evaluated using lung transplant data extracted from the Scientific Registry of Transplant Recipients (SRTR) for predicting post-transplant cancer. The proposed strategy was also evaluated using data from the Behavioral Risk Factor Surveillance System (BRFSS) for predicting diabetes status. Probability calibration plots, calibration slope and intercept, and the Brier score show that the approach improves calibration while maintaining the same discriminatory power according to the area under the receiver operating characteristics curve (AUROC) and the H-measure. The calibrated focal-aware XGBoost achieved an AUROC, Brier score, and calibration slope of 0.700, 0.128, and 0.968 for predicting the 10-year cancer risk, respectively. The miscalibrated focal-aware XGBoost achieved equal AUROC but a worse Brier score and calibration slope (0.140 and 1.579). The proposed method compared favorably to the standard XGBoost trained using cross-entropy loss (AUROC of 0.755 versus 0.736 in predicting the 1-year risk of cancer). Comparable performance was observed with other risk prediction models in the diabetes prediction task. Full article

In this work, we propose a multi-modal speaker change detection (SCD) approach with focal loss, which integrates both audio and text features to enhance detection performance. The proposed approach utilizes pre-trained large-scale models for feature extraction and incorporates a self-attention mechanism to optimize useful features related to speaker change. The extracted features are fused and processed through a fully connected classification network, with layer normalization and dropout for stability and generalization. To address class imbalance, we apply focal loss, which reduces errors for the difficult samples, leading to better balanced performance. Extensive experiments on a multi-talker meeting dataset demonstrate that the proposed multi-modal approach consistently outperforms single-modal models, proving the complementary nature of audio and text for SCD. Fine-tuning pre-trained models (Wav2Vec2 and Bert) for audio and text significantly boosts accuracy, achieving a 21% improvement over frozen models. The self-attention mechanism further improves performance by 2%, highlighting its ability to capture speaker transition cues effectively. Additionally, focal loss enhances the model’s performance, making it more robust to imbalanced data. Full article

Shadows in remote sensing images often introduce challenges in accurate segmentation due to their variability in shape, size, and texture. To address these issues, this study proposes the Supervised Dynamic Kernel U-Net (SDKU-Net), a novel architecture designed to enhance shadow detection in complex remote sensing scenarios. SDKU-Net integrates dynamic kernel adjustment, a combined loss function incorporating Focal and Tversky Loss, and optimizer switching to effectively tackle class imbalance and improve segmentation quality. Using the AISD dataset, the proposed method achieved state-of-the-art performance with an Intersection over Union (IoU) of 0.8552, an F1-Score of 0.9219, an Overall Accuracy (OA) of 96.50%, and a Balanced Error Rate (BER) of 5.08%. Comparative analyses demonstrate SDKU-Net’s superior performance against established methods such as U-Net, U-Net++, MSASDNet, and CADDN. Additionally, the model’s efficient training process, requiring only 75 epochs, highlights its potential for resource-constrained applications. These results underscore the robustness and adaptability of SDKU-Net, paving the way for advancements in shadow detection and segmentation across diverse fields. Full article

Background: Ovarian cancer encompasses a diverse range of neoplasms originating in the ovaries, fallopian tubes, and peritoneum. Despite being one of the commonest gynaecological malignancies, there are no validated screening strategies for early detection. A diagnosis typically relies on imaging, biomarkers, and multidisciplinary team discussions. The accurate interpretation of CTs and MRIs may be challenging, especially in borderline cases. This study proposes a methodological pipeline to develop and evaluate deep learning (DL) models that can assist in classifying ovarian masses from CT and MRI data, potentially improving diagnostic confidence and patient outcomes. Methods: A multi-institutional retrospective dataset was compiled, supplemented by external data from the Cancer Genome Atlas. Two classification workflows were examined: (1) whole-volume input and (2) lesion-focused region of interest. Multiple DL architectures, including ResNet, DenseNet, transformer-based UNeST, and Attention Multiple-Instance Learning (MIL), were implemented within the PyTorch-based MONAI framework. The class imbalance was mitigated using focal loss, oversampling, and dynamic class weighting. The hyperparameters were optimised with Optuna, and balanced accuracy was the primary metric. Results: For a preliminary dataset, the proposed framework demonstrated feasibility for the multi-class classification of ovarian masses. The initial experiments highlighted the potential of transformers and MIL for identifying the relevant imaging features. Conclusions: A reproducible methodological pipeline for DL-based ovarian mass classification using CT and MRI scans has been established. Future work will leverage a multi-institutional dataset to refine these models, aiming to enhance clinical workflows and improve patient outcomes. Full article

