Search Results (81)

Feature tracking is essential for welding crawler robots’ trajectory planning. As welding often occurs in dark environments like pipelines or ship hulls, the system requires low-light image capture for laser tracking. However, such images typically have poor brightness and contrast, degrading both weld seam feature extraction and trajectory anomaly detection accuracy. To address this, we propose a Retinex-based low-light enhancement network tailored for cladding scenarios. The network features an illumination curve estimation module and requires no paired or unpaired reference images during training, alleviating the need for cladding-specific datasets. It adaptively adjusts brightness, restores image details, and effectively suppresses noise. Extensive experiments on public (LOLv1 and LOLv2) and self-collected weld datasets show that our method outperformed existing approaches in PSNR, SSIM, and LPIPS. Additionally, weld seam segmentation under low-light conditions achieved 95.1% IoU and 98.9% accuracy, confirming the method’s effectiveness for downstream tasks in robotic welding. Full article

Objective: This study aimed to compare trained male and female athletes regarding physiological, perceptual, and performance variables at ventilatory thresholds (VT₁ and VT₂). Methods: Twenty-four male and nineteen female trained runners (age: 27.9 ± 6.4 vs. 24.4 ± 4.4 years; body mass: 61.8 ± 4.3 vs. 52.6 ± 4.1 kg; height: 174.6 ± 5.8 vs. 165.0 ± 5.0 cm for males and females, respectively) performed a graded exercise test to exhaustion on a treadmill. During the test, oxygen consumption, respiratory exchange ratio, running power output, heart rate, muscle oxygenation, and rate of perceived exertion were analyzed. Sex differences were evaluated with an unpaired-samples t-test. Results: Males exhibited significantly higher respiratory exchange ratios (0.87 ± 0.04 vs. 0.83 ± 0.03; 1.03 ± 0.06 vs. 1.01 ± 0.06) and absolute running speeds (15.00 ± 1.06 vs. 12.42 ± 1.22 km·h⁻¹; 19.04 ± 1.06 vs. 16.32 ± 1.29 km·h⁻¹) at both thresholds (p < 0.05), whereas women showed higher muscle oxygenation in vastus lateralis (60.44 ± 21.21 vs. 26.38 ± 10.21%) and fractional utilization of maximal aerobic speed (93.64 ± 6.44 vs. 91.43 ± 3.21%) at VT₂ (p < 0.01). Also, rate of perceived exertion was similar between sexes at both thresholds. Conclusion: Males showed higher absolute physiological values, while females demonstrated greater fractional utilization at VT₂ and higher muscle oxygenation. No sex differences were observed in rate of perceived exertion. These findings highlight the importance of using ventilatory thresholds in training prescription. Full article

With increasing attention to sustainability and energy efficiency in transportation systems, advanced intelligent algorithms provide promising solutions for optimizing urban rail transit operations. This study addresses the challenge of optimizing train operation plans for urban rail transit systems characterized by spatiotemporal passenger flow imbalance. By exploring a combined short-turning and unpaired train operation mode, a three-objective optimization model was established, aiming to minimize operational costs, reduce passenger waiting times, and enhance load balancing. To effectively solve this complex problem, an Improved GOOSE (IGOOSE) algorithm incorporating elite opposition-based learning, probabilistic exploration based on elite solutions, and golden-sine mutation strategies were developed, significantly enhancing global search capability and solution robustness. A case study based on real operational data adjusted for confidentiality was conducted, and comparative analyses with Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Grey Wolf Optimizer (GWO) demonstrated the superiority of IGOOSE. Furthermore, an ablation study validated the effectiveness of each enhancement strategy within the IGOOSE algorithm. The optimized operation planning model reduced passenger waiting times by approximately 12.72%, improved load balancing by approximately 39.30%, and decreased the overall optimization objective by approximately 10.25%, highlighting its effectiveness. These findings provide valuable insights for urban rail transit operation management and indicate directions for future research, underscoring the significant potential for energy savings and emission reductions toward sustainable urban development. Full article

