The precise and reliable interpretation of the objects on land is hampered by the presence of clouds in satellite photography. Therefore, before allowing the satellite images to be used for any additional analysis, automatic cloud detection is an essential pre-processing step. The different densities and thicknesses of clouds make this a difficult operation. A Deep Learning (DL)-based system for automatically segmenting clouds from satellite data is proposed in this research to improve the accuracy of climate models. The system uses a Long Short-Term Memory (LSTM) network that is connected with a systematic workflow that includes feature extraction of Histogram of Oriented Gradients (HOG), model training, and data preprocessing of Contrast Limited Adaptive Histogram Equalization (CLAHE). Raw satellite datasets are first preprocessed and pertinent features are extracted. An LSTM-based model uses these features and trained to identify temporal patterns in order to improve cloud formation prediction. Future cloud topologies are predicted by the trained model using historical data, and users can access these predictions. The model is trained and verified using publicly accessible satellite image datasets and is implemented using PyTorch and its related libraries. A comparison with conventional rule-based segmentation techniques demonstrates that the proposed method captures a variety of cloud structures.