Image Area Research Articles

Attention and Multi-Scale Fusion for Shadow Removal of Embedded Basketball Players

Player shadows in basketball games show complexity and diversity due to various factors such as lighting conditions, player movements, camera positions, etc. In the shadow removal task, the attention mechanism aids the model in focusing on the key areas in the image, that is, the players themselves and the shadows around them, and may ignore some details, resulting in poor effects in the shadow removal of embedded basketball game players. To this end, a shadow removal method for embedded basketball game players with design attention and multi-scale fusion is proposed. By extracting the texture of the shadow image area of the local embedded basketball player, setting the gray value threshold of the center point of the image area window, obtaining the Local Binary Patterns (LBP) code of the local texture feature structure of the image, scanning the LBP code of the shadow image of the embedded basketball player, and obtaining the final characteristic parameters of the shadow image of the embedded basketball player, the pixels of the shadow image of the embedded basketball player are balanced and adjusted, and the size of the image is determined by the slope of the transformation function, and the shadow image of the embedded basketball player is within the normal proportion range. The consistency of multi-scale detail pixels is adjusted to achieve pixel segmentation of shadow images of embedded basketball players. Analyzing the basic principles of the attention mechanism, the shadow images of embedded basketball players are collected for key-value pairs to represent action information, and a key-value set is set. Encoding is carried out by the encoder in the attention mechanism, and the dependence of the foul action after encoding is calculated and decoded to determine the shadow image of the embedded basketball player to achieve shadow recognition by obtaining the shadow pixel point chromaticity space of the embedded basketball player color vector value, we determine the chromaticity similarity difference, compare the chromaticity similarity difference with the threshold, design a multi-scale fusion embedded basketball player shadow removal model, and implement the research. Experimental results show that the proposed method can effectively remove the shadow of embedded basketball players.

Optimization of Radiation Dose and Effective Dose for CT Pelvis Examinations in Main Hospitals in ALTaif Region

Introduction: CT scan is one of the medical imaging methods that irradiate patients with significant amounts of radiation. These amount of radiation doses must be well estimated if the imaging area is in part containing radiation-sensitive organs such as the pelvic area. Objectives: the objectives of the current study were to assess radiation dose during pelvis and Abdomen CT imaging and estimate the effective dose as well as to propose diagnostic reference level (DRL).Methodology: 200 adult patients irradiated in two major governmental hospital in Taif city, Saudi Arabia, patients’ demographic data from was collected such as weight, height, and age. Scanner specifications and scan parameters for each pelvis examination were recorded in special data collection sheet. Volume CT dose index (CTDIvol and dose length product (DLP) were utilized to estimate the radiation dose and effective dose. Microsoft Excel was used to analyse the data.Main Results: there was variation in scanning parameters among two hospitals under study and this result in variation in effective dose between two hospitals. The average DLP, CTDIvol and effective dose were 368.5, 390.7 mGy-cm,10.2,10.8 mGy and 7, 7.4 mSv for hospital one and two respectively. Conclusion: Based on the third quartile of DLP and CTDIw, the recommended DRL for both hospitals was 405 mGy-cm and 21.75 mGy, respectively. The findings revealed a reduced effective dose value when compared with previous studies.

Exploring discrepancies in muscle analysis with ImageJ: understanding the impact of tool selection on echo intensity and muscle area measurements.

The aim was to compare the use of different tools within the ImageJ program (polygon vs. segmented line) and their impact on the calculation of muscle area and echo intensity (EI) values in ultrasound imaging of the vastus lateralis muscle. Thirteen volunteers participated in this study. Ultrasound images of the vastus lateralis muscle were acquired using 2D B-mode ultrasonography and analyzed using both the polygon and segmented line tools by the same evaluator. The intraclass correlation coefficient (ICC) and coefficient of variation (CV) assessed the tools' reliability. Bland-Altman plots were employed to verify the agreement between measurements, and linear regression analysis determined proportional bias. A paired t-test was conducted to analyze differences between the tools. The reliability between tools for muscle area calculation was weak (r = 0.000; CV = 138.03 ± 0.34%), while it was excellent for EI (r = 0.871; CV = 15.19 ± 2.96%). The Bland-Altman plots indicated a large bias for muscle area (d = 195.2%) with a proportional bias (p < 0.001). For EI, the bias was (d = 15.2) with proportional bias (p = 0.028). The paired t-test revealed significant differences between the tools for area (p < 0.001) but not for EI (p = 0.060). The study found significant differences in measurements obtained with the polygon and segmented line tools in ImageJ, with the polygon tool showing higher values for muscle area and lower values for EI.