In recent years, the continuous advancement of deep learning technology and its integration into the domain of low-light image enhancement have led to a steady improvement in enhancement effects. However, this progress has been accompanied by an increase in model complexity, imposing significant constraints on applications that demand high real-time performance. To address this challenge, inspired by the state-of-the-art Zero-DCE approach, we introduce a novel method that transforms the low-light image enhancement task into a curve estimation task tailored to each individual image, utilizing a lightweight shallow neural network. Specifically, we first design a novel curve formula based on Gamma correction, which we call the Gamma-based light-enhancement (GLE) curve. This curve enables outstanding performance in the enhancement task by directly mapping the input low-light image to the enhanced output at the pixel level, thereby eliminating the need for multiple iterative mappings as required in the Zero-DCE algorithm. As a result, our approach significantly improves inference speed. Additionally, we employ a lightweight network architecture to minimize computational complexity and introduce a novel global channel attention (GCA) module to enhance the nonlinear mapping capability of the neural network. The GCA module assigns distinct weights to each channel, allowing the network to focus more on critical features. Consequently, it enhances the effectiveness of low-light image enhancement while incurring a minimal computational cost. Finally, our method is trained using a set of zero-reference loss functions, akin to the Zero-DCE approach, without relying on paired or unpaired data. This ensures the practicality and applicability of our proposed method. The experimental results of both quantitative and qualitative comparisons demonstrate that, despite its lightweight design, the images enhanced using our method not only exhibit perceptual quality, authenticity, and contrast comparable to those of mainstream state-of-the-art (SOTA) methods but in some cases even surpass them. Furthermore, our model demonstrates very fast inference speed, making it suitable for real-time inference in resource-constrained or mobile environments, with broad application prospects. Full article

Image style transfer is a challenging task that has gained significant attention in recent years due to its growing complexity. Training is typically performed using paradigms offered by GAN-based image style transfer networks. Cycle-based training methods provide an approach for handling unpaired data. Nevertheless, achieving high transfer quality remains a challenge with these methods due to the simplicity of the employed network. The purpose of this research is to present VariGAN, a novel approach that incorporates three additional strategies to optimize GAN-based image style transfer: (1) Improving the quality of transferred images by utilizing an effective UNet generator network in conjunction with a context-related feature extraction module. (2) Optimizing the training process while reducing dependency on the generator through the use of a depthwise discriminator. (3) Introducing LPIPS loss to further refine the loss function and enhance the overall generation quality of the framework. Through a series of experiments, we demonstrate that the VariGAN backbone exhibits superior performance across diverse content and style domains. VariGAN improved class IoU by $236\%$ and participant identification by $195\%$ compared to CycleGAN. Full article

(1) Objective: As light propagates through water, it undergoes significant attenuation and scattering, causing underwater images to experience color distortion and exhibit a bluish or greenish tint. Additionally, suspended particles in the water further degrade image quality. This paper proposes an improved U-Net network model for underwater image enhancement to generate high-quality images. (2) Method: Instead of incorporating additional complex modules into enhancement networks, we opted to simplify the classic U-Net architecture. Specifically, we replaced the standard convolutions in U-Net with our self-designed efficient basic block, which integrates a simplified channel attention mechanism. Moreover, we employed Layer Normalization to enhance the capability of training with a small number of samples and used the GELU activation function to achieve additional benefits in image denoising. Furthermore, we introduced the SK fusion module into the network to aggregate feature information, replacing traditional concatenation operations. In the experimental section, we used the “Underwater ImageNet” dataset from “Enhancing Underwater Visual Perception (EUVP)” for training and testing. EUVP, established by Islam et al., is a large-scale dataset comprising paired images (high-quality clear images and low-quality blurry images) as well as unpaired underwater images. (3) Results: We compared our proposed method with several high-performing traditional algorithms and deep learning-based methods. The traditional algorithms include He, UDCP, ICM, and ULAP, while the deep learning-based methods include CycleGAN, UGAN, UGAN-P, and FUnIEGAN. The results demonstrate that our algorithm exhibits outstanding competitiveness on the underwater imagenet-dataset. Compared to the currently optimal lightweight model, FUnIE-GAN, our method reduces the number of parameters by 0.969 times and decreases Floating-Point Operations Per Second (FLOPS) by more than half. In terms of image quality, our approach achieves a minimal UCIQE reduction of only 0.008 while improving the NIQE by 0.019 compared to state-of-the-art (SOTA) methods. Finally, extensive ablation experiments validate the feasibility of our designed network. (4) Conclusions: The underwater image enhancement algorithm proposed in this paper significantly reduces model size and accelerates inference speed while maintaining high processing performance, demonstrating strong potential for practical applications. Full article

