Search Results (1,722)

Class-incremental learning allows a model to continuously acquire new classes from sequentially arriving data while preserving its ability to recognize previously learned ones, which is essential for improving adaptability and supporting long-term evolution. However, the class arrival order is inherently random, and highly similar classes may appear consecutively, which intensifies catastrophic forgetting. Although replay-based methods can effectively alleviate this problem, they usually require storing or accessing historical raw samples, which introduces additional data-retention and storage burdens. To address these challenges, this paper proposes ORACIL, an Order-Robust Analytic Class-Incremental Learning framework. First, ORACIL constructs a conflict graph based on class centroids and dynamically partitions newly arriving classes into multiple low-similarity groups, thereby reducing inter-class interference and mitigating forgetting. Second, for each class group, it trains an analytic incremental classification head and performs recursive closed-form updates for the analytic heads using current-stage data and accumulated second-order statistics, without replaying raw historical samples. For group recognition, ORACIL uses feature-derived distance representations rather than raw historical images, making the incremental process raw-sample-free with respect to original image replay. Third, during inference, the group probabilities generated by the group-recognition router are softly fused with the class scores produced by each analytic head, and the class with the highest fused probability is selected as the final prediction. Extensive experiments on CIFAR-100, CUB200, and OmniBenchmark demonstrate the effectiveness of ORACIL. Without replaying historical images, ORACIL achieves final-phase average forgetting rates of 0.16%, 0.77%, and 1.04%, and final-phase accuracies of 95.77%, 93.86%, and 88.12%, respectively. In addition, the MOPD and AOPD results show that ORACIL maintains strong robustness under different class arrival orders. Full article

Deep learning (DL)-based percussion methods in concrete-filled steel-tube (CFST) void detection have gained much attention. However, the detection signal contains a large amount of noise, which affects the accuracy of qualitative and quantitative analyses of the subsequent detection results. To improve the signal-to-noise ratio (SNR) during percussion detection, this study proposes a CFST void detection method using the good point set and vibrational snow ablation optimizer (GVSAO) algorithm and dual-channel parallel convolutional neural networks (CNNs). The proposed method employs the gram angle field (GAF) to transform percussive sound signals into images. It then constructs a dual-channel parallel CNN structure, where the GAF is decomposed into the following two maps: the gram angle sum field (GASF) and the gram angle difference field (GADF). These maps are simultaneously fed into the CNN for training. The outputs from the two channels are concatenated and fused. Finally, the GVSAO algorithm was used for model optimization to improve convergence speed and recognition accuracy. Both the temporal and spatial characteristics of the knocking sound signal are fully preserved, while the interference of different construction noises is effectively avoided. Validation experiments were conducted on CFST specimens with different heights of voids (0, 50, 100, and 150 mm) under different pressure loads. The original sample dataset and the signal-enhanced dataset were obtained by adding background noise with different SNRs. The test results show that the prediction accuracies on the original signal dataset are consistently above 98.74%. Among them, the accuracy achieves 100% at pressure loads of 0 and 50 tons. Additionally, the prediction accuracies on the signal-enhanced dataset are all above 97.2%, indicating that the model maintains a high level of classification performance. This suggests that the model can effectively suppress noise and exhibits excellent robustness. Full article

Deep learning methods that integrate convolutional neural networks (CNNs) and Transformers have achieved remarkable progress in hyperspectral image (HSI) classification. However, existing methods still suffer from insufficient multi-scale spatial–spectral feature modeling, a lack of efficient interaction mechanisms between local and global features, and the inherent high computational complexity and redundant information of Transformers, which limit model performance. To address these issues, a Multi-Scale Spatial–Spectral Convolutional Hybrid Transformer model (MS³CHFormer) is proposed in this article. Specifically, a Multi-Scale Spatial–Spectral Convolution Module (MS³ConvM) is first constructed. Through a multi-branch and multi-receptive-field design, it jointly models spatial and spectral features at different scales, thereby enhancing the representation capability of complex ground objects. Then, a Token-Selective Sparse Transformer Encoder (TSSTE) is designed, which adaptively selects tokens and performs sparse modeling via a Dynamic Correlation-Aware Attention (DCAA) mechanism, effectively reducing computational complexity while suppressing redundant information and further reinforcing key feature representations. Furthermore, a Local–Global Feature Fusion Module (LGFFM) is designed to achieve deep complementary fusion of CNN and Transformer features by mapping them into different representation spaces. Finally, a Detail-Preserving Enhancement Module (DPEM) introduces original detail information through residual connections to compensate for detail loss in high-level semantic representations, thereby enhancing the representation capability of boundaries and fine-grained structures. Experiments and comparative analyses on four public HSI datasets demonstrate that the proposed MS³CHFormer outperforms state-of-the-art methods and achieves superior classification accuracy under limited training samples, exhibiting excellent robustness and generalization ability. Full article

