Everyone

Duman Emre Akın

Gingerbread

“Treelogy” is a mobile application, which can perform leaf-based tree identification among

tree species in Turkey using one picture of a given leaf.

There is only handful of applications interested in tree identification and they are designed

primarily for detecting North American and European tree species. There is no application

which has a good performance and localization support for identifying tree species native to

Turkey. This project is aimed to fill this gap.

While we are constructing this application, we worked on supervised learning for classi-

fication task. We focused on both Deep Learning (specifically Deep Convolutional Neural

Networks) and Support Vector Machines. Tree identification process uses leaf image features

gathered from Caffe, a convolutional neural network framework, and our image processing

module.

After several experiments, we reached the optimal classification accuracy of 93.59% for 57

tree species. Experiments involve 16096 training and 3020 testing leaf images. According to

our findings, we come to the following conclusion. Certain image processing procedures for

extracting features such as shape and texture descriptors, which we have used in our project,

does not produce features as feasible as convolutional neural networks.

"Created by group paY inekereG"


TreeLogy

Последняя версия TreeLogy