Satellite Image Chronology Prediction for Sequence Modeling

Satellite imagery is an important tool in understanding temporal changes in Earth's surface. The chronological sequence of satellite images is very helpful for applications like disaster management, climate monitoring, and urban planning. This paper proposes a novel methodology for the chronology prediction of satellite images using a deep learning model called ConvLSTM. This proposed framework incorporates preprocessing techniques such as resizing, normalization, and grayscale conversion of the satellite imagery in preparation for the temporal sequence model. The model ConvLSTM2D has been designed for efficient capture of spatiotemporal features, and other evaluation metrics, such as SSIM, PSNR, MSE, and RMSE, have been used for robust performance assessment. Other exploration is made by using the U-Net, and Convolutional Autoencoder models to improve prediction accuracy and reconstruction of high-quality images. The experimental results illustrate the efficiency of the ConvLSTM model by showing better performance in predicting high-fidelity sequences of images. This work forms a benchmark to utilize deep learning techniques for predicting satellite image chronology and presents pathways for further research in Earth observation technologies.