Summary: | 博士 === 國立臺灣海洋大學 === 應用地球科學研究所 === 97 === Optical satellite images have the advantages in the studies of earth science which could survey a vast area efficiently. This thesis focuses on the applications of multi-spectral satellite images to the oceanic science survey and also employed 6 published papers to explain the potential and limitation. In the water depth survey, we employed the French optical satellite SPOT images as the basis data. The study had employ the character that the green band of SPOT satellite images could penetrate more through the water column and also developed a set of technique to find out the parameters of sea bottom reflectance and water quality parameters in each pixels to improve the accuracy of shallow water depth detection. Beside the water depth determination, this thesis had also developed a new concept to investigate the shallow water environment. The traditional subpixel mixing concept only represents the plane surface mixing, but when in the shallow water environment the subpixel mixing is not only including the sea bottom surface mixing but also mixing the water quality and water depth information. This study uses the orthogonal subspace projection technique to take apart the information from a vertical mixing pixel to obtain the shallow environment investigation. Beside the exploitation concept of vertical subpixel mixing, the research also employs this new developed method and the local satellite images to investigate the area which can’t be entered such as the foreign country island or some remote districts. Besides the under sea level investigation, this thesis has also developed another method to estimate the oil spill above the sea level. In the virtue of oil spell usually has an unique character in spectrum and the anomaly detection algorithm in which it should able to detect the oil spell distribution. But some marine phenomenon’s interfere usually cause the detection more difficult. This thesis has developed a method which could involve the space information to the calculation and improve the accuracy to estimate the distribution of oil spell. Furthermore, this thesis also developed two algorithms to deal the satellite images analysis on land. The first one is a new algorithm which detected the pure pixel spectrum character. The traditional methods always choose from the images directly and figure out the largest probability pixel as the pure pixel spectrum, but when the pure pixel is not exited in the images and all the images are mixed, actually the traditional one can’t seek out the pure pixel from the images. This study has developed a technique which base on the data images to estimate the largest probability position even if the pure pixel do not exist in the images and the traditional method is able to find out the most possible position of the pure pixel. The purpose of this thesis is to develop the algorithms which is suitable for the marine remote sensing, and these algorithms could help to improve the investigation of information whether above or below the sea level and apply them into the marine geosciences research. The second algorithm we proposed in land area is a shadow remove algorithm. The shadow effects always cause certain error when doing the image classification. This thesis proposed a method to eliminate the shaded effects by a simple project base on subpixel assumption. After the shadow remove procedure the classification result had improved obviously. The last part of this thesis is about hardware design to remote sensing algorithm. This thesis had utilized programmable chip to implement the remote sensing algorithm to the circuit to preprocess the images had recorded on satellite. The process is not only help to remove the information not necessary but also reduce the quantity of data to make the data transmission more convenient.
|