Herramienta software para reconocimiento de objetos como ayuda a los procesos de mercadeo institucional

Abstract

This thesis presents the development of a deep learning application, which allows to identify places or stations in the Universidad Autónoma de Occidente using the tool of Transfer Learning. In the development of the project, a Dataset was created from scratch with more than 20,000 images from 18 different categories using the Data Augmentation technique. The development of the code was done in Python within the Google Colab platform and the application was created in Android Studio, where the Mobilenet convolutional neuronal network was loaded for the Transfer Learning. Subsequently, the representations of the images processed by the network were obtained, a sorting layer of Multi Layer Perceptron type was trained with a Softmax activation function to produce the new predictions. Finally real-time tests were carried out using the application in an Android operating system device by the different stations of the campus, achieving an accuracy of 96%