Melanoma remains one of the most aggressive forms of skin cancer, and its early detection is critical for improving patient survival. Vision Transformers (ViTs) have recently shown strong performance in dermoscopic image analysis; however, their effectiveness often relies on stacking multiple transformer encoder blocks, resulting in large numbers of trainable parameters and increased model complexity. In this study, we propose TRM-ViT, a parameter-efficient recursive Vision Transformer designed for binary melanoma classification. Instead of using multiple independent encoder blocks, TRM-ViT applies a single transformer encoder block recursively with shared weights, enabling effective depth while substantially reducing the number of trainable parameters. Experiments conducted on the HAM10000 dataset demonstrate that TRM-ViT achieves a ROC–AUC of 0.7952, comparable to a standard Vision Transformer (0.7951), while using approximately seven times fewer parameters (2.15 M vs. 14.57 M). Notably, the proposed model maintains high melanoma sensitivity, making it particularly suitable for screening-oriented applications. These results indicate that recursive weight sharing can provide an effective trade-off between diagnostic performance and model compactness, supporting the development of efficient decision-support tools for melanoma screening in resource-constrained environments.

Modern deep convolutional neural networks achieve remarkable performance but require substantial computational resources due to their large parameter counts, limiting their suitability for resource-constrained environments. We propose Tiny Recursive ResNet-50, a parameter-efficient architecture that reduces model complexity through recursive feature refinement with weight sharing across reasoning cycles. The proposed design combines lightweight bottleneck blocks, iterative latent state accumulation, and deep supervision to enhance representation quality without increasing parameter count. Extensive experiments are conducted on melanoma classification using the HAM10000 dataset as the primary training and evaluation benchmark. Results demonstrate that the proposed recursive architecture maintains competitive accuracy while reducing parameters by approximately 49%, confirming its efficiency under constrained settings. To assess robustness under limited data and acquisition variability, we additionally validate on the PH2 dataset (200 images). Due to the small dataset size and class imbalance, evaluation is performed using 5-fold stratified cross-validation, and performance metrics are reported as mean ± standard deviation. This validation confirms that recursive refinement with moderate cycle depth improves stability and generalization in small-data regimes.

Automatic sign language recognition has become an important research area for improving communication between deaf individuals and the hearing community. While vision-based approaches achieve high recognition performance, their dependence on cameras and environmental conditions limits their suitability for portable real-time applications. This paper presents a fully embedded Moroccan Sign Language (MSL) recognition system based on a dual smart glove equipped with ten MPU6050 inertial measurement units (IMUs). A dedicated dataset of approximately 8000 gesture sequences representing 20 common MSL gesture classes was collected from 80 participants. The acquired multivariate inertial signals were preprocessed and used to train a lightweight Long Short-Term Memory (LSTM) network, which was quantized and deployed on a Raspberry Pi Pico microcontroller using TensorFlow Lite Micro. Experimental results achieved an overall recognition accuracy of approximately 98%, with high precision, recall, and F1-score, with an average inference latency of 27.4 ± 1.0 ms on the embedded platform. The proposed platform demonstrates the feasibility of accurate and low-latency MSL recognition on resource-constrained embedded hardware under controlled acquisition conditions, representing an initial proof of concept toward future wearable assistive communication systems.