A Comparative Analysis of Deep Learning Models for Wildlife Prediction

Our surroundings boast abundant wildlife diversity, however, only a handful of species are easily recognizable to most people. This is commonly observed when hikers and trekkers encounter various wildlife species during their adventures but are often unaware of the diverse array of birds, animals, insects, and fish they come across. To tackle this issue, we propose designing an application which employs machine learning techniques such as Convolutional Neural Network (CNN) to identify these species through a simple image input. A dataset consisting of 200 animals, insects and fish species and 125 bird species, totalling to 325 species was trained and tested using various deep learning algorithms such as VGG16, VGG19, ResNet, AlexNet, and DenseNet to determine which algorithm accurately identified the wildlife species with the maximum accuracy. Targeted at hikers and trekkers, the system enhances nature exploration by providing real-time species recognition, enabling users to gain insights into their surroundings. After a detailed research and implementation of several models, it was concluded that ResNet gave the highest accuracy of 98.23% for identifying the wildlife species.