Extreme Learning을 통한 Convolutional Neural Network 분류기의 빠른 학습 및 보행자 검출기로의 적용

Abstract

This thesis presents a fast method to train CNN classifiers through extreme learning and its classification and detection performance is verified with the popular datasets on classification and pedestrian detection. CNN has been one of the best classifiers for images and object recognition. However, the Backpropagation (BP) algorithm, mostly used for training CNN, suffers from slow learning, local minimum, and poor generalization from overfitting. To solve these problems, a new architecture called CNN-ELM has been proposed here. It is based on a local image version of the ELM representation learning. Using MATLAB 2015a, the classification experiments show a comparable or mostly better classification performance compared to the BP trained CNN, with its training speed up to 200 times faster for MNIST, NORB, and CIFAR-10 datasets. The pedestrian detection experiment using INRIA and POSTECH datasets also exhibits much faster training but with similar detection performance than the BP trained CNN.