Improved Convolutional Neural Networks Model in the Identification and Classification of Monkey Species

Abstract

This study demonstrates the utilisation of the Inception V3 model, a deep Convolutional Neural Network (CNN), for automatically identifying and categorising monkey species through computer vision. Our research seeks to improve monitoring and classification by applying modern machine-learning methodologies in light of the essential need for precise species identification for wildlife conservation. The dataset, consisting of images from 10 distinct monkey species, was enhanced by mathematical transformations, including rotations, flips, and colour jittering to increase model generalisation. The Inception V3 model was trained for 50 epochs, attaining a maximum training accuracy of 100% and a peak validation accuracy of 93.75%. The validation loss showed variability, signifying overfitting and difficulties in generalising to unseen data. A comparative study with other research, including those by Brust et al. (2017) and Freytag et al. (2016), validates our model’s competitive strength while reducing prevalent drawbacks, such as reliance on high-quality datasets and susceptibility to environmental fluctuations. To mitigate these constraints, we recommend for future works into techniques like multi-task learning (MTL), attention processes, and synthetic data generation employing Generative Adversarial Networks (GANs). Our results illustrate the capability of the Inception V3 model for real-time, automated wildlife surveillance, establishing a basis for more effective and scalable conservation strategies. This research enhances the precision and efficacy of species identification in dynamic and harsh habitats, hence aiding in the conservation of endangered species.