Research

A digital twin of Aydın Adnan Menderes University’s central campus — modelled in Blender and ray-traced with NVIDIA Sionna — serves as a virtual testbed for the wireless network. Path gain, received signal strength and SINR are analysed across transmitter orientations, showing that the right orientation markedly improves coverage. Among machine-learning methods trained as fast surrogates for the ray tracer, XGBoost performs best, enabling instant predictions without further ray tracing.

Classifies normal, cataract and myopia cases from the ODIR fundus image dataset using state-of-the-art convolutional networks (DenseNet, ResNet, Inception) with contrast-based data augmentation. To cut their heavy computational cost, a lighter, optimised ConvNet with fewer parameters is proposed; it keeps high accuracy while running faster, making it suitable for real-time use and resource-constrained devices.

Compares Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM) and Deep Belief Networks (DBN) for early detection of angular leaf spot, one of the most common diseases in cotton growing; CNN proves the most accurate.