Duman Emre Akın ئاپەکان

“Treelogy” is a mobile application, which can perform leaf-basedtree identification among tree species in Turkey using one pictureof a given leaf. There is only handful of applications interestedin tree identification and they are designed primarily fordetecting North American and European tree species. There is noapplication which has a good performance and localization supportfor identifying tree species native to Turkey. This project isaimed to fill this gap. While we are constructing this application,we worked on supervised learning for classi- fication task. Wefocused on both Deep Learning (specifically Deep ConvolutionalNeural Networks) and Support Vector Machines. Tree identificationprocess uses leaf image features gathered from Caffe, aconvolutional neural network framework, and our image processingmodule. After several experiments, we reached the optimalclassification accuracy of 93.59% for 57 tree species. Experimentsinvolve 16096 training and 3020 testing leaf images. According toour findings, we come to the following conclusion. Certain imageprocessing procedures for extracting features such as shape andtexture descriptors, which we have used in our project, does notproduce features as feasible as convolutional neural networks."Created by group paY inekereG"

Basic Mine Sweeper