Abstract Time series of Normalized Difference of Vegetation Index (NDVI) values, derived from satellite data, are useful for monitoring the vegetation status and can form a basis for more advanced analysis. However, data in these time series is affected by noise caused primarily by atmospheric conditions and acquisition errors, resulting in glitches and prompting the need for developing reconstruction techniques that can efficiently remove the noise. A multitude of approaches have been developed so far, including a variety of temporal-based methods that include filtering techniques. In this letter, a morphological filter with a non-flat structuring element is proposed for NDVI time series reconstruction. This method is applied on two time series obtained from the Copernicus Global Land Service 300 m NDVI product. The experimental results prove the effectiveness of the proposed approach in producing high-quality NDVI reconstructions, highlighted by the significantly better root mean square error (RMSE) values obtained on the considered time series in comparison with three well-established techniques.

Correlative imaging is a powerful analytical approach in bioimaging, as it offers complementary information on the samples measured by different modalities. Particularly, correlative transmission electron microscopy (EM) and nanoscale secondary ion mass spectrometry (NanoSIMS) imaging enable high-resolution morphological and chemical analysis at the subcellular level. However, manual segmentation and correlation of regions of interest (ROIs) in large EM and NanoSIMS data sets are time-consuming, prone to user bias, and limited in throughput. To address this, we developed a computer vision-assisted image analysis pipeline for automatic classification and segmentation of subcellular organelles in EM images, enabling rapid and reproducible correlation with NanoSIMS ion data. Using human neuronal progenitor cells (hNPCs) and differentiated postmitotic neurons, we trained a YOLOv8 deep learning model to recognize six major organelle types. The pipeline included EM image preprocessing, segmentation via YOLOv8, morphological filtering, and image registration with NanoSIMS ion maps. Performance evaluation demonstrated a robust model accuracy. We applied the pipeline to measure 15N-leucine abundance to study protein turnover in single organelles across different cell states. Results showed distinct turnover dynamics among organelles, with slower turnover observed in differentiated neurons compared to hNPCs. The automated pipeline significantly reduced the analysis time (from hours to minutes) while maintaining consistency with manual segmentation. Our approach demonstrates how computer vision can streamline correlative imaging workflows, improve data quality, and enable deeper insights into subcellular processes such as protein turnover, making it especially valuable for SIMS users and broader bioimaging applications.

Deep learning models for atmospheric pattern recognition require spatially consistent training labels that align precisely with input meteorological fields. This study introduces an automatic cold front detection method using the ERA5 reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF) at 850 hPa, specifically designed to generate physically consistent labels for machine learning applications. The approach combines the Thermal Front Parameter (TFP) with temperature advection (AdvT), applying optimized thresholds (TFP < 5 × 10−11 K m−2; AdvT < −1 × 10−4 K s−1), morphological filtering, and polynomial smoothing. Comparison against 1426 manual charts from 2009 revealed systematic spatial displacement, with mean offsets of ~502 km. Although pixel-level overlap was low, with Intersection over Union (IoU) = 0.013 and Dice coefficient (Dice) = 0.034, spatial concordance exceeded 99%, confirming both methods identify the same synoptic systems. The automatic method detects 58% more fronts over the South Atlantic and 44% fewer over the Andes compared to manual charts. Seasonal variability shows maximum activity in austral winter (31.3%) and minimum in summer (20.1%). This is the first automatic front detection system calibrated for South America that maintains direct correspondence between training labels and reanalysis input fields, addressing the spatial misalignment problem that limits deep learning applications in atmospheric sciences.

