Search Results (455)

Object detection technology enables the automatic identification of photovoltaic (PV) panel locations and conditions, significantly enhancing operational efficiency for maintenance teams while reducing the time and cost associated with manual inspections. Challenges arise due to the low resolution of remote sensing images combined with small-sized targets—PV panels intertwined with complex urban or natural backgrounds. To address this, a parallel architecture model based on YOLOv5 was designed, substituting traditional residual connections with parallel convolution structures to enhance feature extraction capabilities and information transmission efficiency. Drawing inspiration from the bottleneck design concept, a primary feature extraction module framework was constructed to optimize the model’s deep learning capacity. The improved model achieved a 4.3% increase in mAP, a 0.07 rise in F1 score, a 6.55% enhancement in recall rate, and a 6.2% improvement in precision. Additionally, the study validated the model’s performance and examined the impact of different loss functions on it, explored learning rate adjustment strategies under various scenarios, and analyzed how individual factors affect learning rate decay during its initial stages. This research notably optimizes detection accuracy and efficiency, holding promise for application in large-scale intelligent PV power station maintenance systems and providing reliable technical support for clean energy infrastructure management. Full article

The construction industry is one of the largest contributors to the world economy. However, the level of automation and digitalization in the construction industry is still at its infancy in comparison with other industries due to the complex nature and the large size of construction projects. Meanwhile, construction projects are prone to cost overruns and schedule delays due to the adoption of traditional progress monitoring techniques to retrieve progress on-site, having indoor activities participating with an accountable ratio of these works. Improvements in deep learning and Computer Vision (CV) algorithms provide promising results in detecting objects in real time. Also, researchers have investigated the probability of using CV as a tool to create a Digital Twin (DT) for construction sites. This paper proposes a model utilizing the state-of-the-art YOLOv8 algorithm to monitor the progress of bricklaying activities, automatically extracting and analyzing real-time data from construction sites. The detected data is then integrated into a 3D Building Information Model (BIM), which serves as a DT, allowing project managers to visualize, track, and compare the actual progress of bricklaying with the planned schedule. By incorporating this technology, the model aims to enhance accuracy in progress monitoring, reduce human error, and enable real-time updates to project timelines, contributing to more efficient project management and timely completion. Full article

Microscopic image automatic recognition is a core technology for mineral composition analysis and plays a crucial role in advancing the intelligent development of smart mining systems. To overcome the limitations of traditional lithium ore analysis and meet the challenges of deployment on edge devices, we propose PS-YOLO-seg, a lightweight segmentation model specifically designed for lithium mineral analysis under visible light microscopy. The network is compressed by adjusting the width factor to reduce global channel redundancy. A PSConv-based downsampling strategy enhances the network’s ability to capture dim mineral textures under microscopic conditions. In addition, the improved C3k2-PS module strengthens feature extraction, while the streamlined Segment-Efficient head minimizes redundant computation, further reducing the overall model complexity. PS-YOLO-seg achieves a slightly improved segmentation performance compared to the baseline YOLOv12n model on a self-constructed lithium ore microscopic dataset, while reducing FLOPs by 20%, parameter count by 33%, and model size by 32%. Additionally, it achieves a faster inference speed, highlighting its potential for practical deployment. This work demonstrates how architectural optimization and targeted enhancements can significantly improve instance segmentation performance while maintaining speed and compactness, offering strong potential for real-time deployment in industrial settings and edge computing scenarios. Full article

Corrosion is a critical issue in civil structures, significantly affecting their durability and functionality. Detecting corrosion at an early stage is essential to prevent structural failures and ensure safety. This study proposes an expert system based on a novel methodology for corrosion detection using vibration signal analysis. The approach employs graphical empirical mode decomposition (GEMD) to decompose vibration signals into their intrinsic mode functions, extracting relevant structural features. These features are then transformed into grayscale images and classified using a Convolutional Neural Network (CNN) to automatically differentiate between a healthy structure and one affected by corrosion. To enhance the computational efficiency of the method without compromising accuracy, different CNN architectures and image sizes are tested to propose a low-complexity model. The proposed approach is validated using a 3D nine-bay truss-type bridge model encountered in the Vibrations Laboratory at the Autonomous University of Querétaro, Mexico. The evaluation considers three different corrosion levels: (1) incipient, (2) moderate, and (3) severe, along with a healthy condition. The combination of GEMD and CNN provides a highly accurate corrosion detection framework that achieves 100% classification accuracy while remaining effective regardless of the damage location and severity, making it a reliable tool for early-stage corrosion assessment that enables timely maintenance and enhances structural health monitoring to improve the long life and safety of civil structures. Full article

