The aim of this project is to classify satellite images into their respective categories i.e. 'Cloudy', 'Desert', 'Green Area' and 'Water' using Convolutional Neural Networks (CNNs) implemented with TensorFlow.
The dataset used in this project is taken from kaggle: https://www.kaggle.com/datasets/mahmoudreda55/satellite-image-classification
The model used for prediction is Pre-trained resnet101 model. The training accuracy of the model is around 99.87% and the testing accuracy is around 99.50%.
The deep learning model of this project is connected with an application created with Gradio for real time prediction and it is deployed on HuggingFace Spaces.
Live Preview: https://som11-satellite-image-classification.hf.space/
While the model of this project can classify images correctly, but in some cases, the model may misclassify the images, therefore, it is strongly advised not to rely solely on the output of this model.