Application of modern architectures of deep neural networks for solving practical problems
Alexander Mityakov, V. K. Varankin, Yury S. Tatarinov · 2017
When classifying images it is important to select a method for extracting features from an image that can be a rather difficult task. Recently deep learning of neural networks has shown good results in automatically features extraction for further classification. In this article the capability of using a modern convolution neural network GoogLeNet for automatically features extraction and further images classification is evaluated. Classification task is reviewed in context of a problem of determining vacant parking places. The neural network has shown a result comparable with a result of classification on manually selected features but more sustainable for image transformations.