Search Results (25)

Background/Objectives: Artificial intelligence (AI) models for glaucoma screening using colour fundus photography have shown strong internal performance; however, their external validity and calibration reliability remain uncertain. This study developed a controlled four-dataset benchmark to evaluate six glaucoma-screening models across RIM-ONE, DRISHTI-GS, the Hillel Yaffe Glaucoma Dataset, and ORIGA. Methods: The evaluated models included a hybrid deep-handcrafted random forest (RF), transfer-learning and semi-supervised VGG16 models, and a compact convolutional neural network (CNN). Performance was assessed in within-dataset and cross-dataset settings using discrimination metrics, including accuracy, area under the receiver operating characteristic curve (AUC), balanced accuracy, and Matthews correlation coefficient (MCC), as well as calibration metrics, including Brier score and expected calibration error (ECE). Threshold stability, preprocessing ablation, and repeated-seed analyses were also performed. Results: Within-dataset evaluation showed strong discrimination, with the complete hybrid CNN + histogram of oriented gradients (HOG) + local binary patterns (LBP) + minimum redundancy maximum relevance (mRMR) + RF pipeline achieving a mean accuracy of 0.8697 and an AUC of 0.8972. However, cross-dataset performance was poor, with a mean accuracy of 0.5685 and an AUC of 0.5798. The denoising autoencoder-enhanced transfer-learning model showed improved probability calibration in transfer settings. Preprocessing effects were dataset-dependent, with raw images, region of interest (ROI) cropping, and Retinex normalisation producing different external performance patterns. Conclusions: High internal accuracy did not translate into reliable cross-domain generalisation. None of the evaluated models were suitable for zero-shot deployment without site-specific validation or recalibration. Full article

Purpose: This study investigates how financial and entrepreneurial capability configurations differentiate internationalized from domestically oriented Thai SMEs. By adopting a decision-science perspective, the study identifies the capability patterns that shape firms’ strategic international orientation. Design/methodology/approach: This research, which utilizes survey and financial data from 179 Thai SMEs (2021–2023), employs an inductive machine learning approach based on Extreme Gradient Boosting (XGBoost). The analytical framework integrates profitability, liquidity, leverage, operational efficiency, international experience, team readiness, market knowledge, and institutional connectivity. Feature importance scores and confirmatory statistical tests are used to validate differentiating capability structures. Findings: Internationalized SMEs are characterized by stronger financial agility, higher profitability, disciplined leverage, and more efficient resource utilization, together with entrepreneurial preparedness reflected in international experience, risk tolerance, team coordination, and institutional embeddedness. Localized SMEs, by contrast, display liquidity-heavy but lower-dynamism profiles, weaker network engagement, and more limited organizational readiness. Practical implications: The findings suggest that SME support should move beyond finance-only assistance and address capability alignment. Programs that strengthen financial agility, team readiness, international market knowledge, digital channels, and institutional networks may improve SME readiness for international engagement. Originality/value: The study contributes to international business theory by reframing SME internationalization as a selective and configurational process rather than a linear or binary outcome. It extends the resource-based view by emphasizing capability bundles rather than isolated resources and contributes to the dynamic capabilities perspective by showing how financial, entrepreneurial, and institutional capabilities are associated with SMEs’ differentiated participation in international markets. Full article

Reliable anomaly early warning for hydropower station sump wells remains challenging due to the strong nonlinearity of water level dynamics and the limited adaptability of conventional fixed-threshold alarms. Here, we present a hybrid deep learning framework—termed CNN–Attention–LSTM–BO—that fuses multi-scale local feature extraction, adaptive temporal weighting, and sequential dependency modeling within a unified architecture, with all critical hyperparameters tuned via Bayesian optimization. A four-dimensional input representation is first constructed from the raw water level signal and its first- and second-order differences together with the drainage pump operating state, capturing both trend and transient information. One-dimensional convolutions at multiple kernel scales encode short-range fluctuation patterns, a Bahdanau-style temporal attention layer selectively amplifies informative time steps, and a stacked LSTM propagates long-horizon risk dependencies. At the decision stage, a dual dynamic thresholding scheme couples an improved $3\sigma$ criterion with kernel density estimation (KDE) to partition the smoothed risk score into three graded alert levels (normal/warning/critical), replacing the binary alarm paradigm. Experiments on the SWaT benchmark yield an average area under the ROC curve (AUC) of 0.9246, an average Accuracy of 0.8812, and a best single-well false alarm rate (FAR) of 3.21% (Well-4), with an average FAR of 8.97% across three wells, outperforming both traditional limit-value alarms and ablated variants lacking CNN or attention modules. Full article

