Search Results (65)

Objectives: We propose a flow-signal-based classification of retinal artery macroaneurysms (RAMs) using Optical Coherence Tomography Angiography (OCTA) and compare the findings with fundus fluorescein angiography (FFA). Methods: A retrospective review of 49 RAM cases observed over 6 years (October 2017–March 2023) at a medical retina clinic at the University Hospital Southampton, UK. Electronic clinical records, FFA, and OCTA images (en face and B-scan) were reviewed to identify pathology and assess RAM flow profiles. Results: In total, 30 eyes from 30 patients were included. The mean age of the patients was 76 years (range 49–91), with 17 females and 13 males. All eyes underwent OCTA, enabling classification of RAMs into three flow signal types: high (9 eyes), low (10 eyes), and absent (9 eyes), while 2 eyes had haemorrhage-related artefacts. A subgroup of 13 eyes also underwent FFA, allowing direct comparison, which showed flow profiles similar to those of OCTA: high (4 eyes), low (6 eyes), and absent (2 eyes), with 1 ungradable case due to subretinal haemorrhage masking. A discrepancy in flow was observed in one case where FFA indicated flow, but OCTA did not. Despite this, FFA and OCTA generally agreed on the flow levels, with a Spearman correlation of r = 0.79 (p = 0.004). Conclusions: OCTA flow profiles were directly comparable to FFA. OCTA effectively identified different levels of blood flow signal behaviour in RAMs. The proposed flow-based RAM classification may aid in prognosis, treatment indications, follow-up, and safe repeat imaging in clinical practice without systemic risk to the patient. Full article

Tidal flats, as critical transitional ecosystems between land and sea, face significant threats from climate change and human activities, necessitating accurate monitoring for conservation and management. However, publicly available tidal flat datasets exhibit substantial discrepancies due to variations in data sources, spectral indices, and classification methods. This study systematically evaluates six widely used 2020 tidal flat datasets (GTF30, GWL_FCS30, MTWM-TP, DCTF, CTF, and TFMC) across China’s coastal zone, assessing their spatial consistency, area estimation differences, and edge classification accuracy. Using a novel edge validation point set (3150 samples) derived from tide gauge stations and low-tide imagery, we demonstrate that MTWM-TP (OA = 0.85) and TFMC (OA = 0.84) achieve the highest accuracy, while DCTF and GTF30 show systematic underestimation and overestimation, respectively. Spatial agreement is strongest in Jiangsu (49.8% unanimous pixels) but weak in turbid estuaries (e.g., Zhejiang). Key methodological divergences include sensor resolution (Sentinel-2 outperforms Landsat in low-tide coverage), spectral index selection (mNDWI reduces false positives in turbid waters), and boundary constraints (high-tide masks suppress inland misclassification). We propose establishing an automated multi-source framework integrating optical (Sentinel-2, Landsat) and radar (Sentinel-1) observation data to enhance low-tide coverage, constructing region-adaptive spectral indices and improving boundary accuracy through the combination of machine learning and thresholding algorithms. This study provides a critical benchmark for dataset selection and methodological advancements in coastal remote sensing. Full article

Facial expression recognition (FER), applied in fields such as interaction and intelligent security, has seen widespread development with the advancement of machine vision technology. However, in natural environments, faces are often obscured by masks, posture, and body parts, leading to incomplete features, which results in poor accuracy of existing facial expression recognition algorithms. Apart from extreme scenarios where facial features are completely blocked, the key information of facial expression features is mostly preserved in most cases, yet insufficient parsing of these features leads to poor recognition results. To address this, we propose a novel joint learning framework that integrates explicit occlusion parsing and feature enhancement. Our model consists of three core modules: a Facial Occlusion Parsing Module (FOPM) for real-time occlusion estimation, an Expression Feature Fusion Module (EFFM) for integrating appearance and geometric features, and a Facial Expression Recognition Module (FERM) for final classification. Extensive experiments under a rigorous and reproducible protocol demonstrate significant improvements of our approach. On the masked facial expression datasets RAF-DB and FER+, our model achieves accuracies of 91.24% and 90.18%, surpassing previous state-of-the-art methods by 2.62% and 0.96%, respectively. Additional evaluation on a real-world masked dataset with diverse mask types further confirms the robustness and generalizability of our method, where it attains an accuracy of 89.75%. Moreover, the model maintains high computational efficiency with an inference time of 12.4 ms per image. By effectively parsing and integrating partially obscured facial features, our approach enables more accurate and robust expression recognition, which is essential for real-world applications in interaction and intelligent security systems. Full article

