Abstract
In recent years, deep learning (DL) algorithms have earned more attention and popularity in image processing, especially in satellite remote sensing analysis, as they can learn the hierarchical and discriminative feature representations within the data. This research aims to enhance the efficiency of the satellite remote sensing image classification by applying deep learning algorithms. The satellite images from the National Agriculture Imagery Program (NAIP) database are initially collected and fed into the system. Consequently, the collected images are pre-processed to enhance the image quality, further improving the developed system's performance. The image pre-processing module utilizes the Patching/Slicing of HS image and the Image Normalization algorithm for performing tasks such as data cleaning, data interpolation, and data discretization, which aids in minimizing the overfitting challenge of the DL algorithm. Further, feature engineering was done to extract the most important features using the pre-trained Autoencoder model, which reduces the data dimensionality. Finally, train the dense Convolutional Neural Network (CNN) with the extracted features to classify the satellite RS images. The experimental results demonstrate that the developed DL strategy obtained an improved accuracy of 93%, which is greater than the existing cutting-edge models. Also, the proposed algorithm attained 96% specificity, 96% sensitivity, 87% precision, and 90% detection rate. This superior performance of the designed methodology highlights its efficiency in analyzing satellite images.