China possesses abundant marine fishery resources, which play a vital role in the national economy. Achieving rapid and high-precision classification of underwater targets in complex aquatic environments is of significant importance for enhancing aquaculture intelligence and operational efficiency. To address the challenges of insufficient feature extraction and inefficient classifier parameter optimization in underwater image classification, this study proposes a classification method integrating local binary patterns (LBP), kernel principal component analysis (KPCA), and an improved sparrow search algorithm (SSA). The method first extracts image texture features using LBP and then applies KPCA for nonlinear dimensionality reduction. Subsequently, three optimization strategies—dynamic weighting, boundary contraction, and adaptive mutation—are introduced to enhance SSA, which is then employed to optimize the core parameters of the Support Vector Machine (SVM). Experiments were conducted on an underwater image dataset containing four types of targets: sea urchins, fish, rocks, and scallops. The results demonstrate that, compared with the traditional KPCA-SVM method, the integration of LBP features and the improved SSA increases classification accuracy from 55% to 94.37%, validating the effectiveness of the proposed approach in extracting underwater image features and optimizing classifier parameters. This provides technical support for improving the feasibility of automatic underwater target recognition in aquaculture applications. Full article

Background/Objectives: Parkinson’s disease (PD) is a progressive neurodegenerative disorder in ageing populations, yet early diagnosis before motor symptoms remains critical. Reliable identification of subtle nigral alterations at early stages of the disease on magnetic resonance imaging (MRI) remains challenging. This limitation is mainly attributed to the subjective and low sensitivity of manual image interpretation in early PD. Here, we demonstrate a deep learning-based framework to enhance early PD detection. The study’s novelty is a lightweight deep learning framework that captures spatial, textural, and frequency-domain PD biomarkers without heavy network architectures or manual region delineation. Methods: The model integrates GhostNet with ensemble learning to combine local and global spatial information. This model employs wavelet-based frequency feature extraction rather than downsampling and incorporates an attention module to focus on relevant image regions, particularly changes in the substantia nigra (SN) region. Segmentation is employed solely as an auxiliary intermediate step to localize the SN and guide discriminative feature extraction. The final output is a binary classification that distinguishes PD patients from healthy controls. T2-weighted MRI data from the PPMI database are employed. Results: The proposed model achieved an F1-score of 0.8762, demonstrating robust performance under class imbalance, outperforming state-of-the-art models with only 2.03 million parameters and 4.36 Giga Floating Point Operations (GFLOPs). The architecture uncovered texture and frequency patterns previously inaccessible with conventional CNN pipelines. Model comparisons demonstrated consistent gains across all evaluated metrics (all p < 0.001), establishing robust diagnostic improvement. Conclusions: These findings establish an efficient, high-performing framework for reliable MRI-based PD identification. The approach provides automated early detection and supports clinically scalable, computationally lightweight screening tools. Full article

