CSI Institute of Technology

Electric vehicles (EVs) are increasingly becoming crucial components of both transportation and energy sectors, necessitating efficient charging to support their growing use. A promising solution is integration of EV charging system with photovoltaic (PV) panels. This is because it offers cost savings, promotes environmental sustainability, and benefits from the continuous advancements in PV module efficiency. This research presents an innovative EV charging system with a novel Bidirectional Cuk converter and a bio-inspired Social Spider Optimized Proportional Integral (SSO-PI) controller. The proposed converter supports in managing the voltage flow between EV battery and grid, enabling charging and discharging operation. The SSO-PI controller effectively regulates the converter and offers better system performance with faster response time and stable control. Additionally, grid system is incorporated to charge EV battery at times of energy demand or failure of PV system. This integration ensures EV battery remains charged using either PV or grid supply, enhancing reliability and system sustainability. The proposed work modelled and simulated using MATLAB to validate its EFFICIENCY. Simulation outcomes reveals that the proposed accomplishes an enhanced efficiency of 96.38

The demand for efficient prediction methods to evaluate and treat post-COVID disorders has increased due to the COVID-19 epidemic. This work presents a novel strategy for anticipating and addressing these health risks that makes use of transformer-based models. To improve the disease prognosis accuracy, this study integrates detailed tabular data and X-ray images of chest, who pretentious with COVID-19.A hybrid transformer model that incorporates the most recent advancements helps to predict the course of the disease. This approach makes it possible for the model to successfully adjust to each patient’s distinctive characteristics and disease progression. Preliminary findings indicate that the proposed approach demonstrates promising results in accurately prognosing post-COVID diseases. Integrating diverse datasets significantly improves the model’s predictive capabilities and treatment efficacy, allowing for tailored recommendations that align with individual patient needs. The Vision Token Transformer (ViToT) architecture is a Hybrid transformer, which contains Vision Transformer in order to train a model with chest X-ray (CXR) images and Feature Tokernizer transformer in order extract features from tabular data to capable of recognizing and categorizing key patterns in CXR images while also extracting optimal features from tabular data. Late fusion technique is applied to combine an extracted feature which leads to cardiovascular disorder. Across 150 training epochs, the model demonstrated a robust performance, achieving a final accuracy of 97.6

Cerebral Microbleeds (CMBs) are small hemorrhages visible in Magnetic Resonance Imaging (MRI) scans and are closely linked to cerebral small vessel disease, stroke, and dementia risk. Detecting CMBs is challenging due to their minute size, heterogeneous appearance, and the presence of imaging artifacts that often mimic pathological signals. Existing detection approaches frequently exhibit limitations in sensitivity, specificity, and overall accuracy, reducing their clinical applicability. To address these challenges, this paper introduces a novel automatic detection framework, termed Hierarchical Triplanar Dual-Path Cascaded U-Net (HTpDPC-UNet), designed for robust and precise CMB identification in MRI scans. The proposed framework integrates a hierarchical triplanar architecture with a dual-path cascaded U-Net to leverage both multi-scale and multi-path feature representations. The Feature Concatenation Module (FCM) path extracts hierarchical features from axial, coronal, and sagittal views through multi-scale convolutions, effectively distinguishing subtle CMBs from surrounding tissues. Complementarily, the Hard Feature Exemplar Learning Module (HFELM) path employs exemplar-based learning and hard sample mining to improve discriminative capability, particularly for challenging CMB cases. A subsequent morphological location-scale identification stage enhances lesion localization and filters false positives, ensuring high reliability and clinical trustworthiness. Extensive experiments conducted on benchmark CMB, AIBL, and ADNI datasets demonstrate that the HTpDPC-UNet significantly outperforms state-of-the-art approaches in both detection and characterization tasks. Quantitative results highlight the superiority of our method, achieving 98.5% sensitivity, 98.2% specificity, and 98.6% detection accuracy, underscoring its potential for clinical translation in automated CMB diagnosis.

The Internet of Things (IoT) has enabled widespread connectivity of smart devices but remains susceptible to cyber intrusions. In this research, a novel dual feature optimization using deep learning network for intrusion Detection (FOUND) technique has been proposed for enhancing security in IoT environments. The proposed method utilizes the bald eagle search (BES) algorithm and butterfly optimization algorithm (BOA) to capture both flow and packet level features to enhance the accuracy of the intrusion detection process. Moreover, a multi-head attention-based bidirectional gated recurrent unit (MHA-BiGRU) is utilized to classify Attack and Non-Attack classes with high precision. The efficacy of the suggested approach is measured utilizing metrics including recall (RC), accuracy, precision (PR), and f1score (F1S). Experimental outcomes utilizing BoT-IoT and UNSW-NB15 datasets demonstrate greater accuracy over existing models. In BoT-IoT, the accuracy of the FOUND approach is 1.5