Background: Most of the existing hyperpolarized (HP) ¹³C MRI analyses use univariate rate maps of pyruvate-to-lactate conversion (k_PL), and radiomic-style multiparametric models extracting complex, higher-order features remain unexplored. Purpose: To establish a multivariate framework based on whole abdomen/pelvis HP ¹³C-pyruvate MRI and evaluate the association between multiparametric features of metabolism (MFM) and clinical outcome measures in advanced and metastatic prostate cancer. Methods: Retrospective statistical analysis was performed on 16 participants with metastatic or local-regionally advanced prostate cancer prospectively enrolled in a tertiary center who underwent HP-pyruvate MRI of abdomen or pelvis between November 2020 and May 2023. Five patients were hormone-sensitive and eleven were castration-resistant. GMP-grade [1-¹³C]pyruvate was polarized using a 5T clinical-research DNP polarizer, and HP MRI used a set of flexible vest-transmit, array-receive coils, and echo-planar imaging sequences. Three basic metabolic maps (k_PL, pyruvate summed-over-time, and mean pyruvate time) were created by semi-automatic segmentation, from which 316 MFMs were extracted using an open-source, radiomic-compliant software package. Univariate risk classifier was constructed using a biologically meaningful feature (k_PL,median), and the multivariate classifier used a two-step feature selection process (ranking and clustering). Both were correlated with progression-free survival (PFS) and overall survival (OS) (median follow-up = 22.0 months) using Cox proportional hazards model. Results: In the univariate analysis, patients harboring tumors with lower-k_PL,median had longer PFS (11.2 vs. 0.5 months, p < 0.01) and OS (NR vs. 18.4 months, p < 0.05) than their higher-k_PL,median counterparts. Using a hypothesis-generating, age-adjusted multivariate risk classifier, the lower-risk subgroup also had longer PFS (NR vs. 2.4 months, p < 0.002) and OS (NR vs. 18.4 months, p < 0.05). By contrast, established laboratory markers, including PSA, lactate dehydrogenase, and alkaline phosphatase, were not significantly associated with PFS or OS (p > 0.05). Key limitations of this study include small sample size, retrospective study design, and referral bias. Conclusions: Risk classifiers derived from select multiparametric HP features were significantly associated with clinically meaningful outcome measures in this small, heterogeneous patient cohort, strongly supporting further investigation into their prognostic values. Full article

To address the challenges in ship plate detection under complex maritime scenarios—such as small target size, extreme aspect ratios, dense arrangements, and multi-angle rotations—this paper proposes a multi-module collaborative detection algorithm, RT-DETR-HPA, based on an enhanced RT-DETR framework. The proposed model integrates three core components: an improved High-Frequency Enhanced Residual Block (HFERB) embedded in the backbone to strengthen multi-scale high-frequency feature fusion, with deformable convolution added to handle occlusion and deformation; a Pinwheel-shaped Convolution (PConv) module employing multi-directional convolution kernels to achieve rotation-adaptive local detail extraction and accurately capture plate edges and character features; and an Adaptive Sparse Self-Attention (ASSA) mechanism incorporated into the encoder to automatically focus on key regions while suppressing complex background interference, thereby enhancing feature discriminability. Comparative experiments conducted on a self-constructed dataset of 20,000 ship plate images show that, compared to the original RT-DETR, RT-DETR-HPA achieves a 3.36% improvement in mAP@50 (up to 97.12%), a 3.23% increase in recall (reaching 94.88%), and maintains real-time detection speed at 40.1 FPS. Compared with mainstream object detection models such as the YOLO series and Faster R-CNN, RT-DETR-HPA demonstrates significant advantages in high-precision localization, adaptability to complex scenarios, and real-time performance. It effectively reduces missed and false detections caused by low resolution, poor lighting, and dense occlusion, providing a robust and high-accuracy solution for intelligent ship supervision. Future work will focus on lightweight model design and dynamic resolution adaptation to enhance its applicability on mobile maritime surveillance platforms. Full article

