Identification of Rodent Species Using Deep Learning

Abstract

In this article, we describe a rodent species identification system using computational tools of the deep learning paradigm. The identified species are 4 different types of rodents and the identification is achieved using artificial intelligence techniques applied to images of these rodents in their natural habitat. These images were captured, using camera systems activatedin automatic mode, hidden in the natural habitat of the species under study, under both daylight and nighttime conditions and labeled by experts. The collected imageset constitutes the data set for supervised training of 1411 images of 4 classes. The identifier developed is a multiclass classifier, based on the deep learning paradigm of the broad topic of machine learning, which allows to build a high performance system. The classifier consists of three stages connected in cascade, being the first stage, a pre-processing stage, then, there is a convolutional neural network (CNN) for feature extraction, implemented with a pre-trained architecture using the method of learning by transfer; specifically, the CNN used is the well-known VGG-16; to this second stage, a support vector machine (SVM) is connected as the next and final stage, which acts as the classification stage. For comparative purposes, the results are contrasted against automatic identification models previously published, the results achieved with our identifier are significantly higher than those achieved in previous research on the subject.