POST-CLASSIFICATION ENHANCEMENT IN THE RESULT OF DEEP LEARNING LAND COVER CLASSIFICATION USING VERY-HIGH RESOLUTION SATELLITE IMAGERY

Land cover is one of the fundamental data utilized in spatial analysis for a range of applications, including climate, environment, natural resources, agriculture, forestry, planning, health, and even social issues. The demand for land cover data is diverse, ranging from global scale, regional scale, and detailed scale. Furthermore, data updates also become a necessity for users. The current trend is the requirement for more accurate and up-to-date land cover data. The growing demand for accurate and up-to-date land cover data is in sync with improvements in satellite imagery data acquisition technology, which now offers improved spatial resolution and more effective satellite imagery data processing. Satellite imagery data processing technology is also growing rapidly with the advance of artificial intelligence for semantic classification or segmentation. This research uses a deep learning approach to classify land cover on Pleiades very high-resolution satellite imagery. A post-classification enhancement is also carried out to improve the consistency and accuracy of the deep learning classification results. The preliminary research results show that post-classification enhancement with the algorithms proposed in this study can increase the accuracy of classification results using deep learning-based approaches by approximately 2%. The original Overall Accuracy and Kappa of the deep learning classification results were 0.84 and 0.79. After the post-classification enhancement process, the Overall Accuracy and Kappa values increased to 0.86 and 0.81.