Project Deep Learning for Terrian Recognition

Accurate terrain recognition is pivotal for applications in autonomous navigation, environmental monitoring, and geological assessments. This project focuses on developing a deep learning-based system to distinguish between four specific terrain types: Grass, Marshy, Rocky, and Sandy. We assessed several convolutional neural network (CNN) models including MobileNet, a traditional CNN, and DenseNet121. Additionally, a hybrid model combining MobileNet with a support vector machine (SVM) was evaluated to explore potential improvements in classification accuracy. Our comparative analysis concluded with the selection of MobileNet as the final model, driven by its balance of high accuracy and computational efficiency. We utilized a well-curated dataset of terrain images, each labeled according to the terrain type, to train and validate our models. The study's findings underscore the effectiveness of MobileNet in performing real-time terrain classification with substantial accuracy while maintaining minimal computational demand, making it exceptionally suitable for deployment in environments with limited processing capabilities.

Keywords: Terrain Recognition, Deep Learning, MobileNet, Convolutional Neural Networks (CNNs), DenseNet121, Support Vector Machine (SVM), Image Classification