Authors: Bandaru Jyothi, M.Radhika
Abstract: This article presents a new approach to automated amphibian and reptile categorization that makes use of deep Convolutional neural networks (CNNs) and transfer learning. By developing a reliable and precise MobileNetV2 model for species identification using deep learning, we tackle the limitations of traditional classification methods while also acknowledging the ecological importance of these two vertebrate groups. Using a transfer learning approach on a massive collection of amphibian and reptile images, we train a pre-trained Convolutional neural network (CNN) to overcome the issue of small dataset size. The model is able to generalize well across several species due to its high extraction efficiency. Additionally, the article delves into the significance of image augmentation techniques for enhancing model performance, particularly in cases when labeled data is scarce. Results are favorable when the proposed method is used to overcome challenges caused by changes in size, posture, and environmental factors. Ecological monitoring, conservation efforts, and biodiversity surveys might benefit from the model's classification accuracy, which we prove by comparing it to a large dataset of amphibians and reptiles. With an accuracy rate of 82%, the proposed MobileNetV2 model cans correctly categories amphibians and reptiles. The growing field of computer vision as it pertains to animal ecology and biology has a scalable and successful approach to automated species identification, which this work adds to it. The results show that deep learning techniques particularly transfer learning, have the potential to address the issues with animal categorization. Additional investigation on the connection between AI and biodiversity protection might result from this.
DOI: http://doi.org/10.5281/zenodo.20285882
