Development of CNN algorithm for detection of endemic plants in Thrace Region from UAV images

The aim of this thesis is to develop and present a plant detection model based on You Only Look Once (YOLO) algorithm, which is a convolutional neural networks algorithm targeting a fast, accessible and high accuracy solution for individuals and organisations who want to keep tracking, detection or statistics of some endemic plants in the Thrace region. In this context, 75 visual data at different resolutions were obtained by using 2 different unmanned aerial vehicles in 2 sample areas of selected endemic plants Dianthus ingoldbyi, Verbascum degenii and Centaurea kilaea. The visual data was increased to 240 by applying certain data augmentation techniques. Two different models were created and compared using natural data and augmented data from the YOLO v11 nano model with certain training parameters, and it was experimentally verified that data augmentation methods not only provide class balance but also positively affect the structural learning and detection ability of the model. Among the selected endemic plants, Verbascum degenii endemic plant was found to be the highest performing endemic plant. The best performing model was facilitated to be used by providing an interface.