Artificial intelligence-assisted food detection is increasingly moving from single-task classification toward integrated analytical systems capable of producing multiple quality-related outputs from one sensing workflow. However, most hyperspectral food detection studies still rely on a single spectral range or simple feature concatenation, which limits their ability to exploit complementary physicochemical information from heterogeneous sensors. In this study, an artificial intelligence-enabled visible–near-infrared and short-wave infrared (VNIR-SWIR) hyperspectral fusion framework is proposed for multi-task fruit detection, using origin authentication and bruise localization as representative tasks. The proposed method first constructs an observation-consistent fused representation from high-resolution VNIR images and low-resolution SWIR images. Collaborative spectral unmixing is used to couple cross-modal material distributions, while abundance-consistency and downsampled observation-consistency constraints are introduced to estimate SWIR-informed features on the VNIR spatial grid without assuming measured high-resolution SWIR ground truth. The fused representation is then processed by a shared spectral–spatial deep encoder with two task-specific heads: a fruit-level classification head for origin authentication and a pixel-level segmentation head for bruise detection. Experiments on apple and kiwifruit datasets show that the proposed framework outperforms VNIR-only, SWIR-only, bicubic-fusion, CNMF-style fusion, and TV-regularized fusion baselines under five fruit-level stratified random splits. For origin-category authentication, the proposed method achieved an accuracy of almost 93.85 for apples and almost 94.35 for kiwifruit. For bruise localization, the proposed method achieved higher overall accuracy, average accuracy, and Cohen’s kappa across the evaluated fruit categories. Full article

Accurate grade detection in graphite ore, which is a strategic and critical mineral resource, plays an important role in improving beneficiation efficiency and overall resource utilization. However, the scarcity of high-grade samples limits the performance of deep learning models in grade identification tasks. This limitation makes it difficult for models to learn stable and representative features. This paper proposes an enhanced CycleGAN-based image augmentation framework designed for graphite ore imagery. The method works within an unpaired image translation architecture. It introduces a distributed reflectance consistency loss. This loss encodes the graphite ore’s typical low reflectance and high optical contrast as explicit statistical constraints. The design enforces consistency in both the intensity distribution and the textural structure of the generated images. The model further integrates a convolutional block attention module into the generator. This module helps refine feature representation under a physics-inspired heuristic. The study constructs augmented training sets using the proposed method. It then evaluates these datasets with a downstream grade classification model. Experimental results show clear improvements. The method reduces Fréchet Inception Distance by 21.9% and Kernel Inception Distance by 39.4%. It also improves peak signal-to-noise ratio by 3.3% and structural similarity index measure by 2.6% compared with the baseline CycleGAN. The classification accuracy in the grade identification task increases by about 2.3 percentage points. These results show that the proposed method improves both the perceptual quality and the statistical consistency of synthetic graphite ore images. It also helps reduce the performance drop caused by limited training data in few-shot learning conditions. Full article

X-ray testing is a cornerstone nondestructive testing (NDT) technique in the nondestructive testing of welds. To address the challenges posed by minute defects such as cracks and pinholes—characterized by small size, weak features, and a tendency to be confused with noise—this paper proposes a minute defect recognition framework based on sparse representation. (1) Median filtering was selected as the basic denoising method. In combination with image enhancement, the discriminability of weld regions and defect features was improved. (2) A segmented ROI extraction method combining Otsu threshold segmentation and Sobel edge detection was proposed. This method can better adapt to inclined or curved weld images and effectively reduce background interference. (3) A micro-defect recognition method based on sparse representation was proposed. By constructing an SDR and combining dictionary learning with sparse solving models, effective representation and classification of micro-defect regions were achieved. Its effectiveness and engineering application value were verified through actual engineering data, third-party witness tests, and competition results. Full article

