Diabetic Retinopathy (DR) is a major complication of Diabetes Mellitus (DM), which increases the possibility of vission loss. The proposes system provides a coherent framework integrating preprocessing, segmentation, feature extraction, and classification methods for DR prediction. Fundus image statistics from the public dataset were used to assess the proposed method. The Wavelet Packet Transform (WPT), which decomposes the images into a number of frequency components, is the primary step in the preprocessing method. WPT enhances both spatial and frequency-domain characteristics of retinal images. The corresponding retinal regions were subsequently identified by segmentation utilizing the Multivariate Kernel Sparse Extreme Learning Machine (MK-SELM) method. The MK-SELM framework improves its capacity to discriminate between DR and non-DR instances by projecting input data into a higher-dimensional space using a Radial Basis Function (RBF) kernel. Wavelet packet coefficients were extracted from segmented retinal regions and combined with domain-specific descriptors, including exudates, hemorrhages, and microaneurysms. A DenseNet-264 model was then used to categorize the data and determine the severity of DR. Using WPT for preprocessing and DenseNet-264 and MK-SELM for segmentation, an enhanced feature-extraction technique achieved an average of 98.9
This study presents an automated ultrasonic inspection method for detecting and quantifying interface defects in honeycomb composite laminate sandwich panels. Using immersion-based pulse-echo ultrasonic testing with full RF-waveform capture, ultrasonic data was acquired from Carbon Fiber Reinforced Polymer (CFRP) panels with a Nomex honeycomb core. Typical interface defects are created, such as missing adhesive, embedded Kapton and polytetrafluorethylene (PTFE) films, and missing core sections, with defects ranging in size from 4.7 mm to 40.8 mm. Two automation strategies to quantify the planar dimensions of the defect were developed: the first using the energy of the interface reflection and the second utilizing an effective facesheet thickness measurement based on the front and back wall reflections. Defect areas were extracted through image thresholding, and their shape was extracted and quantified via a custom algorithm implemented in MATLAB. Results show that the interface energy based detection method (EBD) achieved a consistently higher accuracy, with defect sizing errors typically below 1 mm with a maximum of 1.9 mm over all specimens investigated. In contrast, the thickness-based detection method (TBM) provided excellent accuracy for Kapton inserts (errors less than 0.65 mm) but exhibited larger deviations, up to 4.2 mm, for holes and PTFE defects due to a poor backwall signal. Results are presented that suggest that the backwall echo can be used for defect identification type. The automated techniques for dimensionalization demonstrated good performance across a wide defect size range, with a reliable detection and quantification of features as small as 4.7 mm and provide a reliable alternative to manual signal analysis.
Deforestation in tropical regions poses a significant threat to biodiversity, climate stability, and ecosystem sustainability. This study assesses deforestation susceptibility in Sanjay Gandhi National Park (SGNP), Maharashtra, India, using machine learning and remote sensing techniques. High-resolution satellite imagery from 2013 to 2024 and twelve deforestation determining factors (DDFs) were used to develop and evaluate four models: Random Forest (RF), XGBoost (XGB), Multi-Layer Perceptron (MLP), and Decision Tree Classifier (DTC). Results indicate that distance from forest edges, population density, and forest density are the most influential drivers of deforestation. Among the evaluated models, RF achieved the highest predictive performance (AUC = 0.907), followed by MLP (AUC = 0.899) and XGB (AUC = 0.896). The resulting deforestation susceptibility maps identify high-risk zones concentrated near roads, settlements, and agricultural land, providing valuable spatial insights for conservation planning and resource prioritization. The findings demonstrate the effectiveness of machine learning approaches for modelling deforestation susceptibility and offer practical guidance for policymakers and conservation managers seeking to support sustainable forest management in SGNP and similar ecosystems.
