Automated gastrointestinal (GI) image classification remains challenging because of subtle morphological variation, high inter-class similarity, and the computational constraints of clinical edge devices. We present CA-MobileNetV4, a parameter-efficient teacher-student framework that combines a MobileNetV4-Small student, Channel-Prior Convolutional Attention (CPCA), a ConvNeXt-Tiny teacher enhanced with ACmix and Efficient Channel Attention (ECA), and Class Attention Transfer-based Knowledge Distillation (CAT-KD). CPCA is inserted in Layers 3 and 4 of the student to combine channel priors with multi-scale strip convolutions. During distillation, the pre-trained teacher is frozen; class activation maps (CAMs) from four semantic stages are projected to the class dimension, adaptively pooled to matching spatial sizes, L2-normalized, and aligned across all classes. Focal Loss is used as the classification term, and the CAT loss weight is set to β = 10 after evaluation of 5, 10, 15. Experiments use six Kvasir-v2 classes and eleven Lower-GI Hyper-Kvasir classes with stratified 70
This work presents ZnO nanomaterial-based thin-film fractional-capacitive sensors for measuring urea content in soil. ZnO nanoparticles were synthesized and confirmed to form a well-crystalline hexagonal wurtzite structure by XRD, with SEM imaging showing a densely agglomerated, faceted morphology. Sensors have been fabricated with an epoxy–ZnO nanocomposite dielectric layer with aluminium foil interdigital electrodes at three ZnO concentrations (1.25, 2.5, and 5 wt
This paper proposes a compact low-cost two-port planar microwave sensor on a commercially available substrate to estimate soil moisture in a non-intrusive manner. The proposed sensor consists of a 50-Ω microstrip feed line with two symmetric C-shaped split-ring resonators (SRRs) in order to generate strong localized electric-field confinement around the split-gap regions. This localized evanescent field interacts with the adjacent soil sample and the resonant transmission responds to changes in the effective relative permittivity of the overlying material. Instead of measuring the resonance-frequency shift that requires broadband swept measurements with limited resolution at high moisture levels, the sensor uses the phase of the S21 transmission parameter measured at a single fixed frequency of 2.65 GHz to produce a monotonically increasing sensing output that is nearly linear. A total normalized phase shift of 28.5° over the range of relative permittivity from 6 to 30, which corresponds to the range of interest for normal moist to wet soil conditions, was obtained by electromagnetic simulations, along with an approximately linear phase sensitivity of 2°/unit permittivity over the lower moisture levels. The sensor was then fabricated on a standard FR-4 substrate (εr = 4.4) and characterized by a calibrated vector network analyzer (VNA). The measurements on the three separately-prepared soil samples resulted in a normalized phase shift of 0° to 16.5° with a mean standard deviation of ±0.71° for all samples, indicating good repeatability. A systematic parametric study of the sample-container geometry shows that the phase response depends mainly on resonator geometry and is slightly influenced by dimensions of simulated sample-containers suggesting some tolerance to practical container dimensional changes. The proposed sensor has a competitive combination of an economical substrate, a small footprint, and fixed-frequency measurement simplicity which suggests the potential use for precision-agriculture soil moisture sensing nodes with further experimental calibration.
Self-similar metasurfaces have been considered as one of the best solutions for manipulate and control the light to enhance the sensing accuracy such as viruses, bacteria and anti-bodies in the optical range. In this paper, a combination of two different size of diamond elements array are used to design a self-similar plasmonic metasurface to achieve the Fano response at mid-infrared spectrum including 1900 and 2100 nm. The structure has asymmetric formation and thus result has dependence to TE and TM mode and this feature is considered for developing the optical switch with ratios of 6.2 and 13.8 for 1900 and 2100 nm respectively. The Coupled Mode Theory (CMT) is employed to describe the interaction between the resonant modes generated within the proposed metasurface absorber. The Fano responses of the proposed metasurface makes it interesting for sensing application and so it is examined as a sensor for Flu virus detecting. The sensitivity and Figure of Merit (FOM) for the proposed optical sensor is obtained 660 to 903 nm/RIU and 25 up to 110 RIU− 1 corresponding with material with refractive index of 1.1 to 1.5. The virus ratio finally is checked with this metasurface which shows a high resolution.
Surface plasmon resonance (SPR) is a highly sensitive optical technique for the label-free, real-time detection of biomolecular interactions. In this paper, the finite-difference time-domain (FDTD) method was employed to model and optimize the performance of a gold-based SPR sensor functionalized with glucose oxidase (GOx) for glucose monitoring. The base structure consisted of a BK7 glass substrate, a 2 nm chromium adhesion layer, and a 50 nm gold film. A multilayer functionalization strategy was implemented, incorporating a plasma-polymerized heptylamine (HApp) interlayer and a glutaraldehyde (GA) crosslinker to immobilize the GOx enzyme. Systematic optimization of the sensing architecture determined that a 10 nm HApp layer, a 2 nm GA layer, and a 5 nm GOx layer provided the most balanced optical response between resonance depth and dip sharpness. The simulation results demonstrated that GOx functionalization produced significant resonance angle shifts, achieving a peak sensitivity of 147.167 °/RIU. A strong linear correlation was observed between glucose concentration refractive index and resonance angle shift, highlighting the sensor’s potential for quantitative detection. Furthermore, the functionalized sensor of Au/HApp/GA/GOx exhibited a 1.33-fold sensitivity enhancement compared to a bare gold sensor. These findings confirm the potential of the proposed design for glucose monitoring in urine samples and highlight the utility of FDTD as a predictive tool for biosensor development.
