This study presents an integrated approach combining image processing and deep learning techniques for the accurate classification of vegetation in satellite images. The analyzed satellite data includes detailed information on weather, climate, geographic areas, vegetation, and natural phenomena. The applications of this research extend to environmental surveillance and land-use management, disaster management, and ecological research, covering different land cover types, such as cloudy areas, deserts, green regions, and water bodies. Image processing techniques, including single-band, dual-band, and multi-band processing, are systematically evaluated in conjunction with diverse deep learning approaches, including custom learning and convolutional neural networks (CNNs) architecture. The study's results highlight the effectiveness of background noise elimination in image processing, leading to the optimal training of a neural network model. The proposed model, with a minimal number of epochs, achieves an impressive overall accuracy of 86.67% in classifying satellite images into target categories (cloudy, desert, green area, and water). This accuracy ensures the CNN-based model's capability to discern habitation patterns within each land cover context. The robustness and accuracy of the proposed CNN methodology position it as a valuable tool for decision-makers across diverse environmental contexts, facilitating informed choices for sustainable development and disaster mitigation.