Malaria diagnosis in resource-limited settings requires sensing platforms capable of resolving small changes in the optical properties of biological samples while maintaining simple label-free operation. This study presents a BiB₃O₆/LiB₃O₅-assisted multilayer surface plasmon resonance (SPR) architecture combined with machine-learning-based response prediction for malaria-related refractive-index sensing. The proposed configuration consists of a BK7 prism, Ag plasmonic layer, BiB₃O₆ dielectric layer, and LiB₃O₅ interface layer, whose thicknesses are systematically investigated to control plasmonic coupling, resonance characteristics, and electromagnetic-field confinement. The optical response is evaluated using the Transfer Matrix Method and finite-element simulations, with the dispersive response of Ag incorporated through the Drude–Lorentz model. Based on the reported parametric analyses, a selected configuration of Ag (51 nm)/BiB₃O₆ (6.3 nm)/LiB₃O₅ (1.1 nm) is used for subsequent refractive-index sensing analysis. The sensor exhibits a maximum angular sensitivity of 514.286°/RIU, a figure of merit of 174.334 RIU⁻¹, and a quality factor of 27.898 over the investigated analyte refractive-index range of 1.373–1.402 RIU. The strong electric-field localization near the sensing interface further demonstrates enhanced interaction with changes in the surrounding dielectric environment. In addition, a Gradient Boosting Regressor (GBR) is employed to predict the simulated sensor responses for LiB₃O₅ thickness and analyte refractive-index variations, achieving independent test-set R² values above 0.979 with MAPE below 1
In this paper the Transfer Matrix Method and the Finite Element Method (FEM) are used to theoretically analyse a double-defect one-dimensional photonic crystal (1D-PC) structure as a high-performance gas sensor. The proposed structure is of the form (Air/(BA)ᴺ-BDBD(BA)ᴺB/Air) and is composed of alternating layers of Germanium (Ge) and Lithium Fluoride (LiF) with defect cavities as the sensing region. The coupling of the two identical defect cavities gives rise to two linked resonance modes in the photonic bandgap. The results of both computational methods confirm the accuracy of the suggested model. The effect of the periodic layers on the resonance characteristics shows narrowing of the resonance line width with increasing number of layers. Also, a considerable increase in the quality factor is observed with the rise of the mirror periods; consequently, an optimized period of N = 8 is selected for sensing analysis based on spectral performance with practical detectability. The resonance linewidths are 1.15 femtometers and 1.44 femtometers with quality factors (Q) over 109. The refractive index of the defect cavity is modulated from 1.000 to 1.005 to analyse the sensing performance. The sensitivities corresponding to two resonance peaks are obtained as 1075.58 nm/RIU and 1351.35 nm/RIU, respectively. The ultra-narrow resonance linewidths result in the figures of merit reaching extraordinarily high values of 9.37×108 RIU−1 and 8.90×108 RIU−1 for the two modes. The strong optical confinement, high sensitivity and outstanding sensing capability show that the proposed double-defect photonic crystal structure has the potential to be used in ultra-high resolution refractive index and gas sensing applications. The obtained results represent the maximum performance that can be achieved under ideal loss-less conditions and provide a foundation for designing future photonic crystal sensors.
This work introduces a high-sensitivity chemical sensor for terahertz (THz) applications using a dual-core photonic crystal fiber (DC-PCF) structure. Two solid cores are separated by a circular air hole in the DC-PCF. Materials such as water, ethanol, and benzene can be used to fill this hole. Both elliptical and circular air holes in the design employ exact geometric forms. Comparing this design to conventional photonic crystal fibers (PCFs), the sensitivity is much higher due to improved light confinement and mode coupling. When the chemical composition inside the circular hole changes, the sensor’s transmission spectrum noticeably varies, according to numerical study using finite element methods (FEM) in COMSOL Multiphysics. We demonstrate the exceptional performance of the sensor by achieving a maximum sensitivity of 413,323.12 nm/RIU for a 0.32 m fiber length. The DC-PCF sensor offers great promise for real-time and economical chemical detection in industrial and environmental contexts due to its small size, simplicity of construction, and excellent detection capabilities.
