Design of Hybrid Deep Learning-Based Environmental Monitoring Using Feature Fusion Techniques on Remote Sensing Satellite Imagery

The natural environment is crucial for human development and survival, as it offers climate, land, water, biological resources, and other essential ecosystem services. Therefore, environmental monitoring is of great importance for effective resource management, environmental protection, and human survival and development. Conventional monitoring techniques are labor-intensive, costly, and time-consuming. Owing to advances in remote sensing sensor capabilities, large-scale dynamic observation can quickly capture wide spectral information at both local and global scales. In recent years, satellite remote sensing has increasingly informed the state of and pressures on biodiversity at various spatial scales. Modern Deep Learning (DL) methods, evolved from classical neural networks, have outperformed conventional machine learning models, achieving significant improvements in environmental monitoring performance. This paper presents a Hybrid Deep Learning-based Environmental Monitoring using Feature Fusion Techniques on Satellite Remote Sensing Images (HDLEM-FFTSRSI) model. Initially, the image preprocessing stage involves several steps, including resizing, normalization, dataset division, and augmentation, to improve image quality. Furthermore, the HDLEM-FFTSRSI model implements a fusion of EfficientNetV2-S, Vision Transformer (ViT), and ConvNeXt architectures for feature extraction. Finally, a hybrid of Long Short-Term Memory (LSTM) and Autoencoder (AE) models (LSTM-AE) is employed for classification. The comparative analysis demonstrates that the HDLEM-FFTSRSI method achieves a superior accuracy of 99.11% compared with other models on the EuroSat dataset.