Deep learning-based fully automatic system for segmentation and defect classification of the solar modules using electroluminescence images

Bead nonlinear spiking neural P system for segmentation of multiple brain metastases at magnetic resonance imaging

A novel detection and segmentation system for eichhornia crassipes growth rate using region vision transformer-based adaptive Yolo with Unet++

Precision farming methods require consideration of subtle differences in plant growth processes in different areas of cultivated arable land. Differences in the relief of the field, its water supply, the thickness of the humus layer, etc. cause the need to rank the arable land for the application of agrotechnical operations of different intensities, which ultimately leads to higher yields and lower costs of agricultural crops. The ranking of arable land within a field is usually accomplished by segmentation of remotely sensed data. The creation of a segmentation system requires periodic remote and ground monitoring of fields, collection and processing of the received information with its geographical reference. Both satellite remote sensing systems and unmanned aerial platforms (UAPs) can be used for this purpose. The volume of information received, especially when using UAVs, is very significant, and the requirements for the speed of processing are high. In this regard, efficient methods of systematization and processing of the received images, which rely on sufficiently fast segmentation algorithms, are relevant. This paper considers methods of image segmentation from various remote sensing systems, which allow to increase the economic performance of agronomic measures in the precision farming loop. An example of a threshold segmentation program is given, which can be used separately or as part of an information system to support precision farming processes. The paper presents the results of its application for segmentation of satellite images of a field by NDVI index value. The conducted analysis and recommendations on the segmentation data will contribute to the prevention of environmental violations, yield losses due to sudden changes in weather conditions and differences in the relief of cultivated arable land.

WiFi sensing relies on capturing channel state information (CSI) fluctuations induced by human activities. Accurate motion segmentation is crucial for applications ranging from intrusion detection to activity recognition. However, prevailing methods based on variance, correlation coefficients, or deep learning are often constrained by complex threshold-setting procedures and dependence on high-quality sample data. To address these limitations, this paper proposes a training-free and environment-independent motion segmentation system using commercial WiFi devices from an image-processing perspective. The system employs a novel quasi-envelope to characterize CSI fluctuations and an iterative segmentation algorithm based on an improved Otsu thresholding method. Furthermore, a dedicated motion detection algorithm, leveraging the grayscale distribution of variance images, provides a precise termination criterion for the iterative process. Real-world experiments demonstrate that our system achieves an E-FPR of 0.33% and an E-FNR of 0.20% in counting motion events, with average temporal errors of 0.26 s and 0.29 s in locating the start and end points of human activity, respectively, confirming its effectiveness and robustness.

Aneurysmal subarachnoid hemorrhage (SAH) is a life-threatening neurological emergency with mortality rates exceeding 30%. While deep learning techniques show promise for automated SAH segmentation, their clinical application is limited by the scarcity of labeled data and challenges in cross-institutional generalization. Transfer learning from related hematoma types represents a potentially valuable but underexplored approach. Although Unet architectures remain the gold standard for medical image segmentation due to their effectiveness on limited datasets, Low-Rank Adaptation (LoRA) methods for parameter-efficient transfer learning have been rarely applied to convolutional neural networks in medical imaging contexts. The importance of SAH diagnosis and the time-intensive nature of manual annotation would benefit from automated solutions that can leverage existing multi-institutional datasets from more common conditions. We implemented a Unet architecture pre-trained on computed tomography scans from 124 traumatic brain injury patients across multiple institutions, then fine-tuned on 30 aneurysmal SAH patients from the University of Michigan Health System using 3-fold cross-validation. We developed a novel CP-LoRA method based on tensor canonical polyadic (CP) decomposition and introduced DoRA variants (DoRA-C, convDoRA, CP-DoRA) that decompose weight matrices into magnitude and directional components. We compared these approaches against existing LoRA methods (LoRA-C, convLoRA) and standard fine-tuning strategies across different modules on a multi-view Unet model. Performance was evaluated using Dice scores stratified by hemorrhage volume, with additional assessment of predicted versus annotated blood volumes. Transfer learning from traumatic brain injury to aneurysmal SAH demonstrated feasibility with all fine-tuning approaches achieving superior performance compared to no fine-tuning (mean Dice 0.410 ± 0.26). The best-performing traditional approach was decoding module fine-tuning (Dice 0.527 ± 0.20). LoRA-based methods consistently outperformed standard Unet fine-tuning, with DoRA-C at rank 64 achieving the highest overall performance (Dice 0.572 ± 0.17). Performance varied by hemorrhage volume, with all methods showing improved accuracy for larger volumes (Dice 0.682-0.694 for volumes > 100 mL vs. Dice 0.107-0.361 for volumes < 25 mL). CP-LoRA achieved comparable performance to existing methods while using significantly fewer parameters. Over-parameterization with higher ranks (64-96) consistently yielded better performance than strictly low-rank adaptations. This study demonstrates that transfer learning between hematoma types is feasible and that LoRA-based methods significantly outperform conventional Unet fine-tuning for aneurysmal SAH segmentation. The novel CP-LoRA method offers parameter efficiency advantages, while DoRA variants provide superior segmentation accuracy, particularly for small-volume hemorrhages. The finding that over-parameterization improves performance challenges traditional low-rank assumptions and suggests clinical applications may benefit from higher-rank adaptations. These results support the potential for automated SAH segmentation systems that leverage large multi-institutional traumatic brain injury datasets, potentially improving diagnostic speed and consistency when specialist expertise is unavailable.