The burden of cancer on society is increasing worldwide. The leading cause of cancer-related deaths is detection at a late stage. The search for biomarkers for the early detection of cancer is important. Cancer cell detection via photonic sensing technology is useful for early detection. We propose a light-splitting ring-resonator-coupled Pooja Vikrant Deshmukh design of a photonic crystal sensor for cancer detection. Photonic crystals provide designs with dimensions in the nanometer to micrometer range, making them viable for use in sensor design. We designed a sensor consisting of three waveguides coupled with point defects that act as ring resonators. This unique design provides optimal biomatter interactions. The radius of the ring resonator rods was varied to obtain four different designs. A Monte Carlo simulation was performed to determine the optimized design, and a noise analysis was performed for varying radii. An FDTD simulation was carried out for the optimized design, and a spectrum was obtained for the cancerous cells. This spectrum differentiates cancerous cells from normal cells. The relative sensitivity, figure of merit, limit of detection, and transmission efficiency were measured to be 6435.11 nm/RIU, 8043.89, 6.91e−06, and 99.52
The global impact of COVID-19 has highlighted the urgent need for highly accurate and sensitive diagnostic methods and tools. In this study, we present a surface plasmon resonance (SPR) biosensor designed to enhance the detection of SARS-CoV-2 through optimized material layering. The proposed biosensor incorporates a multilayer structure consisting of copper (Cu), titanium dioxide (TiO₂), silicon nitride (Si₃N₄), and a black phosphorus (BP) layer. After systematic optimization, the ideal thicknesses were determined to be 70 nm for Cu, 13.9 nm for TiO₂, 6.9 nm for Si₃N₄, and 0.47 nm for BP. This configuration demonstrated strong performance in a phosphate-buffered saline (PBS) environment. The sensor achieved a high sensitivity of 500.00 deg/RIU, along with a detection accuracy (DA) of 0.002 deg⁻¹ and a quality factor (QF) of 186.36 RIU⁻¹ at a concentration of 0.10 mM of SARS-CoV-2. Additionally, the biosensor exhibited a linear response to refractive index (RI) variations, indicating its capability for reliable quantification of viral concentrations. These results confirm that the integration of TiO₂, Si₃N₄, and BP layers significantly enhances the performance of SPR biosensors. The proposed design offers improved sensitivity, resolution, and signal quality, making it a promising candidate for early-stage detection and clinical diagnostics, as well as for broader applications in syndromic surveillance.
Distributed sensor networks (DSNs) with multimodal sensors deployed in dense urban environments require drone detection algorithms that are both reliable under poor signal conditions and feasible for execution on resource-constrained edge devices. Although multimodal sensing can improve detection robustness, most existing approaches rely on parallel processing and fusion strategies that incur significant computational overhead and complex inter-modality dependencies. This paper presents a sequential radar–acoustic drone detection framework tailored for low-complexity edge deployment. The proposed method employs radar as the primary detection stage and activates acoustic analysis only when radar confidence falls within a predefined uncertainty margin. Each modality is processed independently using a shallow convolutional neural network. Confidence thresholds derived from radar classifier statistics regulate the sequential decision process. Experimental results demonstrate that the proposed sequential pipeline preserves high specificity while reducing missed detections compared to radar-only processing. Quantitative evaluation shows that the sequential architecture reduces false negatives by up to 41
A novel dumbbell-shaped photonic crystal fiber surface plasmon resonance (PCF-SPR) sensor is proposed for the simultaneous detection of refractive index (RI), temperature, and magnetic field via four independent sensing channels. The sensor structure incorporates gold and silver nanowires into spatially separated regions, facilitating distinct plasmonic excitations. Higher-order core modes are employed to enhance resonance strength by increasing energy transfer to surface plasmon modes. The influence of structural parameters on sensor performance was analyzed using the finite element method (FEM), and the parameters were subsequently optimized to improve sensitivity. The results demonstrate that the sensor achieves high refractive index sensitivity, with maximum wavelength sensitivities of 14,000 nm/RIU and 10,900 nm/RIU in two dedicated channels, enabling effective differentiation between cancerous and normal cells (4,100–8,357 nm/RIU). For temperature measurement, the sensor exhibits a linear response in the range of − 40 °C to 100 °C, with a sensitivity of 2.35 nm/°C. In magnetic field sensing, a sensitivity of 100 pm/Oe is achieved over the range of 40–200 Oe. The proposed sensor offers high sensitivity, strong modal resonance, and multi-parameter sensing capability within a compact fiber structure, demonstrating strong potential for applications in biomedical diagnostics, environmental monitoring, and industrial process control.