Face recognition technology utilizes unique facial features to analyze and compare individuals for identification and verification purposes. This technology is crucial for several reasons, such as improving security and authentication, effectively verifying identities, providing personalized user experiences, and automating various operations, including attendance monitoring, access management, and law enforcement activities. In this paper, comprehensive evaluations are conducted using different face detection and modality segmentation methods, feature extraction methods, and classifiers to improve system performance. As for face detection, four methods are proposed: OpenCV’s Haar Cascade classifier, Dlib’s HOG + SVM frontal face detector, Dlib’s CNN face detector, and Mediapipe’s face detector. Additionally, two types of feature extraction techniques are proposed: hand-crafted features (traditional methods: global local features) and deep learning features. Three global features were extracted, Scale-Invariant Feature Transform (SIFT), Speeded Robust Features (SURF), and Global Image Structure (GIST). Likewise, the following local feature methods are utilized: Local Binary Pattern (LBP), Weber local descriptor (WLD), and Histogram of Oriented Gradients (HOG). On the other hand, the deep learning-based features fall into two categories: convolutional neural networks (CNNs), including VGG16, VGG19, and VGG-Face, and Siamese neural networks (SNNs), which generate face embeddings. For classification, three methods are employed: Support Vector Machine (SVM), a one-class SVM variant, and Multilayer Perceptron (MLP). The system is evaluated on three datasets: in-house, Labelled Faces in the Wild (LFW), and the Pins dataset (sourced from Pinterest) providing comprehensive benchmark comparisons for facial recognition research. The best performance accuracy for the proposed ten-feature extraction methods applied to the in-house database in the context of the facial recognition task achieved 99.8% accuracy by using the VGG16 model combined with the SVM classifier. Full article

Background/Objectives: Age-related macular degeneration (AMD) is the leading cause of visual impairment among the elderly. Optical coherence tomography angiography (OCTA) is a non-invasive imaging modality that enables detailed visualisation of retinal vascular layers. However, clinical assessment of OCTA images is often challenging due to high data volume, pattern variability, and subtle abnormalities. This study aimed to develop automated algorithms to detect and quantify AMD in OCTA images, thereby reducing ophthalmologists’ workload and enhancing diagnostic accuracy. Methods: Two texture-based algorithms were developed to classify OCTA images without relying on segmentation. The first algorithm used whole local texture features, while the second applied principal component analysis ($PCA$) to decorrelate and reduce texture features. Local texture descriptors, including rotation-invariant uniform local binary patterns ($LB{P}^{2riu}$), local binary patterns ($LBP$), and binary robust independent elementary features ($BRIEF$), were combined with machine learning classifiers such as support vector machine (SVM) and K-nearest neighbour (KNN). OCTA datasets from Manchester Royal Eye Hospital and Moorfields Eye Hospital, covering healthy, dry AMD, and wet AMD eyes, were used for evaluation. Results: The first algorithm achieved a mean area under the receiver operating characteristic curve (AUC) of $1.00\pm 0.00$ for distinguishing healthy eyes from wet AMD. The second algorithm showed superior performance in differentiating dry AMD from wet AMD (AUC $0.85\pm 0.02$). Conclusions: The proposed algorithms demonstrate strong potential for rapid and accurate AMD diagnosis in OCTA workflows. By reducing manual image evaluation and associated variability, they may support improved clinical decision-making and patient care. Full article

Emotion analysis based on electroencephalogram (EEG) sensors is pivotal for human–machine interaction yet faces key challenges in spatio-temporal feature fusion and cross-band and brain-region integration from multi-channel sensor-derived signals. This paper proposes MB-MSTFNet, a novel framework for EEG emotion recognition. The model constructs a 3D tensor to encode band–space–time correlations of sensor data, explicitly modeling frequency-domain dynamics and spatial distributions of EEG sensors across brain regions. A multi-scale CNN-Inception module extracts hierarchical spatial features via diverse convolutional kernels and pooling operations, capturing localized sensor activations and global brain network interactions. Bi-directional GRUs (BiGRUs) model temporal dependencies in sensor time-series, adept at capturing long-range dynamic patterns. Multi-head self-attention highlights critical time windows and brain regions by assigning adaptive weights to relevant sensor channels, suppressing noise from non-contributory electrodes. Experiments on the DEAP dataset, containing multi-channel EEG sensor recordings, show that MB-MSTFNet achieves 96.80 ± 0.92% valence accuracy, 98.02 ± 0.76% arousal accuracy for binary classification tasks, and 92.85 ± 1.45% accuracy for four-class classification. Ablation studies validate that feature fusion, bidirectional temporal modeling, and multi-scale mechanisms significantly enhance performance by improving feature complementarity. This sensor-driven framework advances affective computing by integrating spatio-temporal dynamics and multi-band interactions of EEG sensor signals, enabling efficient real-time emotion recognition. Full article

