Evaluation of Object-Oriented, Pixel-Based Classification, and Neural Network in The Separation of Geological Formations Using Landsat 8 Images and Boolean Logic

The preparation of geological maps based on field data and the application of aerial photographs have always been an error because of the structural diversity of the earth and the difficulty of accessing certain regions. But in recent decades, the use of satellite imagery has but in recent decades, the use of satellite imagery has gone a long way to increasing the accuracy and timeliness of geological mapping. The purpose of the current study is to investigate the applicability of Landsat 8 satellite images and object-oriented pixel classification methods in mapping the geological formations of some part of the Shirkuh Mountain range in Yazd province. This area is part of the scattered mountain range of Central Iran with a dry climate and minimal vegetation. Initially, enhanced operations were performed to identify geological information’s using MNF, PCA and FCC processing. Then, the images were classified using object-oriented algorithms (BAYES, SVM, KNN, Decision Tree and Random Forest), neural network (ARTMAP, RBF, MLP and SOM), and base pixels (Maximum Likelihood, Minimum Distance, Mahalanobis and SAM). Next, the error rate of each method was calculated using Boolean logic and kappa coefficient. The results showed that the maximum probability classification with kappa coefficient of 75% in the base pixel category, Fuzzy ARTMAP classification in neural network method with kappa coefficient of 72% and Bayesian classification in object-oriented method with kappa coefficient of 82% have the best results among other methods. These results show that the methods mentioned in the identification and separation of geological formations are effective. The SAM of pixel-based methods, SOM of neural network methods and RF of object-oriented methods with 49%, 64% and 61%, respectively, showed the lowest accuracy in each category.