Deep Learning-Based Species Classification Using Joint ResNet and Data Augmentation

Biodiversity conservation in modern society is of paramount importance, given the decline in species globally. This research delves into the realm of species identification, an essential element for effective biodiversity protection. By leveraging advanced image classification techniques, this study aims to automate and refine species classification. Specifically, a novel methodology integrating Residual Network (ResNet), particularly its ResNet50 and ResNet101 versions, with data augmentation techniques is introduced. The method capitalizes on ResNet's depth and robustness while harnessing data augmentation to enhance dataset versatility. This study is conducted on the Stanford Dogs dataset, a comprehensive collection of over 20,580 images spanning 120 dog breeds. The results highlight a significant improvement in classification accuracy, with ResNet models outperforming several benchmark models. Experimental results show that the combination of ResNet and data augmentation yields superior performance in species classification tasks. The findings of this research not only push the boundaries of image-based species identification but also offer valuable insights for real-world biodiversity conservation initiatives, fostering more informed and effective protection strategies.