@mastersthesis{development-gsi-mastersthesis-20186,
author = "S{\'a}nchez L{\'o}pez, Alberto ",
abstract = "A few years ago, the common practice of artificial intelligence was not sustainable beyond research laboratories and science fiction. Currently, there is widespread demand for advanced
systems capable of simulating human behavior intelligence. Machine learning has been the
factor that has fostered change, covering the need that allows learning and generalizing to
machines by experience.
The evolution of artificial intelligence may go through computer vision, a discipline where
deep learning provides remarkable improvements resembling the human nervous system
with artificial neurons. Among the most important machine vision applications include
object detection, object recognition, scene reconstruction, restoration and image motion
estimation.
The main goal of this project is developing an image classification system based on
transfer learning with convolutional neural networks, focused on object recognition. Are
employed several pre-trained architectures accessible from Keras, a high-level API developed
on TensorFlow highlighted for deep learning research. System evaluation is carried out by
a popular dataset of labeled food (ETHZ Food-101) which measures the quality of each of
the previous architectures.
The practical case of use carried out with this system is the prediction of the class and
labeling of thousands of food images of restaurants mined from TripAdvisor with Selenium.
From this new set of data, a restaurant search engine based on food images is implemented,
that means, the search of restaurants is done by food photographs appearance. Everything
mentioned above is developed in a web application to show all the objectives achieved.",
address = "ETSI Telecomunicaci{\'o}n",
institution = "Universidad Polit{\'e}cnica de Madrid",
keywords = "food;image recognition;deep learning;convolutional neural networks",
month = "June",
title = "{D}evelopment of a {F}ood {I}mage {C}lassification {S}ystem based on {T}ransfer {L}earning with {C}onvolutional {N}eural {N}etworks",
type = "TFM",
year = "2019",
}