Agricultural sustainability is becoming increasingly critical in the face of climate change and resource scarcity. This study presents an innovative method for maize yield estimation, integrating remote sensing data and machine learning techniques to promote sustainable agricultural development. By combining Sentinel-2 optical imagery and Sentinel-1 radar data, accurate maize classification masks were created, and the Weighted Least Squares (WLS) model achieved a coefficient of determination (R²) of 0.89 and a root mean square error (RMSE) of 12.8%. Additionally, this study demonstrates the significant role of water and fertilizer optimization in enhancing agricultural sustainability, with water usage reduced by up to 14.76% in Wuwei and 10.23% in Zhangye, and nitrogen application reduced by 5.5% and 8.5%, respectively, while maintaining stable yields. This integrated approach not only increases productivity and reduces resource waste, but it also promotes environmentally friendly and efficient resource use, supporting sustainable agriculture in water-scarce regions. Full article

Cantonese automatic speech recognition (ASR) faces persistent challenges due to its nine lexical tones, extensive phonological variation, and the scarcity of professionally transcribed corpora. To address these issues, we propose a lightweight and data-efficient framework that leverages weak phonetic supervision (WPS) in conjunction with two pho-neme-aware augmentation strategies. (1) Dynamic Boundary-Aligned Phoneme Dropout progressively removes entire IPA segments according to a curriculum schedule, simulating real-world phenomena such as elision, lenition, and tonal drift while ensuring training stability. (2) Phoneme-Aware SpecAugment confines all time- and frequency-masking operations within phoneme boundaries and prioritizes high-attention regions, thereby preserving intra-phonemic contours and formant integrity. Built on the Whistle encoder—which integrates a Conformer backbone, Connectionist Temporal Classification–Conditional Random Field (CTC-CRF) alignment, and a multi-lingual phonetic space—the approach requires only a grapheme-to-phoneme lexicon and Montreal Forced Aligner outputs, without any additional manual labeling. Experiments on the Cantonese subset of Common Voice demonstrate consistent gains: Dynamic Dropout alone reduces phoneme error rate (PER) from 17.8% to 16.7% with 50 h of speech and 16.4% to 15.1% with 100 h, while the combination of the two augmentations further lowers PER to 15.9%/14.4%. These results confirm that structure-aware phoneme-level perturbations provide an effective and low-cost solution for building robust Cantonese ASR systems under low-resource conditions. Full article

Accurate classification of chest X-ray (CXR) images is crucial for diagnosing lung diseases in medical imaging. Existing deep learning models for CXR image classification face challenges in distinguishing non-lung features. In this work, we propose a new segmentation-assisted fusion-based classification method. The method involves two stages: first, we use a lightweight segmentation model, Partial Convolutional Segmentation Network (PCSNet) designed based on an encoder–decoder architecture, to accurately obtain lung masks from CXR images. Then, a fusion of the masked CXR image with the original image enables classification using the improved lightweight ShuffleNetV2 model. The proposed method is trained and evaluated on segmentation datasets including the Montgomery County Dataset (MC) and Shenzhen Hospital Dataset (SH), and classification datasets such as Chest X-Ray Images for Pneumonia (CXIP) and COVIDx. Compared with seven segmentation models (U-Net, Attention-Net, SegNet, FPNNet, DANet, DMNet, and SETR), five classification models (ResNet34, ResNet50, DenseNet121, Swin-Transforms, and ShuffleNetV2), and state-of-the-art methods, our PCSNet model achieved high segmentation performance on CXR images. Compared to the state-of-the-art Attention-Net model, the accuracy of PCSNet increased by 0.19% (98.94% vs. 98.75%), and the boundary accuracy improved by 0.3% (97.86% vs. 97.56%), while requiring 62% fewer parameters. For pneumonia classification using the CXIP dataset, the proposed strategy outperforms the current best model by 0.14% in accuracy (98.55% vs. 98.41%). For COVID-19 classification with the COVIDx dataset, the model reached an accuracy of 97.50%, the absolute improvement in accuracy compared to CovXNet was 0.1%, and clinical metrics demonstrate more significant gains: specificity increased from 94.7% to 99.5%. These results highlight the model’s effectiveness in medical image analysis, demonstrating clinically meaningful improvements over state-of-the-art approaches. Full article