Objectives: A thorough knowledge of tooth morphology, encompassing the detailed structural complexities, is essential for the practice of dental hygienists in all aspects of their profession. The aim of this study was to assess the efficacy of two instructional approaches in tooth morphology education, by analyzing the performance of dental hygienist students trained with human extracted teeth compared to those educated with plastic teeth models. Methods: This study included two cohorts of undergraduate dental hygienist students: a control group (n = 27) trained using human teeth, and an experimental group (n = 34) trained using plastic teeth models. Each group underwent two consecutive practical exams where they identified all 32 permanent teeth and 8 deciduous molars. Initially, students were tested on the training material that they were assigned (either extracted human teeth or plastic teeth), and, subsequently, they were tested using the alternative material. Both the number and patterns of identification errors were recorded and analyzed. Paired t-tests were used to compare error rates between real and plastic teeth for students trained on either plastic or real teeth, unpaired t-tests were conducted to assess differences in performance between students trained on plastic versus real teeth when tested on both tooth types, and Fisher’s exact tests were employed to examine variations in error proportions across maxillary and mandibular tooth categories. Results: The control group recorded a mean of 6.41 errors per student (total of 173 errors), with three students (11.1%) failing by committing over 12 errors. Their performance improved to a mean of 5.44 errors (total of 147 errors) when tested on plastic teeth, although the improvement was not statistically significant (p = 0.20). Conversely, the experimental group demonstrated high accuracy on plastic teeth, with 19 out of 34 students (55.9%) achieving perfect scores and a total of only 50 errors (mean, 1.47). Their performance, however, declined when tested on real teeth, escalating to a total of 354 errors, with 32 students (94.12%) making more errors on real teeth than on plastic, resulting in a significant increase in errors to an average of 10.41 per student (p < 0.001). Conclusions: This study demonstrates that students perform best when tested on the materials that they initially were trained with, showing that real teeth provide better educational outcomes than plastic models. This advantage underscores the importance of using natural teeth when learning dental anatomy. Full article

Corrosion monitoring is crucial for ensuring the structural integrity of equipment. Acoustic emission (AE) and electrochemical noise (EN) have been proven to be highly effective for the detection of corrosion. Due to the complementary nature of these two techniques, previous studies have demonstrated that combining both signals can facilitate research on corrosion monitoring. However, current machine learning models have not yet been able to effectively integrate these two different modal types of signals. Therefore, a new deep learning framework, CorroNet, is designed to synergistically integrate AE and EN signals at the algorithmic level for the first time. The CorroNet leverages multimodal learning, enhances accuracy, and automates the monitoring process. During training, paired AE-EN data and unpaired EN data are used, with AE signals serving as anchors to help the model better align EN signals with the same corrosion stage. A new feature alignment loss function and a probability distribution consistency loss function are designed to facilitate more effective feature learning to improve classification performance. Experimental results demonstrate that CorroNet achieves superior accuracy in corrosion stage classification compared to other state-of-the-art models, with an overall accuracy of 97.01%. Importantly, CorroNet requires only EN signals during the testing phase, making it suitable for stable and continuous monitoring applications. This framework offers a promising solution for real-time corrosion detection and structural health monitoring. Full article