Abstract BACKGROUND Diffuse infiltration of malignant glioma cells into surrounding brain tissue complicates maximal safe resection. Fluorescence-guided surgery (FGS) with 5-aminolevulinic acid (5-ALA) enhances tumor visualization via protoporphyrin IX (PpIX) fluorescence, but its relationship to cell density remains unclear. We investigate whether PpIX fluorescence correlates with histological cellularity in glioma tissues. MATERIAL AND METHODS We analyzed 243 brain tumor biopsies from patients administered with 5-ALA. Ex vivo hyperspectral imaging captured fluorescence spectra, and a spectral unmixing algorithm quantified abundances of nine fluorophores, including PpIX. Cell density was measured from histopathological slides using an automated cell counting algorithm involving image segmentation and morphological filtering. We focused on glioblastoma samples (n=209) to control for tissue-type variability. For comparison, other tumor types included anaplastic astrocytoma (n=10), gliosarcoma (n=9), metastasis (n=5), and radiation necrosis (n=10). Using linear and quadratic models, we assessed correlations between fluorophore abundances and cell density. RESULTS In glioblastoma samples, weak but statistically significant positive correlations were found between cell density and fluorescence from PpIX634 (R=0.387, p<0.001) and collagen (R=0.403, p<0.001), with collagen showing a slightly stronger correlation. No strong correlation was observed between PpIX fluorescence intensity and cell density across all tumor types (R²=0.17, p<0.001). Quadratic models marginally improved the correlation (R²=0.28) but risked overfitting. Tissue type significantly influenced cellularity and fluorophore abundance (p<0.05), suggesting that factors beyond cell density affect PpIX accumulation. Glioblastoma samples exhibited higher PpIX fluorescence despite lower cell densities than other tumor types, indicating PpIX accumulation in the extracellular matrix or higher intracellular accumulation per cell. CONCLUSION PpIX fluorescence correlates weakly with tumor cell density, suggesting accumulation in the extracellular matrix rather than within tumor cells or higher intracellular accumulation per cell. Collagen fluorescence shows a stronger correlation and may serve as an additional intraoperative biomarker. While PpIX aids tumor visualization, it does not directly reflect cell density. Incorporating biomarkers like collagen fluorescence could enhance tumor delineation and improve surgical outcomes.

Citrus fruits, especially lemons, play a vital economic and nutritional role worldwide but are increasingly threatened by a wide range of diseases that diminish yield quality and quantity. Traditional manual and automated methods for disease detection requires domain expert, ample observation time, and is often ineffective during early infection stages. This paper presents a novel automated approach for the symptom based detection and classification of citrus leaf diseases using a nonlinear Fuzzy Rank-Based Ensemble (NL-FuRBE) methodology, enhanced by image quality improvement techniques. The study emphasizes the significance of timely disease diagnosis in citrus crops, which are vital for global food security and economic stability. The methodology begins with image quality enhancement through Vector-Valued Anisotropic Diffusion (VAD) and morphological filtering, evaluated using PSNR, SSIM, and NIQE metrics to ensure optimal visual clarity for classifier input. The core ensemble integrates three deep learning (DL) architectures–VGG19, AlexNet, and Xception–using a fuzzy rank-based scoring mechanism built on nonlinear transformations (exponential, tanh, and sigmoid functions) to address prediction uncertainty and model bias. A comprehensive dataset of lemon leaf diseases, consisting of 1354 images across nine classes, was utilized for training and evaluation. Experimental results using five-fold cross-validation demonstrate that the proposed model achieves superior performance with an average accuracy of 96.51%, outperforming conventional ensemble and state-of-the-art approaches. The results validate the proposed NL-FuRBE as an effective, automated, and cost-efficient tool for precision agriculture and early disease diagnosis in citrus farming.

Volumetric Visualization of the Dermal Vasculature with Signal and Image-based Feature Extraction on a High-frequency Ultrafast Ultrasound Dataset.