Traditional methods for pest recognition have certain limitations in addressing the challenges posed by diverse pest species, varying sizes, diverse morphologies, and complex field backgrounds, resulting in a lower recognition accuracy. To overcome these limitations, this paper proposes a novel pest recognition method based on attention mechanism and multi-scale feature fusion (AM-MSFF). By combining the advantages of attention mechanism and multi-scale feature fusion, this method significantly improves the accuracy of pest recognition. Firstly, we introduce the relation-aware global attention (RGA) module to adaptively adjust the feature weights of each position, thereby focusing more on the regions relevant to pests and reducing the background interference. Then, we propose the multi-scale feature fusion (MSFF) module to fuse feature maps from different scales, which better captures the subtle differences and the overall shape features in pest images. Moreover, we introduce generalized-mean pooling (GeMP) to more accurately extract feature information from pest images and better distinguish different pest categories. In terms of the loss function, this study proposes an improved focal loss (FL), known as balanced focal loss (BFL), as a replacement for cross-entropy loss. This improvement aims to address the common issue of class imbalance in pest datasets, thereby enhancing the recognition accuracy of pest identification models. To evaluate the performance of the AM-MSFF model, we conduct experiments on two publicly available pest datasets (IP102 and D0). Extensive experiments demonstrate that our proposed AM-MSFF outperforms most state-of-the-art methods. On the IP102 dataset, the accuracy reaches 72.64%, while on the D0 dataset, it reaches 99.05%. Full article

With the growth of data in the real world, datasets often encounter the problem of long-tailed distribution of class sample sizes. In long-tailed image recognition, existing solutions usually adopt a class rebalancing strategy, such as reweighting based on the effective sample size of each class, which leans towards common classes in terms of higher accuracy. However, increasing the accuracy of rare classes while maintaining the accuracy of common classes is the key to solving the problem of long-tailed image recognition. This research explores a direction that balances the accuracy of both common and rare classes simultaneously. Firstly, a two-stage training is adopted, motivated by the use of transfer learning to balance features of common and rare classes. Secondly, a balanced weight function called Balanced Focal Softmax (BFS) loss is proposed, which combines balanced softmax loss focusing on common classes with balanced focal loss focusing on rare classes to achieve dual balance in long-tailed image recognition. Subsequently, a Balanced Weight Learning Mechanism (BWLM) to further utilize the feature of weight decay is proposed, where the weight decay as the weight balancing technique for the BFS loss tends to make the model learn smaller balanced weights by punishing the larger weights. Through extensive experiments on five long-tailed image datasets, it proves that transferring the weights from the first stage to the second stage can alleviate the bias of the naive models toward common classes. The proposed BWLM not only balances the weights of common and rare classes, but also greatly improves the accuracy of long-tailed image recognition and outperforms many state-of-the-art algorithms. Full article