Leveraging CNNs for Automated Dermatological Diagnosis: Advancements in Skin Disease Prediction

Abstract: This is a research paper focused on skin disease prediction using CNN algorithms. Standing on eight classes of data this study accepts the challenge of proper diagnosis of multiple skin diseases. This problem statement supports the call for better and accurate ways of sorting the skin diseases so that dermatologists can diagnose the diseases appropriately and within the right time. With respect to the method, most of the data pre-processing are done on the collected dataset to improve its quality so as to develop a CNN model. Next, testing is performed with using the architecture of a CNN that is trained to search for features and patterns of the skin diseases’ photos. In this study, accuracy rates for each of the eight disease categories reach high figures, according to the results: so, it can be seen that the attainment of the CNN models is quite remarkable to classify diseases. For instance, the suggested method, stands as a fair one because it does not very much depend on the quality of the input image. The paper also examines the possibility of interpreting the CNN models concerning what kind of image features and area contribute to the classification process. This feature makes the model prediction more credible and yields useful information on the way the decision is made, possibly making doctors understand various aspects of pathophysiologic processes associated with different skin ailments. Furthermore, the study seeks to investigate the possibility of the proposed method in proximate actual clinical applications with identification always fast, and with massive data amount. In addition, solving these problems also contributes to the development of not only the dermatology as a branch of medicine, but also improves the understanding of the applicability of CNN based approaches for the analysis of medical images.

An experimental study on laser cleaning of the soot deposition layer on a white marble surface

A coordinated application of the image binarization, area extrapolation, and laser-induced breakdown spectroscopy (LIBS) detection for monitoring and controlling the laser cleaning soot deposition layer process on a white marble surface was reported. The laser cleaning process used a fiber pulsed laser beam with a fixed wavelength of 1064 nm, a pulse repetition frequency of 100 kHz, and a focal spot diameter of 95 μm. Image binarization was conducted to determine 150 ns as the appropriate fixed pulse width for laser cleaning of the soot deposition layers from four discrete values of 50 ns, 100 ns, 150 ns, and 200 ns. The optimal laser energy density of 2.82 J·cm-2 was obtained through area extrapolation, while the optimal spot overlap rate of 20 % was determined to maximize the effective laser cleaning speed. The optimal number of cleanings for a soot deposition layer with a thickness of 77μm was determined to be 5 using LIBS detection and to monitor the laser cleaning process in real time for each small area of the soot deposition layer. Image binarization was used to evaluate the overall cleaning quality of large marble surfaces. By dynamically adjusting the optimal number of cleanings, the cleaning efficiency of the soot deposition layers on half of the surface area of a 20 cm diameter white marble artifact exceeded 95 %. The results indicate that the synergistic use of various online monitoring methods can optimize multiple laser parameters, enabling precise removal of the soot deposition layer on the white marble surfaces caused by anthropic factors, without damaging its surface.

The selection of target areas for orbital imaging, as well as the application of diethylenetriaminepentaacetic acid orbital imaging, clinical factors, alkaline phosphatase, and thyrotropin receptor antibodies in the staging and grading of thyroid-associated ophthalmopathy, are subjects of research.