Facial emotion recognition (FER) is vital for improving human–machine interactions, serving as the foundation for AI systems that integrate cognitive and emotional intelligence. This helps bridge the gap between mechanical processes and human emotions, enhancing machine engagement with humans. Considering the constraints of low hardware specifications often encountered in real-world applications, this study leverages recent advances in deep learning to propose an enhanced model for FER. The model effectively utilizes texture information from faces through Gabor and Local Binary Pattern (LBP) feature extraction techniques. By integrating these features into a specially modified AlexNet architecture, our approach not only classifies facial emotions more accurately but also demonstrates significant improvements in performance and adaptability under various operational conditions. To validate the effectiveness of our proposed model, we conducted evaluations using the FER2013 and RAF-DB benchmark datasets, where it achieved impressive accuracies of 98.10% and 93.34% for the two datasets, with standard deviations of 1.63% and 3.62%, respectively. On the FER-2013 dataset, the model attained a precision of 98.2%, a recall of 97.9%, and an F1-score of 98.0%. Meanwhile, for the other dataset, it achieved a precision of 93.54%, a recall of 93.12%, and an F1-score of 93.34%. These results underscore the model’s robustness and its capability to deliver high-precision emotion recognition, making it an ideal solution for deployment in environments where hardware limitations are a critical concern. Full article

In response to the challenges of bearing fault diagnosis under small sample sizes and variable operating conditions, this paper proposes a novel method based on the two-dimensional analysis of vibration acceleration signals and a Multi-Task Conditional Generative Adversarial Network (MTC-GAN). This method first constructs two-dimensional images of vibration signals by leveraging the physical properties of the bearing acceleration signals and employs Local Binary Patterns (LBP) to extract subtle texture features from these images, thereby generating fault feature signatures with high discriminative power across different operating conditions. Subsequently, MTC-GAN is utilized for data augmentation, and the trained discriminator is used to perform fault classification tasks, improving classification accuracy under conditions with small sample sizes. Experimental results demonstrate that the proposed method achieves excellent fault diagnosis accuracy and robustness under both small sample sizes and varying operating conditions. Compared to traditional methods, this approach exhibits higher efficiency and reliability in handling complex operating conditions and data scarcity. Full article

In order to solve the problems of a high mismatching rate and being easily affected by noise and gray transformation, an improved product defect detection method combining centroid distance and textural information is proposed in this paper. Based on image preprocessing, the improved fuzzy C-means clustering method is used to extract the closed contour features. Then, the contour center distance description operator is used for bidirectional matching, and a robust coarse matching contour pair is obtained. After the coarse matching contour pair is screened, the refined matching result is obtained by using the improved local binary pattern operator. Finally, by comparing whether the number of fine matching pairs is consistent with the number of template outlines, the detection of good and bad industrial products is realized, and the closed contour extraction experiment, the anti-rotation matching experiment, the anti-gray difference matching experiment, and the defect detection experiment of three different products are designed. The experimental results show that the improved product defect detection method has good performance in relation to anti-rotation transformation and anti-gray difference, the detection accuracy can reach more than 90%, and the detection time is up to 362.6 ms, which can meet the requirements of industrial real-time detection. Full article

The exponential increase in waste generation is a significant global challenge with serious implications. Addressing this issue necessitates the enhancement of waste management processes. This study introduces a method that improves waste separation by integrating learning models at various levels. The method begins with the creation of image features as a new matrix using the Multi-Scale Local Binary Pattern (MLBP) technique. This technique optimally represents features and patterns across different scales. Following this, an ensemble model at the first level merges two Convolutional Neural Network (CNN) models, with each model performing the detection operation independently. A second-level CNN model is then employed to obtain the final output. This model uses the information from the first-level models and combines these features to perform a more accurate detection. The study’s novelty lies in the use of a second-level CNN model in the proposed ensemble system for fusing the results obtained from the first level, replacing conventional methods such as voting and averaging. Additionally, the study employs an MLBP feature selection approach for a more accurate description of the HW image features. It uses the Simulated Annealing (SA) algorithm for fine-tuning the hyperparameters of the CNN models, thereby optimizing the system’s performance. Based on the accuracy metric, the proposed method achieved an accuracy of 99.01% on the TrashNet dataset and 99.41% on the HGCD dataset. These results indicate a minimum improvement of 0.48% and 0.36%, respectively, compared to the other methods evaluated in this study. Full article

