Classification of Desert Areas Using Remote Sensing Images and Hybrid Machine Learning.

In order to control desertification, it must first be discovered and classified, and then sustainable agriculture must be promoted to avoid it. It is identified using topographic features of desert areas that vary over time. Therefore, it is necessary to classify and identify desert areas with high accuracy using satellite remote sensing images (SRSI). In this paper, a hybridization was made between Exception transfer learning method, which was used to extract features, and state of art machine learning LightGBM method, which was used to classify (SRSI) dataset approved by Kaggle website. Several preprocessing was also performed on the dataset, such as image cropping to get the important features, as well as performing the data augmentation process to increase the amount of data and make it in different positions. After making a comparison with traditional and previous methods, such as Naive Bayes and K-Nearest Neighbors (KNN), the experiment results showed that LightGBM outperformed them and achieved a high accuracy of 99% and AUC of 100%. [ABSTRACT FROM AUTHOR]