The rapid growth of cities and their populations in recent years has resulted in significant tidal passenger flow characteristics, primarily manifested in the imbalance of passenger numbers in both directions. This imbalance often leads to a shortage of train capacity in one direction and an inefficient use of capacity in the other. To accommodate the tidal passenger flow demand of urban rail transit, this paper proposes a timetable optimization method that combines multiple strategies, aimed at reducing operating costs and enhancing the quality of passenger service. The multi-strategy optimization method primarily involves two key strategies: the unpaired operation strategy and the express/local train operation strategy, both of which can flexibly adapt to time-varying passenger demand. Based on the decision variables of headway, running time between stations, and dwell time, a mixed integer linear programming model (MILP) is established. Taking the Shanghai Suburban Railway airport link line as an example, simulations under different passenger demands are realized to illustrate the effectiveness and correctness of the proposed multi-strategy method and model. The results demonstrate that the multi-strategy optimization method achieves a 38.59% reduction in total costs for both the operator and the passengers, and effectively alleviates train congestion. Full article

General face recognition is currently one of the key technologies in the field of computer vision, and it has achieved tremendous success with the support of deep-learning technology. General face recognition models currently exhibit extremely high accuracy on some high-quality face datasets. However, their performance decreases in challenging environments, such as low-light scenes. To enhance the performance of face recognition models in low-light scenarios, we propose a face recognition approach based on feature decoupling and fusion (DeFFace). Our main idea is to extract facial-related features from images that are not influenced by illumination. First, we introduce a feature decoupling network (D-Net) to decouple the image into facial-related features and illumination-related features. By incorporating the illumination triplet loss optimized with unpaired identity IDs, we regulate illumination-related features to minimize the impact of lighting conditions on the face recognition system. However, the decoupled features are relatively coarse. Therefore, we introduce a feature fusion network (F-Net) to further extract the residual facial-related features from the illumination-related features and fuse them with the initial facial-related features. Finally, we introduce a lighting-facial correlation loss to reduce the correlation between the two decoupled features in the specific space. We demonstrate the effectiveness of our method on four real-world low-light datasets and three simulated low-light datasets. We retrain multiple general face recognition methods using our proposed low-light training sets to further validate the advanced performance of our method. Compared to general face recognition methods, our approach achieves an average improvement of more than 2.11 percentage points on low-light face datasets. In comparison with image enhancement-based solutions, our method shows an average improvement of around 16 percentage points on low-light datasets, and it also delivers an average improvement of approximately 5.67 percentage points when compared to illumination normalization-based methods. Full article

Unpaired image translation with feature-level constraints presents significant challenges, including unstable network training and low diversity in generated tasks. This limitation is typically attributed to the following situations: 1. The generated images are overly simplistic, which fails to stimulate the network’s capacity for generating diverse and imaginative outputs. 2. The images produced are distorted, a direct consequence of unstable training conditions. To address this limitation, the unpaired image-to-image translation with diffusion adversarial network (UNDAN) is proposed. Specifically, our model consists of two modules: (1) Feature fusion module: In this module, one-dimensional SVD features are transformed into two-dimensional SVD features using the convolutional two-dimensionalization method, enhancing the diversity of the images generated by the network. (2) Network convergence module: In this module, the generator transitions from the U-net model to a superior diffusion model. This shift leverages the stability of the diffusion model to mitigate the mode collapse issues commonly associated with adversarial network training. In summary, the CycleGAN framework is utilized to achieve unpaired image translation through the application of cycle-consistent loss. Finally, the proposed network was verified from both qualitative and quantitative aspects. The experiments show that the method proposed can generate more realistic converted images. Full article

Background/Objectives: Apparent Diffusion Coefficient (ADC) maps in prostate MRI can reveal tumor characteristics, but their accuracy can be compromised by artifacts related with patient motion or rectal gas associated distortions. To address these challenges, we propose a novel approach that utilizes a Generative Adversarial Network to synthesize ADC maps from T2-weighted magnetic resonance images (T2W MRI). Methods: By leveraging contrastive learning, our model accurately maps axial T2W MRI to ADC maps within the cropped region of the prostate organ boundary, capturing subtle variations and intricate structural details by learning similar and dissimilar pairs from two imaging modalities. We trained our model on a comprehensive dataset of unpaired T2-weighted images and ADC maps from 506 patients. In evaluating our model, named AI-ADC, we compared it against three state-of-the-art methods: CycleGAN, CUT, and StyTr2. Results: Our model demonstrated a higher mean Structural Similarity Index (SSIM) of 0.863 on a test dataset of 3240 2D MRI slices from 195 patients, compared to values of 0.855, 0.797, and 0.824 for CycleGAN, CUT, and StyTr2, respectively. Similarly, our model achieved a significantly lower Fréchet Inception Distance (FID) value of 31.992, compared to values of 43.458, 179.983, and 58.784 for the other three models, indicating its superior performance in generating ADC maps. Furthermore, we evaluated our model on 147 patients from the publicly available ProstateX dataset, where it demonstrated a higher SSIM of 0.647 and a lower FID of 113.876 compared to the other three models. Conclusions: These results highlight the efficacy of our proposed model in generating ADC maps from T2W MRI, showcasing its potential for enhancing clinical diagnostics and radiological workflows. Full article