Satellite remote sensing provides a potential technology for detecting fog at dawn and dusk on a large scale. However, the spectral characteristics of fog at dawn and dusk are similar to those of the ground surface, which makes satellite-based fog detection difficult. With the aid of time-series datasets from the Himawari-8 (H8)/AHI, this study proposed a novel algorithm of the self-adaptive threshold of visual background extractor (ST-ViBe) model for satellite fog detection at dawn and dusk. Methodologically, the background model was first built using the difference between MIR and TIR (BTD) and the local binary similarity patterns (LBSP) operator. Second, BTD and scale invariant local ternary pattern (SILTP) texture features were coupled to form scene factors, and the detection threshold of each pixel was determined adaptively to eliminate the influence of the solar zenith angles. The background model was updated rapidly by accelerating the updating rate and increasing the updating quantity. Finally, the residual clouds were removed with the traditional cloud removal method to achieve accurate detection of fog at dawn and dusk over a large area. The validation results demonstrated that the ST-ViBe algorithm could detect fog at dawn and dusk precisely, and on a large scale. The probability of detection, false alarm ratio, and critical success index were 72.5%, 18.5%, 62.4% at dawn (8:00) and 70.6%, 33.6%, 52.3% at dusk (17:00), respectively. Meanwhile, the algorithm mitigated the limitations of the traditional algorithms, such as illumination mutation, missing detection, and residual shadow. The results of this study could guide satellite fog detection at dawn and dusk and improve the detection of similar targets. Full article

This paper investigates the use of an unsupervised hybrid statistical–local outlier factor algorithm to detect anomalies in time-series flight data. Flight data analysis is an activity carried out by airlines primarily as a means of improving the safety and operation of their fleet. Traditionally, this is performed by checking exceedances in pre-set limits to the flight data parameters. However, this method highlights single events during a flight, making this analysis laborious. The process also fails to establish trends or reflect potential unknown hazards. This research took advantage of machine learning techniques to recognize patterns in large datasets by implementing the local outlier factor (LOF). In order to minimize human input, a statistical approach was adopted to establish the threshold value above which the flights are considered to be anomalous and interpret the scores. This paper shows that LOF quantifies the degree of outlier-ness of an outlier rather than binary categorizing a point into inlier or outlier, as in the case of clustering algorithms. Thus, with LOF, for the first time, we demonstrated that in the aviation industry, anomalous flights could not only be identified but also be given an anomaly score to compare two anomalous flights in an unsupervised manner. Furthermore, LOF helps to track anomalous behavior in time during the flight. This is insightful when a flight is abnormal, only for some seconds or short duration. For the first time, we attempted to detect flight parameters responsible for anomalous behavior or at least give direction to human experts looking for the cause of abnormal behavior. This was all analyzed with real-life flight data in an unsupervised manner in contrast to simulated data. Full article

The usage of images in different fields has increased dramatically, especially in medical image analysis and social media. Many risks can threaten the integrity and confidentiality of digital images transmitted through the internet. As such, the preservation of the contents of these images is of the utmost importance for sensitive healthcare systems. In this paper, the researchers propose a block-based approach to protect the integrity of digital images by detecting and localizing forgeries. It employs a visual cryptography-based watermarking approach to provide the capabilities of forgery detection and localization. In this watermarking scheme, features and key and secret shares are generated. The feature share is constructed by extracting features from equal-sized blocks of the image by using a Walsh transform, a local binary pattern and a discrete wavelet transform. Then, the key share is generated randomly from each image block, and the secret share is constructed by applying the XOR operation between the watermark, feature share and key share. The CASIA V 1.0 and SIPI datasets were used to check the performance and robustness of the proposed method. The experimental results from these datasets revealed that the percentages of the precision, recall and F1 score classification indicators were approximately 97% for these indicators, while the percentages of the TAF and NC image quality indicators were approximately 97% and 96% after applying several known image processing and geometric attacks. Furthermore, the comparative experimental results with the state-of-art approaches proved the robustness and noticeable improvement in the proposed approach for the detection and localization of image forgeries in terms of classification and quality measures. Full article