The high nutritional and medicinal value of apples has contributed to their widespread cultivation worldwide. Unfavorable factors in the healthy growth of trees and extensive orchard work are threatening the profitability of apples. This study reviewed deep learning combined with computer vision for monitoring apple tree growth and fruit production processes in the past seven years. Three types of deep learning models were used for real-time target recognition tasks: detection models including You Only Look Once (YOLO) and faster region-based convolutional network (Faster R-CNN); classification models including Alex network (AlexNet) and residual network (ResNet); segmentation models including segmentation network (SegNet), and mask regional convolutional neural network (Mask R-CNN). These models have been successfully applied to detect pests and diseases (located on leaves, fruits, and trunks), organ growth (including fruits, apple blossoms, and branches), yield, and post-harvest fruit defects. This study introduced deep learning and computer vision methods, outlined in the current research on these methods for apple tree growth and fruit production. The advantages and disadvantages of deep learning were discussed, and the difficulties faced and future trends were summarized. It is believed that this research is important for the construction of smart apple orchards. Full article

Skin cancer is a class of disorder defined by the growth of abnormal cells on the body. Accurately identifying and diagnosing skin lesions is quite difficult because skin malignancies share many common characteristics and a wide range of morphologies. To face this challenge, deep learning algorithms have been proposed. Deep learning algorithms have shown diagnostic efficacy comparable to dermatologists in the discipline of images-based skin lesion diagnosis in recent research articles. This work proposes a novel deep learning algorithm to detect skin cancer. The proposed CAD-Skin system detects and classifies skin lesions using deep convolutional neural networks and autoencoders to improve the classification efficiency of skin cancer. The CAD-Skin system was designed and developed by the use of the modern preprocessing approach, which is a combination of multi-scale retinex, gamma correction, unsharp masking, and contrast-limited adaptive histogram equalization. In this work, we have implemented a data augmentation strategy to deal with unbalanced datasets. This step improves the model’s resilience to different pigmented skin conditions and avoids overfitting. Additionally, a Quantum Support Vector Machine (QSVM) algorithm is integrated for final-stage classification. Our proposed CAD-Skin enhances category recognition for different skin disease severities, including actinic keratosis, malignant melanoma, and other skin cancers. The proposed system was tested using the PAD-UFES-20-Modified, ISIC-2018, and ISIC-2019 datasets. The system reached accuracy rates of 98%, 99%, and 99%, consecutively, which is higher than state-of-the-art work in the literature. The minimum accuracy achieved for certain skin disorder diseases reached 97.43%. Our research study demonstrates that the proposed CAD-Skin provides precise diagnosis and timely detection of skin abnormalities, diversifying options for doctors and enhancing patient satisfaction during medical practice. Full article

Segmentation neural networks are widely used in medical imaging to identify anomalies that may impact patient health. Despite their effectiveness, these networks face significant challenges, including the need for extensive annotated patient data, time-consuming manual segmentation processes and restricted data access due to privacy concerns. In contrast, classification neural networks, similar to segmentation neural networks, capture essential parameters for identifying objects during training. This paper leverages this characteristic, combined with explainable artificial intelligence (XAI) techniques, to address the challenges of segmentation. By adapting classification neural networks for segmentation tasks, the proposed approach reduces dependency on manual segmentation. To demonstrate this concept, the Medical Segmentation Decathlon ‘Brain Tumours’ dataset was utilised. A ResNet classification neural network was trained, and XAI tools were applied to generate segmentation-like outputs. Our findings reveal that GuidedBackprop is among the most efficient and effective methods, producing heatmaps that closely resemble segmentation masks by accurately highlighting the entirety of the target object. Full article

This paper presents an Enhanced Multilinear Principal Component Analysis (EMPCA) algorithm, an improved variant of the traditional Multilinear Principal Component Analysis (MPCA) tailored for efficient dimensionality reduction in high-dimensional data, particularly in image analysis tasks. EMPCA integrates random singular value decomposition to reduce computational complexity while maintaining data integrity. Additionally, it innovatively combines the dimensionality reduction method with the Mask R-CNN algorithm, enhancing the accuracy of image segmentation. Leveraging tensors, EMPCA achieves dimensionality reduction that specifically benefits image classification, face recognition, and image segmentation. The experimental results demonstrate a 17.7% reduction in computation time compared to conventional methods, without compromising accuracy. In image classification and face recognition experiments, EMPCA significantly enhances classifier efficiency, achieving comparable or superior accuracy to algorithms such as Support Vector Machines (SVMs). Additionally, EMPCA preprocessing exploits latent information within tensor structures, leading to improved segmentation performance. The proposed EMPCA algorithm holds promise for reducing image analysis runtimes and advancing rapid image processing techniques. Full article

