Konvoluční neuronové sítě v analýze obrazu

Abstract

The goal of the Diploma thesis is to get to know Convolutional neural networks and using this knowledge to create a traffic signs detector and a classifier. According to the theoretical knowledge of neural networks function and their learning methods, was created an user application with Python language and Caffe framework, which is a guarantee of a quality implementation of the necessary algorithms and a hight speed calculations on GPU.

This theis and the user application can be further formed and developed. The application is universal, therefore it can be used for more cases than a traffic signs detecting. This text also compares a performance and an accuracy of each and individual model network with notes, which help the readers to understand neuron networks and enable them to create their own classifiers.