Non-intrusive load monitoring (NILM) is essential for smart grid demand-side management and energy conservation, yet existing methods suffer from limited feature discrimination, ambiguous identification of similar electrical appliances, and difficulty balancing model accuracy and lightweight deployment. To address these issues, this paper proposes a dual-branch lightweight load identification method fusing steady-state features and lightweight network. Firstly, V-I trajectory images are generated via standardized transformation and two-dimensional histogram logarithmic mapping, while steady-state characteristics, including active power, reactive power, trajectory area and intermediate section slope, are extracted. Then, a dual-branch network is constructed, where the visual branch adopts depthwise separable convolution and lightweight multi-head attention to mine global trajectory features, and the numerical branch uses fully connected layers to encode steady-state features; feature concatenation fusion is adopted to complete appliance classification. The experimental results on the Plug Load Appliance Identification Dataset (PLAID dataset) show that the proposed method achieves a recognition accuracy of 95.35% with only 0.17M parameters, outperforming standard and medium convolutional neural network (CNN) models. Ablation experiments verify that steady-state feature fusion effectively improves the identification accuracy of easily confused and small-sample loads. The proposed method realizes high-precision and lightweight load identification, which is suitable for edge deployment in smart meters and has practical application value for intelligent power management. Full article

The accurate classification of seabed sediment and benthic covers in shallow-water environments remains a key challenge for marine activities and oceanographic research. However, coastal areas of shallow waters are influenced by complex dynamic environments, making it difficult to obtain authentic sediment and benthic-cover samples. Therefore, to address the problem of few-shot classification of seabed sediment and benthic covers, a few-shot classification algorithm of seabed sediment and benthic covers based on the fusion model of airborne LiDAR bathymetry (ALB) and multispectral images is proposed in this article. Based on the extracted features, a scale-invariant feature transform-progressive sample consensus (SIFT-PROSAC) algorithm and perspective transform model were constructed to achieve feature fusion. Then, multi-modal feature selection is realized using a formal concept analysis-Relief-F (FCA-Relief-F) algorithm. Finally, a graph attention network-prototype network (GAT-PN) model was established to classify five types of sediment and benthic cover (coral reef, stone, sand, vegetation, and coastal zone). To validate the effectiveness of the proposed method, experimental data from actual measurements at Ganquan Island in the Xisha Islands of China were used. Compared to other classical classifiers, the GAT-PN algorithm achieves a higher classification accuracy, with an overall accuracy (OA) and Kappa coefficient of 97.50% and 0.97, respectively. The findings of this study provide effective technical support for marine engineering and related fields. Full article

Railway scene understanding is critical for ensuring train operational safety and advancing intelligent railway systems. Existing railway vision methods mainly focus on perception and classification, while lacking regulation-guided semantic reasoning capabilities in complex environments. To address these limitations, this paper proposes a retrieval-augmented generation (RAG)-enhanced vision–language framework for railway signal cognition and safety reasoning. The proposed method integrates railway signal perception, regulatory knowledge retrieval, and multi-modal reasoning to improve factual consistency, reasoning reliability, and operational interpretability. In addition, a dedicated railway signal dataset comprising 500 standardized railway scene images with structured QA annotations is constructed to support regulation-oriented multi-modal recognition evaluation. Experimental results show that the proposed framework improves reasoning accuracy from 28.40% to 67.20% with an average end-to-end inference latency of 11.31 s per sample, and the inference speed can be further improved by adjusting experimental configurations to trade off between efficiency and accuracy, demonstrating the potential of RAG-enhanced architectures as a foundational step toward reliable multi-modal cognition in intelligent railway systems. Full article

Objective: The objective of this study was to construct a large language model (LLM) for the Response Evaluation Criteria in Solid Tumors (RECIST) assessment using exclusively longitudinal radiology report text. Methods: This study included 258 patients with solid tumors, encompassing 2065 longitudinal CT/MRI examination time points. We developed WiNGPT-32B, an open-source and locally deployable LLM, by infusing it with domain-specific medical knowledge and optimizing it via knowledge distillation, using GPT-4 as the teacher model. Central to its architecture is the Chained Task Execution (CTE) framework, which structures RECIST assessment into four modular components: lesion diameter extraction, sum of longest diameter computation, tumor response classification, and report generation. Model performance (accuracy, recall, precision, and F1 score) was benchmarked against GPT-4 and a single radiologist, utilizing the consensus of three independent radiologists as the reference standard. Results: The number of patients with imaging time points was 212 (82.2%) with 4–10, 36 (13.9%) with 11–20, and 10 (3.9%) with >20 time points. For target lesions, the successful extraction rate of WiNGPT-32B was 0.934 (95% CI: 0.922–0.944), which was slightly higher than that of GPT-4 0.920 (0.907–0.931; p = 0.083). In five-category RECIST classification (complete response, partial response, stable disease, progressive disease, and not evaluable), WiNGPT-32B achieved an overall accuracy of 0.805 (0.786–0.823), significantly higher than GPT-4 (0.699, 0.678–0.720; p < 0.001) but lower than the radiologist (0.915, 0.901–0.928; p < 0.001). For progressive disease, WiNGPT-32B had an F1 score of 0.841 (0.813–0.870), significantly outperforming GPT-4’s 0.755 (0.720–0.790), and approaching the radiologist’s 0.922 (0.902–0.942). Conclusions: WiNGPT-32B demonstrates the feasibility of a text-only, open-source LLM with the CTE framework for longitudinal RECIST assessment, with promising performance in detecting disease progression. Full article

