Identifying and Classifying Mindanao Bird Species using Convolutional Neural Network
Abstract
For many years, classifying and identifying Mindanao Bird species has been limited to surveys, physical visuals, and field guidebooks, all of which are significant factors to account for the accuracy of an observer’s observations comprehensively. However, despite these methods’ effectiveness and traditionalized nature, the process would often be time-consuming and confining. For this reason, the researchers developed a prototypical, deep-learning model in pursuit of classifying and identifying Mindanao Bird Species efficiently and accurately. The researchers developed the study by identifying suitable models for the model, which would then be compared intricately based on the overall performance. The researchers explored several factors contributing to the models’ performance and results. As for the project, the researchers chose AlexNet, ResNet50, and VGG-16 as their basis of exploration. The researchers developed the CNN models accordingly, identifying the appropriate data pre-processing methods, learning rates, and training and validation epochs for the model to accurately identify and classify Mindanao Bird Species. The research has further shown that the ResNet50 was the most consistent model among the three (3) developed models, with an accuracy rate of ≈ 99%-100%. Although AlexNet and VGG-16 have both significantly shown impressive accuracy rates, exceeding ≈ 98% to over 99%, the ResNet50 has remained consistent throughout its repeated training sessions. Nonetheless, the researchers have recommended that future researchers develop the model dynamically with a more diverse dataset.
My Role
- Development of the ResNet model
- Backend development and deployment
Resources
Colab
The model development was carried out using Colab. It leveraged the GPU environment provided by Colab to train the model and save the trained model for reuse. The following link contains the code for the ResNet model.
Leave a comment