This review offers a comprehensive and in-depth analysis of face mask detection and recognition technologies, emphasizing their critical role in both public health and technological advancements. Existing detection methods are systematically categorized into three primary classes: feaRture-extraction-and-classification-based approaches, object-detection-models-based methods and multi-sensor-fusion-based methods. Through a detailed comparison, their respective workflows, strengths, limitations, and applicability across different contexts are examined. The review underscores the paramount importance of accurate face mask detection, especially in response to global public health challenges such as pandemics. A central focus is placed on the role of datasets in driving algorithmic performance, addressing key factors, including dataset diversity, scale, annotation granularity, and modality. The integration of depth and infrared data is explored as a promising avenue for improving robustness in real-world conditions, highlighting the advantages of multimodal datasets in enhancing detection capabilities. Furthermore, the review discusses the synergistic use of real-world and synthetic datasets in overcoming challenges such as dataset bias, scalability, and resource scarcity. Emerging solutions, such as lightweight model optimization, domain adaptation, and privacy-preserving techniques, are also examined as means to improve both algorithmic efficiency and dataset quality. By synthesizing the current state of the field, identifying prevailing challenges, and outlining potential future research directions, this paper aims to contribute to the development of more effective, scalable, and robust face mask detection systems for diverse real-world applications. Full article

Lomami National Park, located in the Democratic Republic of the Congo (DR Congo), is renowned for the integrity of its forest ecosystems, safeguarded by the absence of agricultural activities and limited road access. However, these ecosystems remain under-researched, particularly in terms of forest cover dynamics. This research gap poses a significant challenge to establishing rigorous monitoring systems, which are essential for ensuring the long-term preservation of these valuable ecosystems. This study utilized Google Earth Engine to preprocess Landsat images from 2008, 2016, and 2024, employing techniques such as atmospheric correction and cloud masking. Random Forest classification was applied to analyze land cover changes, using training datasets curated through ground-truthing and region-of-interest selection. The classification accuracy was evaluated using metrics such as overall accuracy, producer’s accuracy, and user’s accuracy. To assess landscape configuration, metrics such as class area, patch number, largest patch index, disturbance index, aggregation index, and edge density were calculated, distinguishing between the park’s core and peripheral zones. Spatial transformation processes were analyzed using a decision tree approach. The results revealed a striking contrast in forest cover stability between Lomami National Park and its surrounding periphery. Within the park, forest cover has been preserved and even showed a modest increase, rising from 92.60% in 2008 to 92.75% in 2024. In contrast, the peripheral zone experienced a significant decline in forest cover, decreasing from 79.32% to 70.48% during the same period. This stability within the park extends beyond maintaining forested areas; it includes preserving and enhancing the spatial structure of forest ecosystems. For example, edge density, a key indicator of forest edge compactness, remained stable in the park, fluctuating between 8 m/ha and 9 m/ha. Conversely, edge density in the peripheral zone exceeded 35 m/ha, indicating that forest edges within the park are considerably more cohesive and intact than those in the surrounding areas. The spatial transformation processes also underscored these contrasting dynamics. In the park, the primary process was the aggregation of primary forest patches, reflecting a trend toward continuous and connected forest landscapes. By contrast, the peripheral zone exhibited dissection, indicating fragmentation and the breakdown of forest patches. These findings highlight the park’s critical role in maintaining both the extent and structural integrity of forest ecosystems, setting it apart from the more degraded periphery. They underscore the resilience of forest ecosystems in the face of limited anthropogenic pressures and the crucial importance of effective land management and rigorous conservation strategies in addressing the challenges posed by urbanization and rural expansion. Additionally, the results emphasize that well-adapted conservation measures, combined with specific demographic and socio-economic conditions, can play a pivotal role in achieving long-term forest preservation and ecological stability. Full article