Brain tumors remain among the most difficult of the medical challenges, and the accurate and timely diagnosis is essential to achieve successful patient outcomes. Over the last decade, artificial intelligence (AI), and specifically deep learning, has profoundly transformed the paradigms of brain tumor detection and segmentation methodologies. This comprehensive review systematically examines the evolution of brain tumor segmentation AI models between 2015 and 2025, covering technological advancements, performance evaluation techniques, and challenges towards clinical translation. We follow the evolution from traditional machine learning approaches to sophisticated deep learning architectures, including Convolutional Neural Networks (CNNs), U-Net architectures, and the more recently emerged Vision Transformers (ViTs). It takes into account the most crucial performance metrics, i.e., Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), accuracy, sensitivity, and specificity, which are primarily tested against benchmarking datasets, such as BraTS. Our findings register noteworthy improvements in performance, wherein the top-performing current ensemble and transformer-based models deliver Dice scores well above 0.95 for whole-tumor segmentation. Despite the stunning progress, limitations in standardization of evaluation, model generalizability across clinical settings and interpretability persist. This review describes the critical views of current capabilities, shortcomings, and directions of AI-based brain tumor segmentation systems with focus on the road to strong clinical deployment of these systems.

Weeds are one of the primary factors that reduce crop productivity by competing for nutrients and water, causing the plant to lose weight and resulting in reduced grain yield. Traditional agricultural practices often rely on uniform herbicide application, which can contaminate soil and raise costs. On agricultural land, selective weed treatment are an efficient and cost-effective way to control weeds that require a deep learning-based crop and weed segmentation system. Many existing crop and weed segmentation research works focus on achieving precise crop and weed segmentation results, rather than building lightweight models to deploy on edge devices. To attain this, we develop an effective and efficient convolutional neural network, namely the Concatenated Attention U-Net with Convolutional Block Attention Module (CAUC). By integrating Linear Concatenated Blocks (LCB), Attention Gate (AG) connections, and Convolutional Block Attention Module (CBAM), the proposed model efficiently utilizes feature maps among its architectural components to achieve superior performance. Depth-wise convolution layers and 1 × 1 convolution layers in LCBs reduce computational complexity. To enable the proposed model to identify the weed portions in multiple crop fields, we integrated three datasets in this research work, namely the Crop/Weed Field Image Dataset (CWFID), Sugar Beet, and Sunflower datasets. Experimental results on carrot, sugar beet, and sunflower crop datasets demonstrate high Accuracy (99.09%), MIoU (81.02%), and F1-score (99.06%), with a modest model size (5.6MB) and computational parameters (0.377million). We developed a lightweight computer vision application (13.7MB) to demonstrate the model's efficacy on low-computational devices.

Abstract This work presents an upgraded autonomous quadruped firefighting system. A hardware payload and set of algorithms are presented to allow a Boston Dynamics Spot robot to autonomously explore an unknown space, detect a fire, and extinguish the fire with a fire extinguisher. The work includes a novel hybrid color space and image segmentation model for locating fires, and an optimized frontier exploration algorithm used to minimize the search time while exploring the space, detecting a fire. Testing of the system was then successfully performed by having Spot autonomously explore a room, detect and then extinguish live fires in a controlled environment.

Modeling and analysis of active thermal regulation in segmented PCM wall systems for optimizing energy efficiency in severe cold regions

Performance of a Standardized Retrograde Urethrogram to Optimize Length, Segment, Etiology (LSE) Anterior Urethral Stricture Disease Classification and Staging.