The assessment and optimization of the MIPA (Morphological Image Processing Approach) algorithm for the retrieval of Atmospheric Boundary Layer Height (ABLH) from Aerosol High-power Lidars (AHL) data are presented. MIPA has been developed at CNR-IMAA in the framework of ACTRIS, and it was tested on several lidar datasets, showing, in general, a good agreement with the traditional ABLH retrieval techniques. The main innovative feature of MIPA with respect to other approaches consists in applying optimized morphological filters and object-oriented analysis on lidar timeseries to obtain ABLH estimates. In this study, we carried out a robust MIPA validation effort based on a dedicated measurement campaign organized at CIAO (CNR-IMAA Atmospheric Observatory) in Spring 2024, where several lidar systems were operating continuously along with a quite complete set of other atmospheric sensors and two radiosounding systems. During the campaign, several case studies were considered for MIPA validation, each characterized by an intensive radiosonde schedule to ensure the establishment of a representative ABLH reference dataset. The ABLH retrieved by MIPA was compared against the corresponding ones obtained by radiosonde data. We observed a good overall agreement under different atmospheric conditions, ranging from intense dust events penetrating the ABL to cleaner atmospheric conditions. The best agreement between MIPA and reference dataset is obtained for longer wavelengths (532 nm and 1064 nm) and during daytime conditions.

Early and accurate detection of retinal pathologies is critical for preventing vision loss and enabling timely clinical intervention. Traditional computer vision techniques, such as thresholding, edge detection, morphological filtering, and Hough transforms, have long been used to extract features from retinal fundus images, yet their performance is often constrained by image variability and complex pathological presentations. This study presents a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNNs) for image-based classification with Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) units, to model geometric and anatomical features derived from classical methods. This architecture allows for the fusion of pixel-level deep features with clinically interpretable descriptors, including optic disc-fovea distance, lesion spatial distribution, and vessel curvature sequences. Comparative analysis demonstrates that the proposed hybrid model achieves superior diagnostic accuracy, reaching 97%, significantly outperforming both conventional image processing approaches and CNN-only baselines. The results indicate that incorporating structured domain knowledge into neural models improves both performance and interpretability, offering a robust framework for real-world retinal disease screening applications.

In CNC machining, the limitations of geometric smoothing and feedrate smoothing methods often result in non-smooth tool motion, leading to workpiece vibrations and deteriorating machining quality. To address the above problem, this article proposes a geometric-feedrate dual-scale planning method that enhances both machining path and feedrate smoothness. At the geometric level, an energy-based approach smooths the machining path while ensuring bounded path errors. By constructing a stiffness matrix, the method formulates path smoothing as an energy minimization problem. A modified Particle Swarm Optimization (MPSO) algorithm effectively solves this problem, significantly improving path smoothness. At the feedrate level, a gradient-based strategy initially updates the feedrate to minimize fluctuations. Traditional S-curve-based feedrate planning methods require recalculations, increasing computational burdens. To overcome this, a four-time hybrid filtering structure (a first-order morphological filter followed by a three-time FIR filter) is introduced. The morphological filter employs an erosion-dilation strategy with a predefined filtering radius to preserve local feedrate minima. The three-time FIR filter smooths the feedrate by applying a predefined time parameter T , ensuring compliance with kinematic constraints. Meanwhile, a bidirectional interpolation method is proposed to reduce feedrate fluctuations caused by interpolation, in which the step size is determined by considering both previous and next interpolation cycles. Finally, Simulations and experiments confirm the effectiveness of the proposed method.

Abstract The present study designs and implements a visual recognition and robotic arm control system based on an FPGA (Field-Programmable Gate Array) platform. The system achieves two core functions by operating on a low-cost, single-chip FPGA with limited hardware resources. First, it captures real-time images using a camera. It performs RGB-to-YCbCr color space conversion, morphological filtering, and image segmentation directly on the FPGA to extract the target objects’ color, shape, and position information in real time. Second, it computes the inverse kinematics of the robotic arm by combining the CORDIC algorithm with lookup tables, enabling accurate calculation of joint angles corresponding to target positions and coordinating multi-degree-of-freedom movements. The system has been successfully deployed and tested on an experimental platform, demonstrating its ability to recognize and manipulate multiple target objects, thereby validating its real-time performance and effectiveness in image processing and motion control. Compared with traditional GPU or CPU+FPGA hybrid platforms, the proposed system offers advantages such as faster response speed, lower hardware cost, higher integration, and reconfigurability, making it a promising solution with strong potential for further research and application.

