High-Resolution Satellite Image Classification Using Deep Learning

Over the past decades, artificial intelligence, particularly Machine Learning and Deep Learning, has become increasingly important in remote sensing and satellite imaging. Machine Learning uses statistical data learning to create models for tasks like image classification and anomaly detection. Deep Learning, a subset of Machine Learning, employs deep neural networks to learn complex representations from unstructured data, supporting advanced tasks such as image segmentation and object recognition. This study aims to use various Deep Learning models to segment satellite images and detect different land use patterns in the Cergy-Pontoise agglomeration, utilizing a large number of samples representing diverse land use classes. The study specifically highlights the performance of the RESNET50 model, a well-known deep convolutional neural network architecture renowned for its depth and ability to capture intricate patterns in visual data. Trained using the fast.ai library (a popular deep learning framework) on high-resolution Sentinel satellite images, RESNET50 has shown exceptional capability in segmenting and identifying different land use patterns in the Cergy-Pontoise area. The results demonstrate that RESNET50 is particularly well-suited for this task, achieving high accuracy in distinguishing between complex and subtle differences in land use, thereby providing valuable insights into the urban planning and environmental management of the region.