HistoTrack++: A Vision-Based System for Temporal Bout Segmentation, Multi-Target Tracking and Kinematic Analysis in Overhead Combat Sports Videos

This study aimed to develop a dataset comprising computed tomography (CT) images and clinical data for melioidosis pneumonia and to utilize machine learning for assisting in prognosis prediction of the disease. We retrospectively analyzed multicenter data from five hospitals to establish a dataset for diagnosing melioidosis pneumonia, including CT images and clinical data. An AI-based CT segmentation system was employed to extract lung tissue and lesions. Quantitative analysis methods were used to derive CT lesion characteristics specific to melioidosis pneumonia. Pearson and Spearman correlation tests were performed to examine the relationships between CT lesion characteristics and clinical parameters. Finally, a machine learning method was applied by combining CT lesion characteristics with clinical parameters to perform prognosis analysis, predicting the progression of melioidosis pneumonia patients to severe or critical illness. Correlation analysis revealed that lung injury was associated with clinical markers from other organ systems, indicating an interrelationship between lung lesions and systemic health. Using a multilayer perceptron classifier, which combined CT lesion characteristics with clinical parameters, the model predicted the progression to severe or critical illness with an AUC of 0.9570 (95% CI: 0.9262–0.9816). This study demonstrated that the CT lesion characteristics of melioidosis pneumonia are correlated with indicators of multi-organ function. Combining CT lesion characteristics with clinical parameters improves the efficiency of prognosis prediction for melioidosis pneumonia. Lung injury CT lesion characteristics were identified as primary markers for predicting the disease prognosis.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-26666-x.

Proton exchange membrane fuel cells (PEMFCs) are energy conversion devices that offer high power densities at low operating temperatures, making them one of the most promising technology for a wide range of applications, including automotive, backup power generating units and portable systems. Commercial PEMFC systems from Nuvera Fuel Cells LLC employ open flow field (OFF) architectures, enabling superior efficiency at high power densities above 1.0 W cm-2 [1, 2]. Understanding the performance of this flow field design in comparison to the conventional land-channel (L-C) benchmark, using modern and advanced diagnostic methods such as segmented cell, electrochemical impedance spectroscopy (EIS) and distribution of relaxation times (DRT) provides critical and important insights for fuel cell stacks developers targeting mobile applications.The experimental work was performed using a test station and a segmented cell system developed at Hawaii Natural Energy Institute [3]. Commercially available 100 cm2 catalyst coated membranes with Pt content of 0.1 and 0.4 mgPt cm-2 for anode and cathode, respectively, were used in this work. 25BC gas diffusion layers were applied for anode and cathode. The OFF- test cell incorporated an OFF on the cathode side, while the anode flow field utilized a 10-channel serpentine with segmentation at the anode. The land-channel test cell (L-C) used a 10-channel serpentine flow field for anode and cathode, with segmentation applied to the cathode. The 10-channel serpentine featured the following geometric parameters: land width of 0.762 mm, channel width of 0.794 mm and channel depth of 0.838 mm. Both test articles utilized co-flow of air and hydrogen gases.Electrochemical diagnostics included determination of electrochemical surface area (ECA) and H2 crossover, as well as measurements of polarization curves (IV curves) under H2/air and H2/O2 gas configurations. The VI curves were used to estimate activation, ohmic and mass transfer overpotentials, as described in our previous work [3]. To determine mass transport resistances, limiting current measurements were performed using 5% O2 diluted with various gases (He, N2, Ar, CF4) as the cathode feed, while pure H2 was supplied to the anode [4]. The performance of the samples was studied under two operating conditions: 1) Anode/cathode: H2 /air, 2/2 stoichiometry, 100/50% RH at the cell inlet,150/150 kPa backpressure and 80°C; and 2) Industry-relevant operating conditions with reduced RH at the cell inlet, 50/50%.Fig. 1 shows polarization curves and high-frequency resistance (HFR) data recorded for both the L-C and OFF architectures and for selected locations of the MEA: segment 1 (inlet), segments 4 and 7 (middle), segment 10 (outlet) and for the total cell.The application of the OFF architecture led to higher overall performance and more uniform performance distribution compared to the L-C flow field design. At the same time, similar HFR values were recorded for both architectures. Additionally, the OFF resulted in a lower temperature gradient across the cell during high current density operation.The performance improvement observed for the OFF architecture was attributed to minimal mass transport losses, underscoring the excellent gas transport and water management capabilities of the open flow field design.Mass transport resistances (RMT ) for the OFF cell were determined at different humidity levels (100/100, 100/50 and 50/50% RH) (Table 1). The results indicated that the mass transport resistance through the ionomer (Rfilm, CL ) was slightly higher for the OFF sample. However, the mass transport resistance in gas phase (Rm, N2 ) was significantly lower in the OFF architecture, contributing to its superior performance under high power-generating conditions. A detailed analysis and comparison of the local performance of PEMFCs employing OFF and land-channel architectures including EIS and DRT data will be presented and discussed.ACKNOWLEDGEMENTSWe gratefully acknowledge funding from Nuvera Fuel Cells LLC and the US Office of Naval Research (N00014-22-1-2045).References K. Srouji, L.J. Zheng, R. Dross, A. Turhan, M.M. Mench, J. Power Sources, 218, 341-347 (2012).Reshetenko, O. Polevaya, Electrochim. Acta, 387 138529 (2021).V. Reshetenko, G. Bender, K. Bethune, R. Rocheleau, Electrochim. Acta 88, 571-579 (2013).T.V. Reshetenko, J. St-Pierre, J. Electrochem. Soc., 161, F1089-F1100 (2014). Figure 1