Flooding is one of the most severe natural hazards, causing widespread environmental, economic, and social disruption. If not managed properly, it can lead to human losses, property damage, and the destruction of livelihoods. The ability to rapidly assess such damages is crucial for emergency management. Near Real-Time (NRT) spatial information on flood-affected areas, obtained via remote sensing, is essential for disaster response, relief, urban and industrial reconstruction, insurance services, and damage assessment. Numerous flood mapping methods have been proposed, each with distinct strengths and limitations. Among the most widely used are machine learning algorithms and spectral indices, though these methods often face challenges, particularly in threshold selection for spectral indices and the sampling process for supervised classification. This study aims to develop an NRT flood mapping approach using supervised classification based on spectral features. The method automatically generates training samples through masks derived from spectral indices. More specifically, this study uses FWEI, NDVI, NDBI, and BSI indices to extract training samples for water/flood, vegetation, built-up areas, and soil, respectively. The Otsu thresholding technique is applied to create the spectral masks. Land cover classification is then performed using the Random Forest algorithm with the automatically generated training samples. The final flood map is obtained by subtracting the pre-flood water class from the post-flood image. The proposed method is implemented using optical satellite images from Sentinel-2, Landsat-8, and Landsat-9. The proposed method’s accuracy is rigorously evaluated and compared with those obtained from spectral indices and machine learning techniques. The suggested approach achieves the highest overall accuracy (OA) of 90.57% and a Kappa Coefficient (KC) of 0.89, surpassing SVM (OA: 90.04%, KC: 0.88), Decision Trees (OA: 88.64%, KC: 0.87), and spectral indices like AWEI (OA: 84.12%, KC: 0.82), FWEI (OA: 88.23%, KC: 0.86), NDWI (OA: 85.78%, KC: 0.84), and MNDWI (OA: 87.67%, KC: 0.85). These results underscore the superior accuracy and effectiveness of the proposed approach for NRT flood detection and monitoring using multi-sensor optical imagery. Full article

Current deep learning models for time series often face challenges with generalizability in scenarios characterized by limited samples or inadequately labeled data. By tapping into the robust generative capabilities of diffusion models, which have shown success in computer vision and natural language processing, we see potential for improving the adaptability of deep learning models. However, the specific application of diffusion models in generating samples for time series classification tasks remains underexplored. To bridge this gap, we introduce the MDGPS model, which incorporates multi-task diffusion learning and gradient-free patch search (MDGPS). Our methodology aims to bolster the generalizability of time series classification models confronted with restricted labeled samples. The multi-task diffusion learning module integrates frequency-domain classification with random masked patches diffusion learning, leveraging frequency-domain feature representations and patch observation distributions to improve the discriminative properties of generated samples. Furthermore, a gradient-free patch search module, utilizing the particle swarm optimization algorithm, refines time series for specific samples through a pre-trained multi-task diffusion model. This process aims to reduce classification errors caused by random patch masking. The experimental results on four time series datasets show that the proposed MDGPS model consistently surpasses other methods, achieving the highest classification accuracy and F1-score across all datasets: 95.81%, 87.64%, 82.31%, and 100% in accuracy; and 95.21%, 82.32%, 78.57%, and 100% in F1-Score for Epilepsy, FD-B, Gesture, and EMG, respectively. In addition, evaluations in a reinforcement learning scenario confirm MDGPS’s superior performance. Ablation and visualization experiments further validate the effectiveness of its individual components. Full article

Few-shot image classification aims to improve the performance of traditional image classification when faced with limited data. Its main challenge lies in effectively utilizing sparse sample label data to accurately predict the true feature distribution. Recent approaches have employed data augmentation techniques like random Mask or mixture interpolation to enhance the diversity and generalization of labeled samples. However, these methods still encounter several issues: (1) random Mask can lead to complete blockage or exposure of foreground, causing loss of crucial sample information; and (2) uniform data distribution after mixture interpolation makes it difficult for the model to differentiate between different categories and effectively distinguish their boundaries. To address these challenges, this paper introduces a novel data augmentation method based on saliency mask blending. Firstly, it selectively preserves key image features through adaptive selection and retention using visual feature occlusion fusion and confidence clipping strategies. Secondly, a visual feature saliency fusion approach is employed to calculate the importance of various image regions, guiding the blending process to produce more diverse and enriched images with clearer category boundaries. The proposed method achieves outstanding performance on multiple standard few-shot image classification datasets (miniImageNet, tieredImageNet, Few-shot FC100, and CUB), surpassing state-of-the-art methods by approximately 0.2–1%. Full article