A NEW APPROACH FOR IMAGE CLASSIFICATION: CONVOLUTIONAL NEURAL NETWORK
Emine Cengil, Ahmet Cevahir Çınar · DergiPark (Istanbul University) · 2016
AbstractIn this work, it has been made image classification process using deep learning that often used with high accuracy rate and low working time about solving problems that belong to machine learning issues such as image processing, signal processing, natural language processing and speech recognition in recent years. Different eight images have been selected from CIFAR-100 dataset. This images are bus, tractor, train, dinosaurs, elephant, butterfly, chair and television. 800 and 40 images for each class in total were respectively used for training and test processes. In work, convolutional neural networks which is a specialized structure of deep learning was used. CNN works by removing hidden attributes thanks to its layered structure. Increasing the number of hidden layer makes a positive impact up to a point. After, it adversely affects performance due to reduce finding speed because of the increase of parameter. We used a AlexNet like structure which has 7 layers in our study. It is found many open source software that accepts CNN as input Caffe that is one of them has been preferred because of compatible with languages such as C++, Python, matlab etc. and applicable to most environments. Method which is recommended shows success of CNN about classification. Also, studies that are made using deep learning techniques have been showed in work. If deep learning improves in this way it will continue to be mentioned in the academic and scientific world, so it will meet expectations