This study aimed to investigate whether patients with thyroid-associated ophthalmopathy (TAO) are in the active phase and explore their correlations with various clinical factors, serological tests, and orbital imaging parameters studied. It also aimed to evaluate the diagnostic value of different target areas in orbital imaging for assessing TAO inflammatory activity according to the American NOSPECS classification. Patients were stratified based on their NOSPECS grade to determine statistical differences among different stages. Receiver-operating characteristic (ROC) curves were used to assess the diagnostic efficacy of statistically significant data. Patients identified as active by orbital imaging were recommended for treatment to observe therapeutic outcomes. A total of 45 patients with thyroid-related eye disease who underwent 99mTc-diethylenetriaminepentaacetic acid single-photon emission computed tomography/computed tomography (CT) fusion imaging, alkaline phosphatase (ALP), thyrotropin receptor antibody (TRAb), free triiodothyronine, free thyroxine, thyroid-stimulating hormone serological tests, and clinical data collection from November 2023 to June 2024 were included. Clinical Activity Score (CAS) assessment, diplopia scores, and NOSPECS grading were conducted the day after imaging. Four rectus muscles were outlined as regions of interest on the optimal frame of extraocular muscle imaging and measured for maximum and mean radioactive counts, normalized against average counts of the patient's optic nerve level occipital bone region (UR1max, UR1mean). Tear gland radioactive uptake was similarly measured for maximum and mean counts normalized against the same region (UR2max, UR2mean). Exophthalmos and extraocular muscle thickening were assessed on CT scans. The correlations between different study parameters and CAS, as well as differences among different NOSPECS grades, were evaluated. ROC curves were used for statistically significant variables. Patients with discrepancies between orbital imaging and CAS results underwent treatment or follow-up to evaluate prognosis and staging. Active TAO status correlated significantly with UR1max, UR1mean, UR2max, UR2mean, ALP, TRAb, and age, using CAS and radiological imaging as standards. Among these, UR2max showed the strongest correlation. Significant differences (except ALP and TRAb) were found among different NOSPECS grades for the studied variables. Orbital volume, CAS results, radiological imaging, and diplopia scores also showed significant differences among different grades. Forty-two of 45 patients exhibited varying degrees of extraocular muscle thickening, confirming its reliability as a diagnostic criterion for TAO. The most common muscles affected were the inferior and medial rectus muscles, observed in 32 cases. High consistency was found between orbital imaging and CAS staging. Three patients diagnosed as active by imaging (including one with inconsistent CAS and imaging results) showed varying degrees of improvement posttreatment. Patients with nonactive imaging but inconsistent CAS results showed no progression during 3-month follow-up. ROC curves demonstrated high diagnostic efficacy for UR2max, with other indices also showing good diagnostic performance. Tear glands and extraocular muscles showed high diagnostic value, with tear glands demonstrating relatively higher value. Orbital imaging showed high consistency with CAS grading, providing an objective and comprehensive assessment of ocular involvement compared to CAS scoring alone. ALP and TRAb also demonstrated value in TAO staging, assisting clinical discrimination. Age may also play a significant role.

Area of Images of Measurable Sets on Depth 2 Carnot Manifolds with Sub-Lorentzian Structure

The paper is devoted to analysis of metric properties of images of measurable sets in sub-Lorentzian geometry introduced on Carnot manifolds. The current research continues the results obtained earlier for classes of compact sets on Carnot groups. The main difference is that, firstly, the mapping is defined on a measurable set (not necessarily compact), and, secondly, the preimage and image of the mapping do not have a group structure. Also, the definition of sub-Lorentzian analog of Hausdorff measure (which is not a measure in general) is modified: in contrast to earlier research, it does not require ``uniform'' sub-Riemannian differentiability. One of results is the property of quasi-additivity of this sub-Lorentzian analog. The latter enables to derive its parameterization by sub-Riemannian Hausdorff measure. In turn, this property means that the sub-Lorentzian analog of Hausdorff measure has classical properties of measure on certain class of sets. The sub-Lorentzian area formula on Carnot manifold is the main result of the paper. We also demonstrate the main ideas of its proof and show their specificity.

Timing Calibration of the Follow-up X-ray Telescope on board the EP satellite

Abstract The Follow-up X-ray Telescope (FXT) is one of the main scientific instruments on board the Einstein Probe (EP) astronomical satellite, which was launched in 2024 January. FXT consists of two Wolter I type nested telescopes (FXT-A and FXT-B) with a focal length of 1600 mm. The focal plane detector adopts PNCCD with 384×384 pixels within an image area of 28.8 mm × 28.8 mm. The timing mode of FXT serves as the primary operating mode for fast X-ray timing observations. To evaluate and validate the timing performance of FXT prior to launch, a comprehensive timing calibration was performed for FXT at the 100-m X-ray test facillity. By simulating various periodic Crab-like profiles using the Grid Controlled X-ray Tube (GCXT) with a pulsar simulation module, it was verified that the relative time accuracy of FXT surpasses 5×10-9. Furthermore, employing GCXT with a voltage pulse generation module allowed us to determine the time resolution of FXT-A and FXT-B as 45.6±2.7 μs and 47.1±2.8 μs, respectively. An absolute time calibration for FXT-B was carried out using GCXT and a time interval analyzer, revealing a measured time delay for FXT-B of 3.9±2.1 μs.