Malaria is still a major global health concern, especially in tropical and subtropical areas where morbidity and mortality are greatly increased by delayed diagnosis. in order to limit the spread of the disease and avoid serious complications, early detection is crucial. For the detection of phase-resolved red blood cells (RBCs), a high-sensitivity surface plasmon resonance (SPR) biosensor based on a BK7 prism–Ag layer integrated with a BP–Franckeite–BP van der Waals heterostructure is numerically investigated. Strong surface plasmon polariton (SPP) excitation and enhanced electric-field confinement at the metal–2D material interface are the reasons for the significantly improved sensing performance of the proposed Structure-4. The obtained sensitivities for the schizont, trophozoite, and ring phases are roughly 259.96, 282.61, and 331.21 deg/RIU, respectively. A stable figure of merit (FoM) of 48.05, 47.82, and 47.52 /RIU is achieved for the detection of the same RBCs. Additionally, the penetration depth (PD) of 178.09 nm for normal and 196.34 nm for Ring phase is attained, allowing for efficient interaction with intracellular RI variations and RBC membranes. The enhanced sensitivity, FoM, and PD show that Structure-4 is an excellent choice for phase-specific and label-free biomedical sensing.
Metastasis (MET) is a critical stage of cancer progression in which malignant cells spread from the primary tumor site to distant organs, affecting millions of people worldwide. Early diagnosis is critical for proper treatment to limit the risk of death associated with the disease. Therefore, it is essential to have a non-invasive test that can diagnose MET with ultra-high sensitivity and rapid detection. In this study, a surface plasmon resonance (SPR)-assisted optical fiber sensor incorporating silver (Ag), platinum sulfide (PtS2), and platinum diselenide (PtSe2) as functional layers is proposed for the highly effective detection of metastatic liver tissue. The refractive index values for normal and metastatic liver tissues were obtained from the literature and employed in the numerical analysis. The sensor performance was investigated using the transfer matrix method (TMM) under wavelength interrogation by analyzing the reflection coefficient and transmitted power characteristics. The thicknesses of the Ag and PtS2 layers were optimized to maximize the sensing performance. The proposed multilayer architecture exploits the synergistic interaction between the Ag plasmonic layer and PtS2/PtSe2 functional layers, resulting in enhanced electromagnetic field confinement and stronger light–matter interactions at the sensing interface. The optimized structure, consisting of a 50 nm Ag layer and a 0.5 nm PtSe₂ layer, achieved a maximum sensitivity of 24,306 nm/RIU with a figure of merit (FoM) of 97.5 RIU ⁻ for detecting metastatic liver tissue. A comparative study with recently reported SPR cancer biosensors demonstrated the competitive sensing performance of the proposed design. The data obtained show that the proposed sensor is a potential platform for rapid, label-free, and high-sensitivity cancer diagnosis and biomedical sensing applications.
