A Comparative Study of Image Classification Models for Terrain Recognition
Krishna Kumar, V Karthika, Mohana T, M. A. · 2024
Terrain recognition is vital for various applications such as autonomous navigation, environmental monitoring, military operations, and urban planning. It enables safe navigation for autonomous vehicles, aids in environmental surveillance, assists in military reconnaissance, and supports urban infrastructure development. Recent years have witnessed the augmentation of machine learning models with online feature extraction techniques and self-supervised learning frameworks. However, a notable challenge lies in managing the complexity inherent in processing extensive datasets or numerous features. In this study, this problem is addressed by employing transfer learning for classifying terrains using images. The effectiveness of three widely used pre-trained convolutional neural network models—ResNet50, VGG16, and InceptionV3—was compared. The findings demonstrate the strong performance of the models, emphasizing the effectiveness of transfer learning in tasks focused on terrain recognition.