Accurate cultivar identification of commercial dried goji berries is essential for raw material sorting, batch consistency assessment, and quality control during processing and distribution. Conventional approaches based on manual judgment or physicochemical analysis are often subjective, labor-intensive, time-consuming, and costly, making them unsuitable for rapid commercial sorting and quality inspection. To develop a rapid, low-cost, and nondestructive method for dried goji berry cultivar identification, this study proposes a visual recognition framework that integrates RGB imaging with lightweight deep learning. A dataset comprising 25,899 RGB images from five cultivars of commercial dried goji berry samples, namely Ningqi No. 7, Linqi No. 5, Ningqi No. 1, Keqi 6082, and Jingqi No. 1, was constructed. Given the pronounced surface shrinkage, complex texture, and subtle inter-cultivar appearance differences of dried goji berries, an image quality enhancement method was designed to strengthen the representation of color gradation, textural details, and edge information. For model development, CoAtNet was selected as the baseline network and redesigned for lightweight deployment. By integrating an improved feature extraction module and an information-preserving downsampling module, the proposed LSH-CoAtNet model enhances fine-grained feature representation while reducing computational cost. On the quality-enhanced image dataset, the proposed method achieved an accuracy of 98.80%, a precision of 98.81%, a recall of 98.80%, and an F1-score of 98.80%. The model contained only 6.41 M parameters and required 1.60 GFLOPs, outperforming the baseline model in both classification performance and computational efficiency. Ablation experiments and five-fold cross-validation further confirmed the effectiveness of the image quality enhancement strategy, the contribution of each improved module, and the stability of the model. Overall, the proposed method, which combines RGB image quality enhancement with LSH-CoAtNet, provides a low-cost, nondestructive, and efficient technical solution for rapid cultivar identification, raw material sorting, batch consistency assessment, and quality control of commercial dried goji berries during processing and distribution. It may also serve as a reference for intelligent classification and quality inspection of other specialty dried horticultural products. Full article

Melon (Cucumis melo L.) is a globally significant horticultural crop, characterized by high nutritional value and substantial commercial status. However, frequent outbreaks of powdery mildew severely threaten its yield and fruit quality. Current early detection methods primarily focus on detached leaf assays, which often lack sufficient model generalization. This study proposes a temporal 3D multispectral point cloud reconstruction method for melon plants by integrating multispectral imaging with 3D reconstruction technology. An Artificial Neural Network (ANN) model for 3D spatial light field distribution was developed based on a hemispherical white reference to achieve precise reflectance calibration of the multispectral point clouds. Post-calibration, the coefficient of variation (CV) for the spectral reflectance of the hemispherical reference in 3D space was reduced to less than 2.4%. On this basis, an early classification model for melon powdery mildew was constructed using Partial Least Squares Discriminant Analysis (PLS-DA) based on the mean reflectance spectra of individual plant point clouds. The results demonstrate that the average recognition accuracy reaches 85.94% from 4 days post-inoculation onwards, enabling disease early warning three days in advance. This research provides critical theoretical support and technical reference for the non-destructive early monitoring and precision smart plant protection of crops in facility agriculture. Full article

This study aims to develop eco-efficient ceramic tiles through the valorization of recycled glass (GW; soda–lime glass cullet) as a partial raw material substituent, enabling a reduction in sintering temperature and, consequently, a decrease in thermal energy demand, carbon-equivalent emissions, and the depletion of virgin mineral resources. Ceramic tiles were elaborated by partially substituting natural bentonite with 30–50 wt.% GW and fired at 900 °C and 950 °C. Use of GW promoted liquid-phase sintering, driving significant densification evidenced by a marked reduction in open porosity and water absorption. SEM images confirm a denser, more homogeneous structure with reduced porosity, leading to improved mechanical strength and chemical durability. Compositions containing 30–35 wt.% bentonite exhibit the most optimized microstructure, characterized by well-dispersed crystalline phases embedded within a dense vitreous matrix. These findings demonstrate that high-performance ceramic tiles meeting standard classification thresholds can be manufactured at sub-1000 °C firing temperatures through judicious incorporation of recycled glass waste. This approach offers a viable pathway toward reduced energy consumption, diminished reliance on primary mineral resources, and enhanced circularity within the construction ceramics industry. Full article

