Robust periocular recognition in contactless settings remains challenging due to cross-sensor and cross-spectral appearance variations, as well as unconstrained factors such as pose, blur, and occlusion. This paper proposes a parameter-efficient periocular recognition framework based on a DINOv2-pretrained foundation Vision Transformer (ViT) adapted using Low-Rank Adaptation (LoRA). The LoRA modules are injected into the query, key, and value projections of the transformer backbone, while an ArcFace head is employed to learn identity-discriminative embeddings. During enrollment, multiple gallery samples of each subject are encoded and combined using a Quality-Aware Gated Fusion (QAGF) strategy to construct a reliability-aware identity template, where informative samples receive higher contribution and degraded samples are automatically down-weighted. The proposed framework is evaluated on two benchmark periocular datasets, UBIPr and UTIRIS, under visible-spectrum, near-infrared-spectrum, and cross-spectrum protocols. Ablation studies involving frozen backbone transfer, full fine tuning, and LoRA-based parameter-efficient fine-tuning further validate the effectiveness of the proposed adaptation strategy in terms of both recognition performance and trainable parameter reduction. Experimental results demonstrate consistent improvements over CNN baselines and representative prior methods, achieving an Equal Error Rate (EER) as low as 1.2% and a True Accept Rate (TAR) of 99.1% at False Accept Rate (FAR) = 10-3, while reducing trainable parameters by more than 90%. These results indicate that LoRA-adapted foundation vision transformers with quality-aware template fusion provide an effective and scalable solution for next-generation periocular biometric recognition.
Leukemia is a bone marrow-derived malignancy that is distinguished by aberrant white blood cell proliferation. Leukemic cells proliferate rapidly in the blood and travel to the brain, neurological system, spleen, liver, and lymph nodes, among other areas of the body. In the early stages of the disease, leukemia is difficult to diagnose because current technologies are less time-efficient. To overcome this, a novel LUK-SEGNET has been proposed for leukemia classification using blood smear images from the ALL-IDB dataset. The input images are pre-processed by resizing and cropping black edges from the blood smear images. The Duo-SegNet feature extraction technique extracts the high-dimensional features from the images. The proposed method uses an elephant herding optimization algorithm to select the relevant features from the images. The performance of the proposed method is evaluated using accuracy, precision, recall, specificity and f1-score. The proposed LUK-SEGNET method achieves an accuracy rate of 99.74% in the normal class and 99.87% in the abnormal classes. The proposed LUK-SEGNET method achieves an overall accuracy of 0.05%, 0.19%, and 0.11% better than DL4ALL, ResNet-152 and Inception V3, respectively.
This study presents a comprehensive numerical design and small-signal modeling of a 100-nm T-gate AlGaN/GaN Fin High Electron Mobility Transistor (HEMT) incorporating a Transitional Recess Gate (TRG) with recess depths ranging from 5 to 20 nm. The device structure, modeled using Silvaco TCAD, integrates an n-doped GaN cap layer to enhance carrier transport and electrostatic control. The numerical simulations were performed under varied gate and drain biases to analyze the impact of recess depth and cap doping on the intrinsic device parameters. The extracted small-signal parameters ( gm , gd , Cgs , and Cgd) were obtained using the Dambrine method and validated across multiple bias conditions. The numerically derived small-signal model accurately reproduces the simulated S-parameter characteristics and provides physical insight into the bias-dependent transconductance and output conductance variations. Results show that the TRG configuration achieves improved transconductance linearity and an extended gate voltage swing (GVS), while maintaining stable high-frequency behavior with simulated fT and fmax values of 61 and 137 GHz, respectively.
Globally, breast cancer is considered as a major cause of cancer related illness and death among women. Hence, it requires an early and precise diagnosis system to lower the mortality rate. The conventional deep learning models pose privacy and security concerns under healthcare regulations. These issues are addressed by a novel Federated Flash Attention Hierarchical Deep Capsule Network model for robust detection and classification. The proposed model is based on three factors such as Hierarchical Deep Capsule Network, which models spatial and structural relationships within mammographic features, Flash Attention mechanism to improve diagnostic focus, and Federated Learning, which enables multi-institutional collaboration without transferring raw data. The privacy preservation in the proposed framework is achieved through federated learning, where raw data is not shared between clients. The model is evaluated on benchmark mammography datasets using an 80:20 training-testing split and compared with existing machine learning and federated approaches. The experimental analysis is conducted using different measures to demonstrate improved performance compared to baseline models. The proposed Federated Flash Attention Hierarchical Deep Capsule Network attains an accuracy of 98.89%, a precision of 98.78%, a recall of 98.51%, an F1-score of 98.64%, and a Specificity of 98.19%.