Image datasets acquired from orchards are commonly characterized by intricate backgrounds and an imbalanced distribution of disease categories, resulting in suboptimal recognition outcomes when attempting to identify apple leaf diseases. In this regard, we propose a novel apple leaf disease recognition model, named RFCA ResNet, equipped with a dual attention mechanism and multi-scale feature extraction capacity, to more effectively tackle these issues. The dual attention mechanism incorporated into RFCA ResNet is a potent tool for mitigating the detrimental effects of complex backdrops on recognition outcomes. Additionally, by utilizing the class balance technique in conjunction with focal loss, the adverse effects of an unbalanced dataset on classification accuracy can be effectively minimized. The RFB module enables us to expand the receptive field and achieve multi-scale feature extraction, both of which are critical for the superior performance of RFCA ResNet. Experimental results demonstrate that RFCA ResNet significantly outperforms the standard CNN network model, exhibiting marked improvements of 89.61%, 56.66%, 72.76%, and 58.77% in terms of accuracy rate, precision rate, recall rate, and F1 score, respectively. It is better than other approaches, performs well in generalization, and has some theoretical relevance and practical value. Full article

This paper proposes a new Hepatocellular Carcinoma (HCC) classification method utilizing a hyperspectral imaging system (HSI) integrated with a light microscope. Using our custom imaging system, we have captured 270 bands of hyperspectral images of healthy and cancer tissue samples with HCC diagnosis from a liver microarray slide. Convolutional Neural Networks with 3D convolutions (3D-CNN) have been used to build an accurate classification model. With the help of 3D convolutions, spectral and spatial features within the hyperspectral cube are incorporated to train a strong classifier. Unlike 2D convolutions, 3D convolutions take the spectral dimension into account while automatically collecting distinctive features during the CNN training stage. As a result, we have avoided manual feature engineering on hyperspectral data and proposed a compact method for HSI medical applications. Moreover, the focal loss function, utilized as a CNN cost function, enables our model to tackle the class imbalance problem residing in the dataset effectively. The focal loss function emphasizes the hard examples to learn and prevents overfitting due to the lack of inter-class balancing. Our empirical results demonstrate the superiority of hyperspectral data over RGB data for liver cancer tissue classification. We have observed that increased spectral dimension results in higher classification accuracy. Both spectral and spatial features are essential in training an accurate learner for cancer tissue classification. Full article

With the rapid evolution of mobile communication networks, the number of subscribers and their communication practices is increasing dramatically worldwide. However, fraudsters are also sniffing out the benefits. Detecting fraudsters from the massive volume of call detail records (CDR) in mobile communication networks has become an important yet challenging topic. Fortunately, Graph neural network (GNN) brings new possibilities for telecom fraud detection. However, the presence of the graph imbalance and GNN oversmoothing problems makes fraudster detection unsatisfactory. To address these problems, we propose a new fraud detector. First, we transform the user features with the help of a multilayer perceptron. Then, a reinforcement learning-based neighbor sampling strategy is designed to balance the number of neighbors of different classes of users. Next, we perform user feature aggregation using GNN. Finally, we innovatively treat the above augmented GNN as weak classifier and integrate multiple weak classifiers using the AdaBoost algorithm. A balanced focal loss function is also used to monitor the model training error. Extensive experiments are conducted on two open real-world telecom fraud datasets, and the results show that the proposed method is significantly effective for the graph imbalance problem and the oversmoothing problem in telecom fraud detection. Full article

Deep learning-related technologies have achieved remarkable success in the field of intelligent fault diagnosis. Nevertheless, the traditional intelligent diagnosis methods are often based on the premise of sufficient annotation signals and balanced distribution of classes, and the model structure is so complex that it requires huge computational resources. To this end, a lightweight class imbalanced diagnosis framework based on a depthwise separable Laplace-wavelet convolution network with variable-asymmetric focal loss (DSLWCN-VAFL) is established. Firstly, a branch with few parameters for time-frequency feature extraction is designed by integrating wavelet and depthwise separable convolution. It is combined with the branch of regular convolution that fully learns time-domain features to jointly capture abundant discriminative features from limited samples. Subsequently, a new asymmetric soft-threshold loss, VAFL, is designed, which reasonably rebalances the contributions of distinct samples during the model training. Finally, experiments are conducted on the data of bearing and gearbox, which demonstrate the superiority of the DSLWCN-VAFL algorithm and its lightweight diagnostic framework in handling class imbalanced data. Full article