Object-based Detection of Impervious Area in Agriculture Land Using High-Resolution Satellite Image

Abstract

Agricultural land is important for the food security of a nation However the total agriculture land area across the world has been decreasing every year due to human activities Developing technology especially in businesses and other industries is resulting in increased construction of buildings which then results in loss of land People build factories houses etc on agricultural areas which then become impervious areas as the water absorption ability of the ground underneath decreases The loss of farmland could further pose a threat to national food production leading to shortages and soil pollution Thus periodical assessments to record the change in the total farmland area need to be carried out Traditional manual digitation is usually conducted to detect impervious areas in agricultural land However this process is laborious in Taiwan a country with a large agricultural land area Thus this study uses an object-based approach that employs high-resolution satellite images to detect the impervious areas A pan-sharpened Pleiades image with 0 5-meter resolution and four spectral bands were utilized The HSV (hue-saturation-value) bands derived from the RGB bands were added as object features to extract the impervious area The spectral feature i e HSV NDVI NDWI the soil extraction algorithms and the shape feature i e size and compactness were deployed to extract the impervious area within the agricultural land An F1-score of 0 70 was obtained from this proposed method Furthermore the transferability test was carried out by testing two different conditions The first condition was tested by slicing one-image subset into three different sizes The second condition was tested by analysing four Pleiades image subsets with various scenes and different acquisition times The result shows that the method is stable enough to process various image scenes

Date of Award	2018 Sept 5
Original language	English
Supervisor	Chi-Kuei Wang (Supervisor)