The Advanced Topographic Laser Altimeter System (ATLAS) on the Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) employs a photon-counting detection mode with a 532 nm laser to obtain high-precision Earth surface elevation data and offers a new remote sensing method for nearshore bathymetry. The key issues in using ATLAS ATL03 data for bathymetry are achieving automatic and accurate extraction of signal photons in different water environments. Especially for areas with sharply fluctuating topography, the interaction of various impacts, such as topographic fluctuations, sea waves, and laser pulse direction, can result in a sharp change in photon density and distribution at the seafloor, which can cause the signal photon detection at the seafloor to be misinterpreted or omitted during analysis. Therefore, an improved size and direction adaptive filtering (ISDAF) method was proposed for nearshore bathymetry using ATLAS ATL03 data. This method can accurately distinguish between the original photons located above the sea surface, on the sea surface, and the seafloor. The size and direction of the elliptical density filter kernel automatically adapt to the sharp fluctuations in topography and changes in water depth, ensuring precise extraction of signal photons from both the sea surface and the seafloor. To evaluate the precision and reliability of the ISDAF, ATLAS ATL03 data from different water environments and seafloor terrains were used to perform bathymetric experiments. Airborne LiDAR bathymetry (ALB) data were also used to validate the bathymetric accuracy and reliability. The experimental findings show that the ISDAF consistently exhibits effectiveness in detecting and retrieving signal photons, regardless of whether the seafloor terrain is stable or dynamic. After applying refraction correction, the high accuracy of bathymetry was evidenced by a strong coefficient of determination (R²) and a low root mean square error (RMSE) between the ICESat-2 bathymetry data and ALB data. This research offers a promising approach to advancing remote sensing technologies for precise nearshore bathymetric mapping, with implications for coastal monitoring, marine ecology, and resource management. Full article

Background: Soybean is an important global crop in food security and agricultural economics. Accurate estimation of soybean pod counts is critical for yield prediction, breeding programs, precision farming, etc. Traditional methods, such as manual counting, are slow, labor-intensive, and prone to errors. With rapid advancements in artificial intelligence (AI), deep learning has enabled automatic pod number estimation in collaboration with unmanned aerial vehicles (UAVs). However, existing AI models are computationally demanding and require significant processing resources (e.g., memory). These resources are often not available in rural regions and small farms. Methods: To address these challenges, this study presents a set of lightweight, efficient AI models designed to overcome these limitations. By integrating model simplification, weight quantization, and squeeze-and-excitation (SE) self-attention blocks, we develop compact AI models capable of fast and accurate soybean pod count estimation. Results and Conclusions: Experimental results show a comparable estimation accuracy of 84–87%, while the AI model size is significantly reduced by a factor of 9–65, thus making them suitable for deployment in edge devices, such as Raspberry Pi. Compared to existing models such as YOLO POD and SoybeanNet, which rely on over 20 million parameters to achieve approximately 84% accuracy, our proposed lightweight models deliver a comparable or even higher accuracy (84.0–86.76%) while using fewer than 2 million parameters. In future work, we plan to expand the dataset by incorporating diverse soybean images to enhance model generalizability. Additionally, we aim to explore more advanced attention mechanisms—such as CBAM or ECA—to further improve feature extraction and model performance. Finally, we aim to implement the complete system in edge devices and conduct real-world testing in soybean fields. Full article

