Chaitanya Gupta

Terrain Classification

Terrain Classification
Personal Project
Machine Learning
Computer Vision

Project Overview

A supervised deep-learning system that classifies satellite/terrain images into terrain categories using CNNs and transfer learning.

Developed a terrain classification methodology leveraging convolutional neural networks (CNNs) and transfer learning. The approach fine-tunes pre-trained CNN models (e.g., ResNet, DenseNet) on a domain-specific terrain dataset, using convolution + pooling feature extractors and fully connected classification heads. The project includes data preprocessing, augmentation, training pipelines, and evaluation with confusion matrices and accuracy metrics.

Features

  • CNN-based supervised classifier with transfer learning
  • Data preprocessing and augmentation for satellite/terrain imagery
  • Training pipelines with model checkpointing and evaluation
  • Visualization of predictions and class-wise performance

Key Outcomes

  • Applied transfer learning to speed up convergence and improve accuracy
  • Achieved strong classification performance across the target terrain classes
  • Provided end-to-end training and evaluation scripts for reproducibility

Tech Stack

PythonTensorFlowKerasPyTorch (optional)NumPyOpenCVMatplotlib