Citas bibligráficas
Gil, A., Vásquez, S. (2024). Comparativa de un sistema de visión por computador a un clasificador de desechos controlado por una red neuronal convolucional para la categorizacion de desechos de manera predeterminada en un entorno exterior [Tesis, Universidad Privada Antenor Orrego]. https://hdl.handle.net/20.500.12759/44491
Gil, A., Vásquez, S. Comparativa de un sistema de visión por computador a un clasificador de desechos controlado por una red neuronal convolucional para la categorizacion de desechos de manera predeterminada en un entorno exterior [Tesis]. PE: Universidad Privada Antenor Orrego; 2024. https://hdl.handle.net/20.500.12759/44491
@misc{renati/784495,
title = "Comparativa de un sistema de visión por computador a un clasificador de desechos controlado por una red neuronal convolucional para la categorizacion de desechos de manera predeterminada en un entorno exterior",
author = "Vásquez Cabel, Sigifredo Martín",
publisher = "Universidad Privada Antenor Orrego",
year = "2024"
}
The present study aimed to compare and evaluate four different Convolutional Neural Network (CNN) architectures for the detection of calcifications in recyclingrelated images. A total of 150 internet images and 50 photographs of recyclable waste were collected, obtained from pavilion G of the Antenor Orrego Private University (UPAO), to be used in the training of the different neural network architectures. From this data set, a random separation was carried out into two different groups: one for training and another for testing the neural networks. To carry out the training and evaluation of these networks, the TensorFlow library was used, using TensorBoard to display the precision and loss graphs of each of the evaluated architectures. The results obtained revealed significant differences between the different neural network architectures in terms of their performance in the task of detecting calcifications in images related to recycling. In conclusion, after exhaustive analysis of the results, it was determined that the ""ResNet50"" architecture stood out as the most promising in relation to accuracy and loss during training. This choice is based on the search for a deep neural network with an exceptional ability to represent high-level features in images related to recycling, making it the optimal choice for future research in this field
Este ítem está sujeto a una licencia Creative Commons Licencia Creative Commons