Background/Objectives: Preoperative differentiation among benign thyroid nodules, follicular variant papillary thyroid carcinoma (FV-PTC), and classical papillary thyroid carcinoma (C-PTC) remains clinically challenging. FV-PTC is particularly difficult to identify due to its substantial sonographic and cytological overlap with both benign nodules and other malignant subtypes, frequently resulting in overtreatment or delayed diagnosis. This study aimed to develop and validate an interpretable multimodal model for accurate three-class discrimination using routine ultrasound images, with a specific focus on improving the preoperative identification of FV-PTC. Methods: This retrospective study included 479 pathologically confirmed thyroid nodules from 462 patients. Conventional ultrasound features and radiomics features extracted from grayscale ultrasound and color Doppler flow imaging were used to construct three predictive models: a Conventional Ultrasound model (conventional ultrasound features only), a Radiomics model (radiomics features only), and an Integrated model (combined features). Each model was trained using four machine learning classifiers. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) analysis, and clinical usefulness was evaluated using decision curve analysis (DCA). Results: The support vector machine (SVM)-based Integrated Model achieved the best overall performance. In the independent testing cohort, the AUCs were 0.853 for FV-PTC, 0.882 for C-PTC and 0.928 for benign nodules. The Integrated Model showed the greatest improvement for FV-PTC, with a ΔAUC of 0.141 compared with the Conventional Ultrasound Model. SHAP (SHapley Additive exPlanations) analysis identified wavelet-HL_gldm_Dependence and wavelet-HH_glcm_InverseVariance as the two most important radiomics predictors in both the Radiomics Model and the Integrated Model, demonstrating robust cross-model stability and high discriminative power. Conclusions: The SVM-based Integrated Model demonstrated promising performance for three-class classification of thyroid nodules and enhanced the preoperative identification of FV-PTC. This approach may provide an interpretable and noninvasive decision-support tool for refining subtype-specific risk stratification and supporting individualized clinical management. Full article

Maritime search and rescue is an important component of emergency response frameworks and primarily relies on Unmanned Aerial Vehicles (UAVs) for maritime object detection. However, maritime accidents frequently occur in low-visibility environments, such as foggy or low-light conditions, which lead to low contrast, blurred object boundaries, and degraded texture representations. Most existing maritime object detection algorithms are developed for natural light scenes, and their performance deteriorates markedly when deployed directly in low-visibility environments, primarily due to reduced image quality that hinders feature extraction and semantic information aggregation. Although several studies incorporate image enhancement techniques prior to detection to improve image quality, these approaches often introduce significant additional computational overhead, limiting their practical deployment on UAV platforms. To tackle these challenges, this paper proposes a lightweight model built upon a recent YOLO framework, termed Multi-Scale Adaptive YOLO (MSA-YOLO), for maritime detection using UAVs in low-visibility environments. The proposed model systematically optimizes the backbone, neck, and detection head networks. Specifically, an improved StarNet backbone is designed by integrating Efficient Channel Attention (ECA) mechanisms and multi-scale convolutional kernels, which strengthen feature extraction capability while maintaining low computational overhead. In the neck network, a high-frequency enhanced residual block branch is inserted into the C3k2 module to capture richer detailed information, while depthwise separable convolution is utilized to further reduce computational cost. Moreover, a non-parametric attention module is incorporated into the detection head to adaptively optimize features in the classification and regression branches. Finally, a joint loss function that combines bounding box regression, classification, and distribution focal losses is utilized to improve detection accuracy and training stability. Experimental results on the constructed AFO, Zhoushan Island, and Shandong Province datasets demonstrate that, relative to YOLOv11-s, MSA-YOLO reduces model parameters and FLOPs by 52.07% and 41.36%, respectively, while achieving improvements of 1.11% and 1.33% in mAP@0.5:0.95 and mAP@0.5. These results indicate that the proposed method effectively balances computational efficiency and detection accuracy, rendering it suitable for practical maritime search and rescue applications in low-visibility environments. Full article