In recent times,an important task for comprehending the usefulness of the process of identifying and keeping track of a location's physical qualities in the scene categorization of a remote sensing image has been widely used in many fields. Due to the close association between the scene content in a remote sensing image and the spatial relationship characteristics of CNNs, they have been significantly hampered by the absence of big datasets with accurate annotations. Using such a large dataset for direct training, on the other hand, can result in the network overfitting to noisy input. Therefore, it is necessary to remove noise and enhance the remote sensing image classification of the network. In this work, the remote sensing scene images are enhanced from two datasets, namely the AID and WHU-RS datasets, and the performance is analysed using normalisation of pre-processing. The image is effectively extracted and selected by applying VGG-19 and IACO. Finally, LSTM for effective remote sensing image detection and classification When compared to existing models, the proposed work performed better in remote scene image classification. The experimental results achieved higher recognition accuracy for the AID (96.79%) and WHU-RS (94.79%) of the remote sensing classification.