A Transfer Learning Applied for Malaria Disease Detection on Blood Smear Images

In this work, we developed a convolutional neural network, which we used to diagnose malaria parasites in images of red blood cells in blood smear samples. A method was developed from the study of the sizes of the images to modify them with the minimum change to their original proportions to process them with a convolutional neural network previously trained using transfer learning techniques. In this paper, we also use data augmentation techniques to improve the efficiency of our predictor. The comparison of our method and the convolutional neural network architecture developed with the other twenty-one classification models reported in the literature is presented. Our proposal achieves an efficiency of 96% in image classification. This result is equal to the best result obtained with models reported in the literature, and it also has the advantage that it is a small convolutional neural network that does not consume much time and resources in its training.