Aging-in-Place monitoring in small, high-density apartments faces three persistent engineering constraints: severe occlusions, strict privacy requirements, and unstable home networks. Under these conditions, cloud-centered closed-loop schemes become unreliable, while alerts are often difficult to verify. To address these issues, this paper proposes an Edge-First Digital Twin-based Cyber-Physical Home System (DT-CPHS), in which uplink semantic synchronization is explicitly decoupled from downlink local control. The system uploads only low-dimensional semantic state vectors and evidence metadata, while alert auditing and replay are supported through an event-sourced Twin Event Log. To improve robustness under occlusion and conflicting sensor indications, DT-CPHS adopts decision-level multi-source fusion based on Dempster–Shafer evidence theory. In parallel, a connectivity-aware local rule engine provides deterministic offline fallback, together with hysteresis-based anti-flapping and seq-based backfill for consistency maintenance. A full-stack prototype was implemented in a real apartment using non-intrusive heterogeneous sensing (mmWave radars, wearable, PIR/door sensors) and a low-cost gateway (Raspberry Pi 4) with HA/Node-RED/MQTT. Experimental results validate four engineering goals. First, for offline survivability, local safety actions maintained 100
The setting degree and interlayer strength of cement concrete are key indicators governing construction efficiency and structural safety, both of which are highly dependent on concrete moisture content. Therefore, this study developed a novel fringing electric field-based array sensor for depth-resolved moisture detection of cement concrete. A geometric model of the sensor array was simulated within COMSOL, and its structural configuration was optimized via orthogonal experimental design. With the optimal parameter configuration, the device achieved a penetration depth of 66.86 mm, a signal strength of 11.387pF, and a sensitivity of 0.267pF/
Efficient irrigation management and timely plant disease identification are essential for improving agricultural productivity while minimizing water consumption. Typical smart agriculture systems based on the Internet of Things (IoT) perform these tasks individually and are unable to offer holistic decision support. In this study, we propose an adaptive agriculture system based on IoT, which consists of LSTM+XGBoost for irrigation prediction and Lightweight CNN for plant disease classification, in a single decision-support framework. The irrigation prediction model was trained using the IoT sensor data, which consists of soil moisture, soil temperature, soil humidity, and environmental variables; the disease classification model was evaluated with the PlantVillage dataset, which contains more than 54,000 images of plant leaves belonging to 38 disease–crop classes. Experiments were conducted with a data splitting of 70:15:15 and 5fold cross-validation. The proposed framework resulted in a classification accuracy of 93.5
The design and development of a potable, sensitive Circular Cylindrical Resonator (CCR) based dual port sensor for the detection of mustard oil as Oil under Test (OuT) with palm oil as Adulterant under Test (AuT) characterisation is presented in this article. High sensitivity, compactness, and optimization are the main considerations for sensor design in the current state-of-the-art development. The integration of a CCR-based dual-port resonator with a Least Squares Estimation framework for quantitative dielectric characterisation of edible oil adulteration is the feature that makes the designed sensor distinctive. The suggested sensor offers a affordable substitute for edible oil characterization, low sample volume (< 2 mL), experimental validation, and mathematical permittivity estimation in a single measurement platform. To examine the characteristics of the oil sample under test (SuT), a cylindrical cavity of acrylic material with a radius of 14 mm has been introduced above the patch. With dimensions of 51.5 × 40 × 1.6 mm³, the sensor is meant to function at a resonance frequency of 2.98 GHz. The least squares approach has also been used to confirm the device’s sensing capabilities. Additionally, the experimental findings validated the sensor’s precision in identifying the AuT. The maximum achieved sensitivity of the sensor is 0.21. These excellent performance outcomes and the close correspondence between simulation and actual test data demonstrate how successfully the sensor detects the Adulteration.