Applications of Denoising Diffusion Probability Model in Image Processing

Abstract. In 2020, Jonathan Ho et al. from the University of Berkeley proposed the Denoising Diffusion Probabilistic Models (DDPM), which improved on the shortcomings of the previous-generation Deformable Part Models (DPM). At the same time, they also surpassed previous generative models in image synthesis effects by using noise prediction, such as Generative Adversarial Networks (GAN), Variational Auto-Encoders (VAE), flow-based models and energy-based models, etc. After that, the denoising diffusion probabilistic models gradually received more discussions and research. In 2022, OpenAI launched the DALL-E 2, realizing Text-to-Image generation combined with the denoising diffusion probabilistic model. Two years later, with the release of Sora, the first Text-to-Video large models based on DALL-E, the denoising diffusion probabilistic models have received unprecedented attention in the multi-modal field. This article first provides the background and development process of the denoising diffusion probabilistic models. Secondly, it introduces the algorithmic principles of the diffusion models. Following that, it lists the applications of the models in the area of images. In the last section, it expounds on the advantages, disadvantages and future trends of the denoising diffusion probabilistic models.

Training ESRGAN with multi-scale attention U-Net discriminator

In this paper, we propose MSA-ESRGAN, a novel super-resolution model designed to enhance the perceptual quality of images. The key innovation of our approach lies in the integration of a multi-scale attention U-Net discriminator, which allows for more accurate differentiation between subject and background areas in images. By leveraging this architecture, MSA-ESRGAN surpasses traditional methods and several state-of-the-art super-resolution models in terms of Natural Image Quality Evaluator (NIQE) scores as well as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) across various benchmark datasets, including BSD100, Set5, Set14, Urban100, and OST300. Additionally, subjective evaluations further confirm the enhanced visual quality delivered by MSA-ESRGAN, particularly in terms of preserving texture and overall image realism. To ensure a fair comparison with Real-ESRGAN, we initialized our generator with a pre-trained Real-ESRNET model and followed the same training setup. Our model was trained on the DIV2K dataset using high-resolution image patches and the Adam optimizer, incorporating exponential moving average (EMA) for stability and performance enhancement. Evaluations on multiple benchmark datasets demonstrate that MSA-ESRGAN consistently delivers superior perceptual quality, as evidenced by higher NIQE, PSNR, and SSIM scores compared to other methods. Specifically, our model shows significant improvements in both objective and subjective measures of image quality. Furthermore, an ablation study highlighted the critical role of our multi-scale attention U-Net discriminator in enhancing the model’s performance. The results underscore the effectiveness of MSA-ESRGAN in maintaining image naturalness and perceptual quality, providing a robust benchmark for blind super-resolution tasks.

Multi-scale graph harmonies: Unleashing U-Net’s potential for medical image segmentation through contrastive learning

Medical image segmentation is essential for accurately representing tissues and organs in scans, improving diagnosis, guiding treatment, enabling quantitative analysis, and advancing AI-assisted healthcare. Organs and lesion areas in medical images have complex geometries and spatial relationships. Due to variations in the size and location of lesion areas, automatic segmentation faces significant challenges. While Convolutional Neural Networks (CNNs) and Transformers have proven effective in segmentation task, they still possess inherent limitations. Because these models treat images as regular grids or sequences of patches, they struggle to learn the geometric features of an image, which are essential for capturing irregularities and subtle details. In this paper we propose a novel segmentation model, MSGH, which utilizes Graph Neural Network (GNN) to fully exploit geometric representation for guiding image segmentation. In MSGH, we combine multi-scale features from Pyramid Feature and Graph Feature branches to facilitate information exchange across different networks. We also leverage graph contrastive representation learning to extract features through self-supervised learning to mitigate the impact of category imbalance in medical images. Moreover, we optimize the decoder by integrating Transformer to enhance the model’s capability in restoring the intricate image details feature. We conducted a comprehensive experimental study on ACDC, Synapse and BraTS datasets to validate the effectiveness and efficiency of MSGH. Our method achieved an improvement of 2.56–13.41%, 1.04–5.11% and 1.77–3.35% of dice on the three segmentation tasks respectively. The results demonstrate that our model consistently performs well compared with state-of-the-art models. The source code is accessible at https://github.com/Dorothywujie/MSGH.

