Parasite classification in copro images with transfer learning and data augmentation

Humans can harbor parasites; hence, it is fundamental an early detection to prevent diseases. Parasites can be observed in microscopic images, and computer vision may be a helpful approach to detect and classify those parasites in digital images. Deep learning models have shown to have a high perfor...

Descripción completa

Detalles Bibliográficos
Autores: Gutiérrez Velázquez, Miguel Ángel, Chacon Murguia, Mario Ignacio, Ramirez Quintana, Juan Alberto, Quintana, Carlos Arzate, Corral Saenz, Alma Delia
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2022
País:México
Institución:UNIVERSIDAD DE GUADALAJARA
Repositorio:ReCIBE. Revista Electrónica de Computación, Informática, Biomédica y Electrónica
Idioma:español
OAI Identifier:oai:ojs.recibe.cucei.udg.mx:article/235
Acceso en línea:http://recibe.cucei.udg.mx/index.php/ReCIBE/article/view/235
Access Level:acceso abierto
Palabra clave:Parasite Classification
Data Augmentation
Transfer Learning
GAN
AlexNet
Clasificación de parásitos
Aumento de datos
Transferencia de aprendizaje
,AlexNet.
Descripción
Sumario:Humans can harbor parasites; hence, it is fundamental an early detection to prevent diseases. Parasites can be observed in microscopic images, and computer vision may be a helpful approach to detect and classify those parasites in digital images. Deep learning models have shown to have a high performance in image classification. Therefore, this article presents various multi-class deep learning classifiers to recognize 8 classes: 7 types of parasites and non-parasite class. The designed classifiers are based on transfer learning from an AlexNet modified architecture. By having a reduce amount of parasite images samples, a data augmentation was done, employing traditional methods and images generation with an adversarial neural network (GAN) designed for this purpose. The classifier with best performance presented a 99.94%, 98.97% and 98.18% accuracy in the for training, validation and testing sets, respectively.