Automatic image segmentation model for indirect land use change with deep convolutional neural network

• Published in: Spatial information research (Online) Vol. 32; no. 3; pp. 327 - 337
• Main Authors: Vatresia, Arie, Utama, Ferzha, Sugianto, Nanang, Widyastiti, Astri, Rais, Rendra, Ismanto, Rido
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.06.2024; 대한공간정보학회
• Subjects: Database Management; Earth and Environmental Science; Geographical Information Systems/Cartography; Geography; Remote Sensing/Photogrammetry; 공학일반; Bengkulu flood; Deep convolutional neural network; Indirect land use change; Image segmnetation; Landsat
• ISSN: 2366-3286; 2366-3294
• DOI: 10.1007/s41324-023-00560-y

Inherent limitations of Landsat images restrict the accuracy of land categorization efforts for a better comprehension involved in transforming raw image data into relevant land cover information. Managing mixed pixels and complex spectrum responses, the introduction of advanced algorithmic approaches is essential. Hybrid classification methods that include spectral, spatial, and contextual data can increase the precision of assigning class labels to pixels with confusing features. This paper demonstrate the construction of automatic images segmentation based on deep convolution neural networks with object-oriented integration. Harnessing machine learning approaches in remote sensing images, an exper imental phase showed the best fit model that can be implemented further into larger areas with a Kappa validation value of 99.466% and errors of 0.015 on average. The model used to classify land use to see land degrada tion in Air Bengkulu watershed, the main source of annual flood disaster in Bengkulu area. We found that the area has been reduced by 20%, 1.9%, 48%, and 7.9% for forest, bare land, plantations, and rice fields, respectively, from its initial area of year 2000. Furthermore, increase value of palm oil plantations (7.6%) over the area showed indirectly reason for replacement of agriculture allocation with correlation value of 97.12%.