Improved fast deconvolution algorithms based on functional beamforming for gas leakage sound source imaging

Fast deconvolution algorithms have high computational efficiency, but they often overlook spatial variations in the point spread function (PSF) and wrap-around errors, resulting in the degradation of sound source imaging performance. Accordingly, this paper proposes two improved fast deconvolution algorithms based on functional beamforming (FB). The proposed algorithms establish a linear equation involving the FB output, sound source distribution, and raised-power spatial shift-invariant PSF, specially with irregular focus grid point. Furthermore, iterative solutions are obtained based on fast Fourier transform, enhancing imaging performance while maintaining computational efficiency. Simulations and experimental results demonstrate that compared to the fast deconvolution algorithms, the proposed algorithms expand the effective imaging area to 36.3°, and the Rényi entropy is reduced by approximately 45 %, which has higher imaging spatial resolution and dynamic range. Finally, the compressed gas leakage source imaging results validate the effectiveness of the proposed algorithms, indicating that they have promising engineering application prospects.

A transferable approach for quantifying benthic fish sizes and densities in annotated underwater images

Abstract Benthic fishes are a common target of scientific monitoring but are difficult to quantify because of their close association to bottom habitats that are hard to access. Advances in image‐acquisition technologies, machine vision and deep learning have made capturing and quantifying fishes with cameras increasingly feasible. We present a method and open‐source software called ‘FishScale’ to estimate benthic fish lengths, numeric abundance and biomass density in underwater environments assessed with down‐looking monocular images. ‘FishScale’ estimates fish abundances and size frequencies from near‐nadir monocular images where fish have already been semantically segmented. The software accounts for lens distortion, underwater magnification effects and fish body curvature to automatically estimate fish lengths and the areas of images where they were captured. Numeric and biomass density are estimated through a deterministic machine vision algorithm that requires a user‐provided length‐weight relationship for species of interest and calibration images. Results from validation studies show that lengths and weights can be estimated with high accuracy and precision for round goby (Neogobius melanostomus) captured in distorted action camera images, and from large‐bodied lake trout (Salvelinus namaycush) imaged with a machine vision camera. The real‐world utility of the approach is demonstrated in a case study estimating round goby abundances and size frequencies along a 10.7‐km transect surveyed with an autonomous underwater vehicle in Lake Michigan, USA. Our validation studies demonstrate that the approach estimates benthic and benthopelagic fish lengths and weights with little bias and good accuracy and precision for species with much different body shapes and sizes. The method is applicable to data collected using a variety of nadir imaging approaches with widespread applications to fisheries monitoring and quantification of any species or object for which nadir images and working distances between the camera and feature of interest are available.

VLSI implementation of a modified min-max median filter using an area and power competent tritonic sorter for image denoising

The prominence of image processing in today’s cutting-edge technology is undeniable. Integrating software with hardware leverages both strengths, resulting in a real-time processing system that is efficient and streamlined. Raw images are usually affected by noise, which hinders the acquisition of good-quality and detailed images; hence, denoising becomes necessary. This paper proposes a modified min-max median (MMM) filter to remove impulse noise and a Tritonic sorter to localize corrupted pixels. The proposed denoising method focuses on localizing noisy pixels, unlike traditional denoising approaches, which focus only on noise detection and filtering. A min-max sheet provides the location of the corrupted pixels, and filtering is performed on them. The Tritonic Sorter, consisting of a max locator and a min locator, compares three input values and finds the minimum, maximum and median values among them. Compared to other state-of-the-art methods, the proposed method minimizes the number of comparators needed to carry out the sorting process. The proposed method was synthesized in the ZedBoard Zynq kit using the Vivado tool. The results show that the area improved by 27%, and the power improved by 16.23% compared with those of the existing method.

A Computational Protocol for the Knowledge-Based Assessment and Capture of Pathologies.