Coronary algorithm for cardiac sub structures and prospective real-time surveillance of cardiac dose exposure. Retro and prospective study to validate AI auto-segmentation. A 3D UNet was trained on 560 thoracic CT scans from a single institution (2003-2014) and validated internally (n=70). External validation was performed in 283 patients treated at an independent institution (2005-2020). Clinical implementation comprised (1) retrospective analysis of 3,399 lung cancer patients treated in 2014-2022 and (2) prospective surveillance of 1,386 consecutive patients in 2023. Geometric accuracy, concordance of dose-volume parameters; association of AI-derived substructure metrics with outcome; temporal dose trends; and the proportion of patients exceeding prespecified risk. Median (inter-quartile range) Dice/ASSD were 0.95 (0.94-0.96)/1.1 mm for the heart and 0.87 (0.82-0.90)/1.9 mm for the LAD; the median absolute difference between AI and manual LAD V15 was 1%. AI-derived LAD V15 remained independently associated with MACE (sub distribution hazard ratio [HR], 1.03%; 95% CI, 1.01-1.05) and ACM (adjusted HR, 1.02; 95% CI, 1.00-1.03), internally and externally. Retrospective deployment showed a 32% relative decline in median LAD V15 from 2014 to 2022 (12% to 8%) and identified high- risk doses in 1,086 of 3,399 patients (32%). Prospective surveillance flagged 264 contemporary patients (19%) for potential cardiology referral. A validated AI system accurately segments cardiac substructures, reproduces dose-outcome relationships, enables large-scale surveillance, and point-of-care alerts for high-risk patients. Automated cardiac dose monitoring could facilitate adoption of coronary-sparing therapy and follow-up.

To advance surface error correction and high-precision surface shape maintenance in large-aperture segmented telescopes, this study introduces the multi-point radial-axial hybrid force-driven active correction technique (MP-RHFACT) for sub-mirrors. In contrast to conventional sub-mirror correction technologies (e.g., warping harness), the proposed technique demonstrates a significantly enhanced capability for correcting low-order surface shape errors, such as defocus and astigmatism. The active correction system mainly consists of 12 sets of edge hybrid force actuator assemblies and 1 set of central axial force actuator assembly, achieving active correction of low-order aberrations by coordinating 6 pairs of radial forces and 13 sets of axial forces. For a hexagonal sub-mirror with a circumscribed diameter of about 1.1 m and a vertex curvature radius of approximately 10.6 m, simulation analysis reveals that the technique reduces defocus error by a factor of 85.4, 0°/45° astigmatism by factors of 93.8/94.5 respectively, and horizontal/oblique trefoil aberrations by factors of 16.3 and 8.9, thereby demonstrating excellent correction performance-particularly for defocus and astigmatism errors. It should be noted that the reduction factor is defined as the ratio of the initial surface error RMS to the residual RMS after correction. This method offers robust support for achieving and maintaining high-precision co-phasing in segmented systems. Further investigations indicate that, even in the presence of correction force magnitude errors (±0.2 N) and directional errors (inclination angle ±1°), the RMS values of sub-mirror surface shape error induced by these deviations remain below 6.9 nm, confirming the system's robust performance. In summary, the MP-RHFACT proposed in this paper reduces the complexity, fabrication, and alignment challenges, and manufacturing costs associated with sub-mirrors, thereby providing an effective solution for the engineering application of large-aperture segmented telescopes.

