Satellite Image Classification Using Convolutional Neural Networks
Vijay Madaan, Neha Sharma · 2024
This effort searches extensively for cloud, desert, green, and water images by CNN-based satellite photo classification. Kaggle preprocessed its many satellite images using data augmentation techniques including rescaling, zooming, and flipping thereby strengthening the model. With numerous convolutional and pooling layers and thick layers, the CNN model on the test set acquired $\mathbf{8 8. 3 8 \%}$ accuracy. The model’s satellite image recognition was demonstrated utilizing the classification performance across classes of confusion matrix analysis. The fundamental difficulty of this project is precisely arranging satellite images into clouds, deserts, vegetation, and water bodies. Applications in environmental monitoring, land use analysis, disaster management, and others as well are covered here. The effort generates significant data augmentation and CNNs are used to improve classification accuracy. Architectural fine-tuning and model training on a well-augmented dataset help to enhance classification performance to 88.38%. This approach guarantees accurate and efficient classification of satellite images, hence enhancing environmental monitoring and analysis.