Comparative Analysis of Multispectral Satellite Image Segmentation for Land Cover Classification: Novel Naive Bayes vs. Convolutional Neural Networks

This project explores the enhancement of machine learning accuracy in identifying multispectral satellite images by comparing novel convolutional neural networks (CNN) with a newly developed Naive Bayes segmentation method. The study involved multiple tests on both techniques, each using a sample size of 20 images. The accuracy of the multispectral satellite image segmentation was the primary focus. For pre-test analysis, a power value was computed using the G Power software, yielding an 80 % power level. In terms of accuracy, the novel Naive Bayes method demonstrated superior performance, with an accuracy of 91.5050 %, compared to the CNN technique. Additionally, the standard deviation for the Naive Bayes method was 0.05916, indicating consistent performance, with a minimal standard error. The significance of these findings lies in the fact that the novel Naive Bayes approach outperformed the established CNN method, particularly in its ability to accurately segment multispectral satellite images. This improvement is statistically significant, as confirmed by the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{G}$</tex> Power settings (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$a=0.05$</tex>, power <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$=0.1$</tex>), which highlight the difference between the two independent methods. These results suggest that the novel Naive Bayes method could be a more effective tool for satellite image analysis, offering higher accuracy and reliability.