Abstract Stochastic resonance (SR) systems are widely used for early weak fault detection; however, their output saturation characteristics restrict the nonlinear amplification effect, which reduces the ability to extract signal features. To overcome output saturation limitations, an improved segmented anti-saturation bistable stochastic resonance system (ISABSR) is proposed for early fault diagnosis. The paper constructs a second-order bistable potential function, which effectively mitigates the output saturation problem in SR systems by discarding the quartic term in the classic SR potential function that causes this phenomenon. In fault extraction, the Grey Wolf Optimization (GWO) algorithm is employed, using the system output signal-to-noise ratio (SNR) as the fitness function. This enables multi-parameter adaptation of the ISABSR, thereby enhancing the system's fault extraction capability and robustness. This method, combined with Variational Mode Decomposition (VMD), is experimentally validated. The proposed method effectively alleviates the output saturation phenomenon in stochastic resonance systems and demonstrates superior fault signal feature extraction ability compared to other comparative methods.

Deep learning-based segmentation system is proposed that exploits three variants of YOLOv11 architecture, namely YOLOv11n-seg, YOLOv11s-seg, and YOLOv11m-seg for automated detection and localization of the oral disease conditions from photographic intraoral images. Dataset has been created by combining publicly available data from sources like Roboflow resulting in an initial version (v1) having 582 images annotated at the pixel level. To mitigate class imbalance issues as well as increase generalization capability by the model, progressively this dataset was augmented to create version 2 (v2) and then further extended into version 3 (v3). Training was done in three separate stages: Feature extraction, partial fine-tuning, and full fine-tuning. YOLOv11m-seg model performed best in the partial fine-tuning phase with results of box mAP@50 = 0.521 and mask mAP@50 = 0.500. Training was done on Google Colab's free tier using an Intel Xeon CPU with 13 GB RAM, 15 GB T4 GPU, and 120 GB storage allowance. For application, the best performing model was exported to ONNX format with NMS enabled and deployed as a fully client-side responsive web app built in React.js and ONNX Runtime Web. The tool enables both clinicians and non-experts to detect oral diseases from intraoral images with a single click.

The quality of urban space influences the sustainability of cities and the well-being of their inhabitants, but quantifying this attribute through numerical representations of urban conditions has proved difficult in urban planning. This article describes how integrating qualitative and quantitative datasets through analytical and generative methods can enhance the comprehension and evaluation of urban space quality. Focusing on the city of Sydney, Australia, the research employed a public survey to assess the urban conditions of 11 suburbs against key qualitative traits of beauty, comfort, safety and ambience. The data was analysed using image segmentation and geographical information systems, and correlations between the survey responses and the urban characteristics present in each image were calculated. The results include nine characteristics of urban spaces that reflect the listed qualitative traits and a percentile ratio for each urban condition that represents the perception of each trait, offering a comprehensive understanding of the determinants of the quality of urban spaces. The research contributes to ongoing efforts to improve the quality of life in urban environments by providing a highly specific and clear quantification of four highly subjective perceptions of urban space. The proposed quality measurement method represents a valuable tool for policymakers, urban planners and designers to use to inform decision-making and ultimately create more liveable, sustainable and inclusive cities.

Abstract. Urban environments are continually evolving, which presents significant challenges for 3D semantic segmentation systems that must adapt to emerging object categories. In this paper, we address the problem of Novel Class Discovery (NCD) in 3D semantic segmentation for urban scenes. We introduce a feature-driven framework that leverages the Dynamic Multi-level Feature Synthesis Module (D-MFSM) to extract and integrate multi-scale, cross-view structural information from raw urban point clouds. D-MFSM dynamically partitions point clouds via an adaptive grouping mechanism that utilizes a learnable spatial weight vector, and subsequently constructs local neighborhoods by means of an improved farthest point sampling strategy. The extracted local features are then processed by a dual-path adaptive synthesis mechanism and further refined through a novel cross-axis reordering strategy, which together yield comprehensive aggregated feature representations. These representations facilitate robust novel class discovery while maintaining high segmentation accuracy on known classes. Comprehensive evaluations on the DALES dataset demonstrate that the proposed approach yields substantial improvements in segmentation performance across diverse urban scenarios. The proposed framework, therefore, offers a complementary solution to existing methods and contributes to the development of more adaptive and accurate 3D semantic segmentation systems in complex urban settings.