Background/Objectives: This study explores the integration of wearable sensors and artificial intelligence (AI) for Human Activity Recognition (HAR) in the diagnosis and rehabilitation of Parkinson’s disease (PD). The objective was to develop a proof-of-concept model based on internal reproducibility, without external generalization, that is capable of distinguishing pathological movements from healthy ones while ensuring clinical relevance and patient safety. Methods: Nine subjects, including eight patients with Parkinson’s disease and one healthy control, were included. Motion data were collected using the Motigravity platform, which integrates inertial sensors in a controlled environment. The signals were automatically segmented into fixed-length windows, with poor-quality segments excluded through preprocessing. A hybrid CNN-LSTM (Convolutional Neural Networks—Long Short-Term Memory) model was trained to classify motion patterns, leveraging convolutional layers for spatial feature extraction and LSTM layers for temporal dependencies. The Motigravity system provided a controlled hypogravity environment for data collection and rehabilitation exercises. Results: The proposed CNN-LSTM model achieved a validation accuracy of 100%, demonstrating classification potential. The Motigravity system contributed to improved data reliability and ensured patient safety. Despite increasing class imbalance in extended experiments, the model consistently maintained perfect accuracy, suggesting strong generalizability after external validation to overcome the limitations. Conclusions: Integrating AI and wearable sensors has significant potential to improve the HAR-based classification of movement impairments and guide rehabilitation strategies in PD. While challenges such as dataset size remain, expanding real-world validation and enhancing automated segmentation could further improve clinical impact. Future research should explore larger cohorts, extend the model to other neurodegenerative diseases, and evaluate its integration into clinical rehabilitation workflows. Full article

Early detection of morbidities that complicate pregnancy improves health outcomes in low- and middle-income countries. Automatic revision of electronic health records (EHRs) can help identify such morbidity risks. There is a lack of corpora to train models in Spanish in specific domains, and there are no models specialized in maternal EHRs. This study aims to develop a fine-tuned model that detects clinical concepts using a built database with text extracted from maternal EHRs in Spanish. We created a corpus with 13.998 annotations from 200 clinical notes in Spanish associated with EHRs obtained from a reference institution of high obstetric risk in Colombia. Using the Beginning–Inside–Outside tagging scheme, we fine-tuned five different transformer-based models to classify between 16 classes associated with eight entities. The best model achieved a macro F1 score of 0.55 ± 0.03. The entities with the best performance were signs, symptoms, and negations, with exact F1 scores of 0.714 and 0.726, respectively. The lower scores were associated with those classes with fewer observations. Even though our dataset is shorter in size and more diverse in entity types than other datasets in Spanish, our results are comparable to other state-of-the-art named entity recognition models fine-tuned in Spanish and the biomedical domain. This work introduces the first fine-tuning of a model for named entity recognition specifically designed for maternal EHRs. Our results can be used as a base to develop new models to extract concepts in the maternal–fetal domains and help healthcare providers detect morbidities that complicate pregnancy early. Full article

Accurate polyp segmentation plays a vital role in the early detection and prevention of colorectal cancer. However, the diverse shapes, blurred boundaries, and varying sizes of polyps present significant challenges for automatic segmentation. Existing methods often struggle with effective local feature extraction and modeling long-range dependencies. To overcome these limitations, this paper proposes PolypFormer, which incorporates a local information enhancement module (LIEM) utilizing multi-kernel self-selective attention to better capture texture features, alongside dense channel attention to facilitate more effective feature fusion. Furthermore, a novel cross-shaped windows self-attention mechanism is introduced and integrated into the Transformer architecture to enhance the semantic understanding of polyp regions. Experimental results on five datasets show that the proposed method has good performance in polyp segmentation. On Kvasir-SEG datasets, mDice and mIoU reach 0.920 and 0.886, respectively. Full article

