Materials, Methods & Technologies, Volume 17, 2023

Abstract: The classification and identification of wildlife for monitoring and conservation purposes have become increasingly important, particularly as environmental degradation becomes a growing concern. Several studies have employed manual methods that are time-consuming, erroneous, and laborious for classification on wildlife datasets. A fully automated classification system is required to address this issue. The use of deep learning for the automatic identification of wildlife has been suggested. However, studies that compare and validate the use of different models in real-world monitoring scenarios are lacking. In this study, we collected wildlife image datasets from the Animals with Attributes repository, and we evaluated the performance of two mainstream Convolutional Neural Network (CNN) architectures, EfficientNetB0 and VGG16, in multi-species classification and identification. We deployed a multi-species classification model, based on deep-learning techniques, that could effectively recognize 37 distinct species categories. Our analysis showed that the EfficientNetB0 model outperformed the VGG16 model overall. The model was trained on a diverse dataset of 185,111 images and tested on 3,131 images, achieving over 80% accuracy and a top-5 accuracy of more than 90%. The F1-score, precision, and recall values for each species category exceeded 0.90, indicating high accuracy in identifying individual species. The model’s prediction performance was validated through experimentation and through the Gradio and Tkinter interfaces, which showed the model to be highly accurate and reliable in image classification. This model could be used in various wildlife monitoring and conservation applications and could make a significant contribution to the field of computer vision. The high accuracy of the model in identifying individual species categories and its reliability in managing a large number of species categorizations make it a valuable tool for conservationists and researchers who seek to monitor and protect wildlife species.