Rapid, label-free detection of acetone is essential for environmental monitoring, industrial safety, and non-invasive biomedical diagnostics, motivating the development of highly sensitive terahertz (THz) sensing platforms. This work proposes a hybrid multi-material THz metasurface biosensor comprising a silver-coated plus-shaped resonator, symmetrically arranged molybdenum ditelluride/copper-coated star resonators, tungsten diselenide-coated rectangular resonators, and graphene/gold layers on a silicon dioxide substrate. The sensor is analyzed using full-wave finite-element simulations in COMSOL Multiphysics, supported by coupled-mode theory, transfer matrix modeling, and Floquet–Bloch periodic boundary conditions. Graphene chemical-potential tuning (0.1–0.9 eV) yields a modulation-depth variation from 0.094
Surface Plasmon Resonance (SPR) and Localized Surface Plasmon Resonance (LSPR)-based probes are gaining traction as sensors for the detection of various analytes. Here, we have synthesized L-tyrosine-capped silver nanoparticles, which are more affordable and have a longer shelf life than gold nanoparticles. The synthesized nanoparticles have also been used to detect Mg2+ ions in water at a basic pH (10–13). The Mg2+ ions were detected by naked-eye colorimetry, smartphone digital image analysis, and UV-Visible spectroscopy. The shape and size of the synthesized nanoparticles were also studied at various stages of detection using HR-TEM images and DLS, which clearly showed nanoparticle aggregation upon binding to Mg2+ ions. The limit of detection values were observed to be 4.56 ± 0.06 µM (digital image analysis) and 1.17 ± 0.11 µM (UV-Visible spectroscopy), both well below the permissible limit. The L-TyrAgNPs were also able to detect Mg2+ ions in spiked real water samples (tap, drinking, and groundwater). The overall results signify that the synthesized nanoparticles are a selective LSPR-based probe for colorimetric detection of magnesium ions in water.
Low-dose and sparse-angle computed tomography (CT) reduces radiation exposure but makes image reconstruction challenging due to noisy and limited projection data. Popular reconstruction methods are based on two-stage approaches, typically involving filtered backprojection (FBP) followed by a neural network to enhance the image. FBP, however, amplifies noise and struggles with irregular sampling. Therefore, we explore filter-free initial reconstructions, shifting the filtering step to the neural network. In particular, we investigate how two-stage methods can be adapted for cases where implementing explicit filters is difficult, such as with irregular sampling. Specifically, we propose backprojection (BP) or a small number of Landweber iterations as the initial reconstruction, followed by a fine-tuned DRUNet model, referred to as BP-DRUNet and Landweber-DRUNet, respectively. For evaluation, we consider both regular and irregular sampling conditions: For regular sampling, we compare BP-DRUNet with FBP-DRUNet (using FBP as the initial stage) in order to benchmark against standard two-stage approaches. BP-DRUNet performs comparably to FBP-DRUNet under regular sampling. In irregular sampling, Landweber-DRUNet improves reconstruction quality with more iterations, though at the cost of longer training and inference times. Experiments are carried out on synthetic and real CT datasets with parallel- and fan-beam acquisitions across different sparse-angle setups.
A neurological disorder refers to any condition that disrupts the structure and normal functioning of the brain, spinal cord, or nerves, which are the key components of the nervous system. Detecting such disorders is often challenging, since symptoms frequently overlap, early indicators are often subtle, and the brain’s intricate functions differ greatly across individuals. Therefore, a new approach called Stacked Xception Convolutional Neural Network (SXcp-CNN) is devised for detecting brain neurological disorders utilizing Magnetic Resonance Imaging (MRI). Initially, the input MRI image is passed to noise reduction using the Median Filter to eliminate unwanted artifacts. Concurrently, Histogram Normalization is applied to enhance image contrast and improve visual quality. Afterwards, image augmentation is conducted by Generative Adversarial Networks (GAN), flipping, and rotation. Subsequently, image segmentation is accomplished by exploiting Znet with Weighted Tversky loss (WTL), developed by combining Weighted Binary Cross-Entropy loss (WBCE) and Focal Tversky loss. Next, the extraction of features is accomplished by exploiting EfficientNet-B0 and Haralick texture features. Finally, brain neurological disorder detection is done using the proposed SXcp-CNN, designed by integrating Stacked CNN (S-CNN), XceptionNet, and the Taylor concept. Furthermore, the proposed SXcp-CNN attains a better accuracy, True Positive Rate (TPR), True Negative Rate (TNR), F1-Score and precision of 96.703