We propose that one of the main hurdles in delivering comprehensively informed care results from the challenges surrounding the extraction, representation, and retention of prior clinical experience and basic medical knowledge, as well as its translation into time- and context-informed actionable interventions. While emerging applications in artificial intelligence-based techniques, for example, large language models, offer impressive pattern association capabilities, they often fall short in producing human-readable explanations crucial to their integration into clinical care. Moreover, they require large well-defined and well-integrated data sets that typically conflict with the availability of such data in all but a few areas of medicine, for example, medical imaging and neuroimaging, noninvasive monitoring of bio-electrical activity, etc. In this chapter, we argue that approximate reasoning rooted in the knowledge that is explainable to the human clinician may offer attractive avenues for the introduction of such knowledge in a systematic way that supports formal retention, sharing, and reuse of new clinical and basic medical experience. We outline a conceptual protocol that targets the use of sparse and disparate data of different types and from different sources, seamlessly drawing on our collective experience and that of others. We illustrate the utility of such an integrative approach by applying the latter to the assessment and reconciliation of data from different experimental models, human and animal, in the example use case of a complex health condition.

Novel Quantification Protocol for Cardiovascular Calcification Progression Using Longitudinal MicroPET/MicroCT Images.

Micro positron emission tomography (PET) and micro computed tomography (CT) imaging are powerful, ideal research tools for following the progression of cardiovascular calcification. Due to their non-invasive nature, small research animals can be imaged at multiple time points. The challenge lies in the accurate quantification of cardiovascular calcification. Here, we provide a protocol, using images from the later disease stages as a template, to accurately quantify the progression of cardiovascular calcification in longitudinal studies. The protocol involves 1) the alignment of the chest area in multiple images from the same animal during a longitudinal study as the first step, 2) the definition of a region of interest (ROI) situated within the heart and the aorta at the site of larger calcium deposits that become apparent in later images, and 3) simultaneous segmentation and quantification of calcium deposits across all images acquired during the longitudinal study. This streamlined method enhances the accuracy of image analysis in following the progression of cardiovascular calcification by improving the precision of ROI definition and reducing the variability associated with earlier techniques that analyze individual scans independently.

Mathematical model of liver cirrhosis formation during morphological and molecular-genetic preclinical studies

BACKGROUND: Currently, researchers describe challenges in developing new treatments for fibrosis and cirrhosis: poor quality of preclinical models, insufficient trial duration, and lack of markers of therapeutic response. A separate task is to standardize the process of liver cirrhosis formation in preclinical trials, which is necessary to obtain accurate quantitative estimates in a short timeframe. AIM: This study aimed to develop a mathematical model for the formation of liver cirrhosis during preclinical trials. MATERIALS AND METHODS: Liver fibrosis and cirrhosis were induced in male Wistar rats using freshly prepared thioacetamide solution for 17 weeks. The area of connective tissue was determined as a percentage of the image area. The area of interlobular veins was measured in µm2. The numbers of cells expressing the FAP marker and the α-SMA marker were counted. The level of mRNA expression of the Vegfa and Yap1 genes was assessed by real-time polymerase chain reaction. A mathematical model for classifying observations into stages was constructed using multiple logistic regression with stepwise selection of predictors, followed by calculation of sensitivity, specificity, and area under the curve with a 95% confidence interval based on ROC analysis. RESULTS: As a result of the analysis, a mathematical model of liver cirrhosis formation was developed. The model is based on the values of two indicators: FAP+ cells and Yap1 mRNA and demonstrated good quality. The resulting value of the area under the ROC curve of 0.883 suggests good results for classifying cases. CONCLUSIONS: The mathematical model makes it possible to differentiate the stage of liver cirrhosis from the stage of fibrosis during preclinical studies. It provides a foundation for studying the pathogenesis of liver fibrosis and cirrhosis, identifying new potential molecular targets for antifibrotic therapy, and reducing the number of expensive, labor-intensive laboratory tests.

