The pdf and printed version of article PV-BASED RAPID CHARGING AND BATTERY SWAPPING STATION FOR SMALL TRANSPORT EVS IN RURAL AREAS OF WEST BENGAL Bidrohi Bhattacharjee, Shibabrata Mukherjee, Rupanjali Bhattacharjee, Subha Bhattacharya, Ankur Ganguly, Arabinda Das Facta Universitatis, Series: Electronics and Energetics, Vol. 38, No 3, September 2025, pp. 469-485, contain an incorrect DOI: https://doi.org/10.2298/FUEE2503457B Correct DOI is: https://doi.org/10.2298/FUEE2503469B Link to the corrected article 10.2298/FUEE2503469B
Breast cancer is a widespread and life-threatening disease that occurs when abnormal cells in the breast tissue grow and divide uncontrollably. Breast screening using ultrasound is a good method. In this paper, a modified version of the multi-stage residual U-Net, referred as Residual U-Net-based Architecture for Breast Cancer Detection (RABC-Net), is proposed as a segmentation network for breast tumor detection from ultrasound images. This U-net variant is built using the ResNet architecture. In the model architecture the skip connections at multiple stages within the encoder-decoder network are incorporated strategically. These skip connections facilitate the preservation of fine-grained details from the encoder layers, enhancing the accuracy of the reconstructed images. Additionally, the decoder architecture is employing convolutional layers with regularization techniques and batch normalization, to ensure stable training and preventing overfitting. The model is trained using various types of activation functions. Finally, a modified version of the Mish activation function is proposed in this architecture, named Parametric Mish (P-Mish), which prevents neuron death and loss. The data imbalanced problem has been solved using a generative adversarial networks (GAN) model. The experimental results are demonstrating the effectiveness of the method in producing highfidelity reconstructions, outperforming existing approaches. The effectiveness of the proposed RABC-Net and classical segmentation models is evaluated on two different datasets such as BUSI (Breast Ultrasound Images) & UDIAT. On the BUSI and UDIAT datasets, experimental findings are showing that RABC-Net outperforms the other state-ofthe-art models. The RABC-Net architecture exhibited promising outcomes in automated breast cancer segmentation, demonstrating high accuracy and robustness across diverse imaging conditions.
This paper presents one solution for device configuration and calibration in remote power meter reading and controlling system. The configuration and calibration challenges in mass production settings are discussed, along with the implementation and core features of the Initz software. In real production environments, Initz is used for device configuration in several substations located in the regions of Novi Sad and Zaje & ccaron;ar.
The BT-CEAB2 Pulse Assessment Simulator was evaluatedusing a novelflexible Marija V Perga https://orcid org/0000 0001 6078 200 MXene/polyurethane resistive sensor positioned at the radial site during 1 h sessions in Absrracr: The BT-CEAB2 Puls Asessment Simultor was evalated usng a novel fl three operating modes: Palpation, Auscultation, and Palpation-to-Auscultation. For each MXene/polyurethane r sistive sensor positioned at th ra ial site during 1 h sessions in mode, approximately 1 min recordings acquired at 0, 30, and 60 min were analyzed in operating modes: Palpation, Auscultation, and Palpation-to-Auscultation For each m Python using baseline-drift removal and morphology-oriented filtering, with the HeartPy approx mately 1 min recordings acquired at 0, 30, nd 60 mi were analyoed i Py toolkit used for systolic peak detection and heart-rate estimation. Across all modes, the using baseline-drift removal and morphologyoriented filtering,withthe HeartPyoolki signals exhibited clear pulsatile patterns without dropouts or abrupt discontinuities. Inter-for systolic peak detectionand heart rate esimat on. A rossall modes,the ignals exh beat intervals stayed tightly clustered with only sparse isolated deviations, indicating c e r p l a i pat r highly stable pulse timing during sustained simulator operation. In contrast, systolic peak h u pot r a upt d n e te b t n e s igh ly lu r d i h nly p rs i lated d ia ins,ndi g highly able p l i during t i d i ulat p r i amplitudes showed larger relative variability, plausibly influenced by changes in sensor In cont t sytoli p k a plitud h d fixation, contact pressure,and the coupling between the simulator and the sensor over rltive ibilityluilifl dby cha i en fi atio t t s ur the coupling between the simulator and the sensor over time Representative pulse wavef time. Representative pulse waveforms retained a repeatable morphology with identifiable retained a repeatable morpho ogy with identifiable systolic and diastolic features ena systolic and diastolic features, enabling consistent systolic-to-diastolic timing metrics to consisten systolic o diast verify the simulator output.
Smartphones, as the most adaptable and widely used IoT devices, are increasingly targeted by sophisticated malware due to the sensitive data they carry. Traditional Android malware detection approaches, particularly those relying solely on static features such as permissions or solely on network flow, are often susceptible to evasion techniques. To address these limitations, this study introduces a hybrid deep learning framework named DroidNet, which effectively combines both static (Android permissions) and dynamic (network traffic flow) features through feature-level concatenation. By fusing data from both domains, DroidNet is capable of capturing comprehensive behavioral patterns-leveraging the declarative intent of an application (via permissions) alongside its runtime behavior (via network activity). This hybrid design significantly enhances the model's ability to distinguish between benign and malicious applications, even in cases where permissions alone may be deceptive or insufficient. Feature selection was performed using the Chi-Square test to retain only statistically relevant inputs, followed by standard preprocessing steps to ensure data integrity. While baseline classifiers such as K-Nearest Neighbors, Gaussian Na & uml;& imath;ve Bayes, and Support Vector Machines were tested on static features, they were outperformed by DroidNet. The proposed model achieved superior performance with an F1 Score of 97.05%, Accuracy of 99.7%, Cohen's Kappa Score of 0.9289, and an AUC-ROC Score of 0.9643-demonstrating its robustness and suitability for real-time Android malware detection in both smartphone and IoT environment.