The paper presents a mathematical model for real-time object detection and recognition in video streams, based on stepwise analysis of inter-frame changes. The proposed approach integrates basic linear and morphological operations with an efficient inter-frame differencing procedure, enabling the localization of moving or newly appearing objects across consecutive frames, followed by their classification using neural networks. The formalized algorithmic structure of the model covers all essential stages: image scaling, grayscale conversion, absolute difference computation, threshold filtering, morphological cleanup, extraction of regions of interest, object classification, and subsequent temporal tracking. The model is structured as a sequence of functional transformations addressing both spatial and temporal aspects of video data processing. The use of inter-frame differencing as a core activity detector is justified as it significantly reduces the computational burden in comparison with fully convolutional deep learning models such as SSD or YOLO. Classical morphological filters (opening and closing) are employed to refine object contours, while size-based region filtering helps exclude noisy or irrelevant areas. At the final stage, validated regions are passed to a classification module, allowing identification of object types and enabling tracking without repeated detection. An experimental evaluation was conducted using footage from a static camera to assess the model’s effectiveness. The results demonstrate an average frame processing time of 5.4 ms, meeting real-time operational requirements, and a recognition accuracy of 71.2%. Profiling indicates that the most computationally intensive operations are associated with morphological processing, whereas classification accounts for less than half of the total processing time. This highlights the efficiency of the hybrid approach, where simple linear preprocessing significantly reduces the data load for classification without substantial accuracy loss.

Abstract Solar-blind ultraviolet(UV) imaging is an effective discharge detection method. To quantitatively describe discharge characteristics, a new method using digital image processing algorithms to extract image parameters for characterizing discharge characteristics is proposed. The discharge areas were extracted by threshold segmentation, morphological filtering, and small area elimination algorithms on UV images. The boundary point coordinates of each discharge area are obtained using a multi area boundary tracking algorithm, and the spot area parameter is defined to quantify the discharge intensity, and the total running time for the above image processing is 0.59 seconds. The parallel line projection algorithm in the Radon algorithm was used to obtain the spatial distribution curve of discharge intensity along the surface of the insulator.

Abstract Mapping the internal stratigraphy of ice sheets serves a variety of glaciological applications, from the study of past ice flows to current distribution of surface mass balance and melting to contemporary ice dynamics, all of which are crucial for improving future projections of sea level rise. The method of choice for investigating the internal structure of ice sheets is radio‐echo sounding (RES). Mapping englacial stratigraphy has been carried out so far mostly by time‐consuming manual or semi‐automatic methods. Although reliable, such approaches are not feasible for comprehensive analysis of the wealth of data available. Therefore, (semi‐)automatic mapping of the internal stratigraphy from RES radargrams has been a field of interest for about two decades. Here, our goal is to present a complete pipeline for automatic tracing of internal reflection horizons (IRH) of intermediate to large depths in the ice sheet from radargrams using deep learning. We introduce IRHMapNet, which is a deep learning framework that uses a U‐Net‐based architecture to trace IRHs, based on airborne RES data with preprocessing steps such as noise removal and data augmentation, and postprocessing techniques such as morphological filtering and skeletonization. We use a combination of manually fully‐traced radargrams, results of image processing and thresholding method, and layer slope inference for training U‐Net architectures. We evaluate the successful performance of our approach, also for deep and extended IRH analysis, using a variety of metrics and discuss remaining progress and persistent shortcomings of machine learning approaches. The results of our experiments demonstrate that IRHMapNet successfully achieves the objective with specific architecture and data sets.