SteriCNN: Cloud native stego content sterilization framework

Modern robust steganography-based cyber attacks often bypass intrinsic cloud security measures, and contemporary steganalysis methods struggle to address these covert threats due to recent advancements in deep learning (DL)-based steganography techniques. Existing steganography removal methods are constrained by trade-offs involving high processing times, poor quality of sanitized images, and insufficient removal of steganographic content. This paper introduces SteriCNN, a lightweight deep residual neural network model designed for steganography removal. SteriCNN effectively eliminates embedded steganographic information while preserving the visual integrity of the sanitized images. We employ a series of convolutional blocks with three residual connections for feature extraction, feature learning, feature attention, and image reconstruction from the residue. The proposed model utilizes the correlation of channel features to achieve a faster learning rate, and by varying the dilation rate in convolutional blocks, the model achieves wider receptive fields, enabling it to cover larger areas of the input image at each layer. SteriCNN is targeted for blind image sterilization for real-time use cases due to its low training and prediction time costs. Our study shows impressive results for both traditional and deep learning-based stego vulnerabilities, with approximately 90% of steganograms eliminated while maintaining an average PSNR value of 46 dB and an SSIM of 0.99 when tested with popular steganography methods.

Identification of sweetpotato virus disease-infected leaves from field images using deep learning.

Sweetpotato virus disease (SPVD) is widespread and causes significant economic losses. Current diagnostic methods are either costly or labor-intensive, limiting both efficiency and scalability. The segmentation algorithm proposed in this study can rapidly and accurately identify SPVD lesions from field-captured photos of sweetpotato leaves. Two custom datasets, DS-1 and DS-2, are utilized, containing meticulously annotated images of sweetpotato leaves affected by SPVD. DS-1 is used for training, validation, and testing the model, while DS-2 is exclusively employed to validate the model's reliability. This study employs a deep learning-based semantic segmentation network, DeepLabV3+, integrated with an Attention Pyramid Fusion (APF) module. The APF module combines a channel attention mechanism with multi-scale feature fusion to enhance the model's performance in disease pixel segmentation. Additionally, a novel data augmentation technique is utilized to improve recognition accuracy in the edge background areas of real large images, addressing issues of poor segmentation precision in these regions. Transfer learning is applied to enhance the model's generalization capabilities. The experimental results indicate that the model, with 62.57M parameters and 253.92 Giga Floating Point Operations Per Second (GFLOPs), achieves a mean Intersection over Union (mIoU) of 94.63% and a mean accuracy (mAcc) of 96.99% on the DS-1 test set, and an mIoU of 78.59% and an mAcc of 79.47% on the DS-2 dataset. Ablation studies confirm the effectiveness of the proposed data augmentation and APF methods, while comparative experiments demonstrate the model's superiority across various metrics. The proposed method also exhibits excellent detection results in simulated scenarios. In summary, this study successfully deploys a deep learning framework to segment SPVD lesions from field images of sweetpotato foliage, which will contribute to the rapid and intelligent detection of sweetpotato diseases.

Spatio-temporal reconstruction of droplet impingement dynamics by means of color-coded glare points and deep learning

Abstract The present work introduces a deep learning approach for the three-dimensional reconstruction of the spatio-temporal dynamics of the gas–liquid interface on the basis of monocular images obtained via optical measurement techniques. The method is tested and evaluated at the example of liquid droplets impacting on structured solid substrates. The droplet dynamics are captured through high-speed imaging in an extended shadowgraphy setup with additional glare points from lateral light sources that encode further three-dimensional information of the gas–liquid interface in the images. A neural network is trained for the physically correct reconstruction of the droplet dynamics on a labeled dataset generated by synthetic image rendering on the basis of gas–liquid interface shapes obtained from direct numerical simulation. The employment of synthetic image rendering allows for the efficient generation of training data and circumvents the introduction of errors resulting from the inherent discrepancy of the droplet shapes between experiment and simulation. The accurate reconstruction of the three-dimensional shape of the gas–liquid interface during droplet impingement on the basis of images obtained in the experiment demonstrates the practicality of the presented approach based on neural networks and synthetic training data generation. The introduction of glare points from lateral light sources in the experiments is shown to improve the reconstruction accuracy, which indicates that the neural network learns to leverage the additional three-dimensional information encoded in the images for a more accurate depth estimation. By the successful reconstruction of obscured areas in the input images, it is demonstrated that the neural network has the capability to learn a physically correct interpolation of missing data from the numerical simulation. Furthermore, the physically reasonable reconstruction of unknown gas–liquid interface shapes for drop impact regimes that were not contained in the training dataset indicates that the neural network learned a versatile model of the involved two-phase flow phenomena during droplet impingement.