| Summary: | 碩士 === 國立中央大學 === 水文與海洋科學研究所 === 106 === The purpose of the present study is to analyze UAV-images and apply image processing techniques to identify the regions of algal reefs. In this study, we applied the K-means cluster method for reef classification. We also investigated the accuracy and computational efficiency of different image techniques in reef identification. Color intensity, texture analysis, and histogram equalization were used to filter out the non-reef components. We also adjusted the image’s brightness to improve the recognition rate. Two images were randomly selected for testing the computational efficiency. Three combinations of image analysis methods based on the following three main groups were tested: a) the color intensity only, b) the color intensity and texture of image, and c) the color intensity, texture of image and brightness. Our results show that the reef and wet sand cannot be easily distinguished when only the color intensity is used for classification. The accuracy of the reef identification using group b has significantly improved to 83%, with slightly low accuracy when the brightness distribution is uneven. After the brightness is adjusted, the accuracy of applying group c increases to 89%, which is the best method among the three groups. We also found that images with resolution of 9.5 cm can significantly reduce the computational cost with better accuracy. Finally, we applied the improved image processing technique to compare the change of the algal-reef region along Taoyuan coast between year 2016 and 2017. Our result proved that the improved image processing technique can significantly reduce the labor cost and increased the classification accuracy, as compare to the previous manually annotation of the reef region.
|
|---|