Abstract Accurate flood mapping in urban environments remains a critical yet challenging task. The primary challenge lies in the accuracy of widely available topographic data, a key constraint when employing hydrodynamic models for large‐scale inundation mapping. Recent advances in Very High‐Resolution satellite Photogrammetry Digital Elevation Models offer a promising solution to mitigate this limitation. However, effective preprocessing is imperative before integrating these data sets into flood inundation modeling, as the presence of artifacts such as buildings and trees can lead to inaccurate simulations. Here, we evaluated the potential of using the 0.5 m‐resolution GoogleDEM for flood inundation modeling using New Delhi, India—a densely populated urban area with pronounced flood risk—as a case study. We first examined the feasibility of extracting flood defenses from GoogleDEM. We then developed an iterative morphological filter to generate a bare‐earth GoogleDEM. Lastly, we assessed the flood inundation accuracy of our processed GoogleDEM by simulating a flood event with a 25‐year return period flood. We find that the errors of GoogleDEM‐derived flood defense heights are mostly within 1 m, but the automated extraction of defenses remains challenging. Our proposed approach reduced the artifact‐related errors of the original GoogleDEM by 85% (RMSE). The error of the simulated water surface level in the bare‐earth GoogleDEM (with flood defenses) was reduced by 97% compared to the original GoogleDEM, down to 0.16 m. This study presents a comprehensive evaluation of integrating cutting‐edge DEM data to enhance flood mapping accuracy, particularly in complex urban areas that are otherwise extremely data‐poor.

Autism Spectrum Disorder (ASD) is a neuro-developmental disorder characterized by varying degrees of difficulty in social interaction and communication and repetitive behaviors. Early confirmation of the diagnosis of ASD leads to early appropriate treatment. However, confirming ASD diagnosis is challenging due to its wide spectrum and challenging behavior assessment. This research proposes a technology-based ASD diagnosis on children utilizing thermal facial analysis. This is conducted subject to the uniqueness of facial expression that is typically applied to children with ASD. A modified Inception V3 architecture did the intended thermal facial analysis for ASD diagnosis. Homomorphic filters and morphological filters are applied to the data pre-processing to improve the classification ability. The proposed identification method shows better sensitivity to the false-positive aspect. It is indicated by better performance in terms of precision, with a rate of 90% to 91%. This research is expected to support medical experts in confirming early diagnosis in children with ASD.

An accurate thermal measurement of low-frequency stimulation (LFS) pads for thermotherapy was investigated using background subtraction (BGS) methods. The safety of LFS thermal pads must be investigated to prevent low-temperature burns (LTBs), because they frequently contact the sensitive skin in neck, shoulder and abdominal regions. The thermal measurement was based on thermal imaging using the active region-of-interest (ROI) from a foreground. The shape of the LFS thermal pad consists of complicated curves, thus it is difficult to extract the foreground using conventional shapes of ROIs. We proposed the foreground extraction using background subtraction (BGS) and digital and morphological filters to time-variant thermal images. The foreground extraction was implemented using open sources and experimented for abdominal, cervical and patellar pads. The results showed that the foreground can be separated from background regardless of the size, position, orientation and shape of the pad. The thermal characteristics of the LFS thermal pads were evaluated from the complicated shapes of the foreground with high accuracy. This study demonstrated that standard deviation of pixel history (SDPH) is a simple method for the BGS, but the SDPH is useful to find the safety risk of LTBs and prevent them in advance. The results also showed that the proposed SDPH was simple but had remarkable accuracy compared with the conventional BGS methods. These BGS methods are expected to increase the reliability of products used on the human body. Further, the BGS methods can be used to inspect the temperatures of static products in industrial processes.

