Abstract
A considerable portion of the natural inland lakes has been gradually transformed into aquaculture ponds to meet the enormous demand for aquaculture products. The changes in ponds area can be used to measure the impact of human activities on inland lakes. However, aquaculture ponds and inland lakes are often intermingled with each other especially in the areas close to the lake shore, posing great difficulties for the extraction of aquaculture ponds from medium resolution (15-30 m) multispectral imagery, such as Landsat TM, OLI, and Geofen-1 WFV images. This study proposes a contour-based regularity measurement for water segments, which evaluates the zero-curvature portions of the boundaries, to distinguish aquaculture ponds from natural water. Water surfaces are firstly extracted from satellite images, and then boundary trace of each water segment is carried out to evaluate the geometrical feature of its contour, including perimeter, curvature and the proposed contour-based regularity. Eventually, SVM classification based on these geometrical features separates the aquaculture ponds from inland lakes. Experiments on Landsat TM, OLI, and Geofen-1 WFV images showed that the combination of perimeter, area and proposed contour-based regularity outperforms other feature combinations and produced the most accurate classification. Therefore, the proposed method can be used to extract all aquaculture ponds from all historic Landsat images to monitor the changes in inland aquaculture.
NSTL
Elsevier