As the primary method used in open-pit mining, blasting has a direct impact on the efficiency and cost of subsequent operations. Therefore, dynamic identification of rock fragment size after blasting is essential for evaluating blasting quality and optimizing mining plans. This study presents a YOLOv8-based binocular vision model for real-time recognition of blasting fragmentation. The model is trained on a dataset comprising 1536 samples, which were annotated using an automatic labeling algorithm and expanded to 7680 samples through data augmentation techniques. The YOLOv8 instance segmentation model is employed to detect and classify rock fragments. By integrating binocular vision-based automatic image capture with Welzl’s algorithm, the actual particle size of each rock fragment is calculated. Furthermore, region of interest (ROI) extraction and shadow-based data enhancement techniques are incorporated to focus the model on the blasting fragmentation area and reduce environmental interference. Finally, software and a system were independently developed based on this integrated model and successfully deployed at engineering sites. The dynamic recognition Mean Average Precision of this integrated model is 0.84, providing a valuable reference for evaluating blasting effects and improving work efficiency. Full article

Due to issues such as low resolution, scattering noise, and background clutter, ship detection in Synthetic Aperture Radar (SAR) images remains challenging, especially in inshore regions, where these factors have similar scattering characteristics. To overcome these challenges, this paper proposes a novel SAR ship detection framework that integrates adaptive channel attention with large kernel adaptation. The proposed method improves multi-scale contextual information extraction by enhancing feature map interactions at different scales. This method effectively reduces false positives, missed detections, and localization ambiguities, especially in complex inshore environments. Also, it includes an adaptive channel attention block that adjusts attention weights according to the dimensions of the input feature maps, enabling the model to prioritize local information and improve sensitivity to small object features in SAR images. In addition, a large kernel attention block with adaptive kernel size is introduced to automatically adjust the receptive field designed to extract abundant context information at different detection layers. Experimental evaluations on the SSDD and Hysid SAR ship datasets indicate that our method achieves excellent detection performance compared to current methods, as well as demonstrate its effectiveness in overcoming SAR ship detection challenges. Full article

Orthotropic steel box girders and steel bridge decks are commonly applied to bridges. Because of the coupling of original defects and alternating forces, fatigue cracks are likely to appear in the structures. In order to ensure the life span of bridges, methods for automatic crack identification are needed. In this paper, we present a novel approach for crack detection and bridge condition monitoring by integrating convolutional neural networks (CNNs) with graph attention networks (GATs). At first, the original large-sized images are divided into small-sized patches, and these patches are input into a CNN architecture to extract features by decreasing dimensions. Then, the output features of the CNN model are considered as nodes of the graph. Considering the spatial relationship among the patches in the original image, the node from the central patch is connected to the nodes from its neighboring patches to constitute a graph structure, which can be input into a GAT model to learn the relationship among the nodes and update the features. Finally, the output features of GAT can judge whether the central patch contains cracks. Forty original large-sized images are cropped into abundant patches for the training of the CNN-GAT model. With the use of a sliding window technique, the trained CNN-GAT model is capable of finding the patches containing cracks in the test images with large sizes. From the test results, the location and the size of the cracks are exhibited, which indicates that the proposed approach is effective for crack identification in bridge structures. Full article

The recognition of Dongba pictographs presents significant challenges due to the pitfalls in traditional feature extraction methods, classification algorithms’ high complexity, and generalization ability. This study proposes a convolutional neural network (CNN)-based image classification method to enhance the accuracy and efficiency of Dongba pictograph recognition. The research begins with collecting and manually categorizing Dongba pictograph images, followed by these preprocessing steps to improve image quality: normalization, grayscale conversion, filtering, denoising, and binarization. The dataset, comprising 70,000 image samples, is categorized into 18 classes based on shape characteristics and manual annotations. A CNN model is then trained using a dataset that is split into training (with 70% of all the samples), validation (20%), and test (10%) sets. In particular, data augmentation techniques, including rotation, affine transformation, scaling, and translation, are applied to enhance classification accuracy. Experimental results demonstrate that the proposed model achieves a classification accuracy of 99.43% and consistently outperforms other conventional methods, with its performance peaking at 99.84% under optimized training conditions—specifically, with 75 training epochs and a batch size of 512. This study provides a robust and efficient solution for automatically classifying Dongba pictographs, contributing to their digital preservation and scholarly research. By leveraging deep learning techniques, the proposed approach facilitates the rapid and precise identification of Dongba hieroglyphs, supporting the ongoing efforts in cultural heritage preservation and the broader application of artificial intelligence in linguistic studies. Full article