Laser vision sensors for weld seam extraction face critical challenges due to arc light and spatter interference in welding environments. This paper presents a real-time weld seam extraction method. The proposed framework enhances robustness through the sequential processing of historical frame data. First, an initial noise-free laser stripe image of the weld seam is acquired prior to arc ignition, from which the laser stripe region and slope characteristics are extracted. Subsequently, during welding, a dynamic region of interest (ROI) is generated for the current frame based on the preceding frame, effectively suppressing spatter and arc interference. Within the ROI, adaptive Otsu thresholding segmentation and morphological filtering are applied to isolate the laser stripe. An optimized RANSAC algorithm, incorporating slope constraints derived from historical frames, is then employed to achieve robust laser stripe fitting. The geometric center coordinates of the weld seam are derived through the rigorous analysis of the optimized laser stripe profile. Experimental results from various types of weld seam extraction validated the accuracy and real-time performance of the proposed method.

Automatic surveillance systems have become essential tools for urban centers. These technologies enable intelligent monitoring that is both versatile and non-intrusive. Today, these systems can analyze various aspects, such as urban traffic, citizen behavior, and the detection of unusual activities. Most intelligent systems utilize photo sensors to gather information and assess situations. They analyze data sequences from these photo sensors over time to detect moving objects or other relevant information. In this context, background modeling approaches are crucial for efficiently detecting moving objects by differentiating between the foreground and background, which serves as the basis for further analysis. Although current methods are effective, the dynamic nature of outdoor environments can limit their performance due to numerous external variables that affect the collected information. This paper introduces a novel algorithm that uses mathematical morphology to create a background model by analyzing texture information in discrete spaces, leading to an efficient solution for the background subtraction task. The algorithm dynamically adjusts to global luminance conditions and effectively distinguishes texture information to label the foreground and background using morphological filters. A key advantage of this approach is its use of discrete working spaces, which enables faster implementation on standard hardware, making it suitable for a variety of devices. Finally, our proposal is tested against reference datasets of surveillance and common background subtraction algorithms, demonstrating that our method adapts better to outdoor conditions, making it more robust in detecting different moving objects.

Abstract This study proposes a method for recognizing sugarcane field ground height based on 3D-LiDAR and an improved RANSAC algorithm, aiming to accurately extract and identify the height of sugarcane growth points from the sugarcane field environment. First, ROI region extraction and Gaussian filtering preprocessing were performed on the experimental bench data collected by LiDAR. Based on the improved RANSAC algorithm and combined with the point cloud morphology characteristics, a multi-model parallel fitting strategy was adopted to simultaneously fit both ground and sugarcane models. The terrain obscured by sugarcane leaves was reconstructed through cubic polynomial fitting. Traversing the Z-direction minimum of the sugarcane point cloud and labeling the sugarcane growth points, morphological filtering was used to remove the residual noise, and finally calculating the ground height of the sugarcane growth points. The results of indoor simulations and sugarcane field experiments show that the ground height recognition algorithm can accurately identify the boundary point between sugarcane and the ground. The average errors between terrain inversion height and actual measured height are 0.09cm and 0.1cm, respectively. This provides reliable technical support for the adaptive adjustment of sugarcane harvester header height.

Developing method for generating accurate Digital Terrain Model (DTM) of streams is necessary due to the limitations of traditional field survey methods, which are time-consuming and costly and do not provide continuous data. The objective of this study was to develop an advanced method for generating high-quality DTM of streams using Structure from Motion (SfM) data. A leveling survey was conducted on four cross-sections of the Bokha stream in Icheon City, S. Korea, and SfM-based DTM was produced using the Pix4Dmapper program and Phantom 4 multispectral drone. Two vegetation filters (NDVI and NDI) and two morphological filters (ATIN and CSF) were applied to the data, and the best filter combination was identified based on MAE and RMSE analyses. The integration of NDVI and CSF showed the best performance for the vegetation area, while a single application of NDVI showed the lowest MAE for the bare area. The effectiveness of the SfM method in eliminating waterfront vegetation was confirmed, with an overall MAE of 0.299 m RMSE of 0.375 m. These findings suggest that generating DTMs of riparian zones can be achieved efficiently with a limited budget and time using the proposed methodology.