Existing approaches for image deraining often rely on synthetic or unpaired real-world rainy datasets, leading to sub-optimal generalization ability when processing the complex and diverse real-world rain degradation. To address these challenges, we propose a novel iterative semantic-guided detail fusion model with implicit neural representations (INR-ISDF). This approach addresses the challenges of complex solution domain variations, reducing the usual negative impacts found in these situations. Firstly, the input rainy images are processed through implicit neural representations (INRs) to obtain normalized images. Residual calculations are then used to assess the illumination inconsistency caused by rain degradation, thereby enabling an accurate identification of the degradation locations. Subsequently, the location information is incorporated into the detail branch of the dual-branch architecture, while the normalized images obtained from the INR are used to enhance semantic processing. Finally, we use semantic clues to iteratively guide the progressive fusion of details to achieve improved image processing results. To tackle the partial correspondence between real rain images and the given ground truth, we propose a two-stage training strategy that utilizes adjustments in the semantic loss function coefficients and phased freezing of the detail branch to prevent potential overfitting issues. Extensive experiments verify the effectiveness of our proposed method in eliminating the degradation in real-world rainy images. Full article

Shipbuilding drawings, crafted manually before the digital era, are vital for historical reference and technical insight. However, their digital versions, stored as scanned PDFs, often contain significant noise, making them unsuitable for use in modern CAD software like AutoCAD. Traditional denoising techniques struggle with the diverse and intense noise found in these documents, which also does not adhere to standard noise models. In this paper, we propose an innovative generative approach tailored for document enhancement, particularly focusing on shipbuilding drawings. For a small, unpaired dataset of clean and noisy shipbuilding drawing documents, we first learn to generate the noise in the dataset based on a CycleGAN model. We then generate multiple paired clean–noisy image pairs using the clean images in the dataset. Finally, we train a Pix2Pix GAN model with these generated image pairs to enhance shipbuilding drawings. Through empirical evaluation on a small Military Sealift Command (MSC) dataset, we demonstrated the superiority of our method in mitigating noise and preserving essential details, offering an effective solution for the restoration and utilization of historical shipbuilding drawings in contemporary digital environments. Full article

Specular highlight removal is a challenging task in the field of image enhancement, while it can significantly improve the quality of image in highlight regions. Recently, deep learning-based methods have been widely adopted in this task, demonstrating excellent performance by training on either massive paired data, wherein both the highlighted and highlight-free versions of the same image are available, or unpaired datasets where the one-to-one correspondence is inapplicable. However, it is difficult to obtain the corresponding highlight-free version of a highlight image, as the latter has already been produced under specific lighting conditions. In this paper, we propose a method for weakly supervised specular highlight removal that only requires highlight images. This method involves generating highlight-free images from highlight images with the guidance of masks estimated using non-negative matrix factorization (NMF). These highlight-free images are then fed consecutively into a series of modules derived from a Cycle Generative Adversarial Network (Cycle-GAN)-style network, namely the highlight generation, highlight removal, and reconstruction modules in sequential order. These modules are trained jointly, resulting in a highly effective highlight removal module during the verification. On the specular highlight image quadruples (SHIQ) and the LIME datasets, our method achieves an accuracy of 0.90 and a balance error rate (BER) of 8.6 on SHIQ, and an accuracy of 0.89 and a BER of 9.1 on LIME, outperforming existing methods and demonstrating its potential for improving image quality in various applications. Full article