Using GAOT to Establish Interpretation Mode of Vegetation by SPOT Satellite Imagery - A Case Study in Taipei City

• Main Authors: CHEN, KUANG-JEN, 陳匡仁
• Other Authors: CHEN, LI
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/uj7qb9

碩士 === 中華大學 === 土木工程學系 === 104 === The development of remote sensing technology has led many experts and scholars to launch applications of such technology on the monitoring of water quality of a water body, the usage of lands, as well as the interpretation of the earth surface. The technology of remote sensing is applied as remote sensing is conducted by the sensors carried on satellites. Since there are a wide variety of sensors carried on satellites, a satellite corresponding to different requirements ought to be selected. Nevertheless, there are multiple methods of interpreting the usage of lands, but there are pros and cons to each method due to the lack of uniform satellite image scale, which makes selection rather difficult. Hence, this research aims to use artificial intelligence to strengthen the ability to interpret. The paper utilizes SPOT-2 and SPOT-5 satellite images of Taipei City taken in multiple years, including 1993, 2003, and 2014, and 192 identical locations from these years are sampled. Because the main focus is on the interpretation of vegetation, the data directly associated with bodies of water are isolated. Two thirds of the data from each year are treated as the training sample, and the remaining one third of the data are treated as the testing sample. The interpretation of images is based on the multispectral reflectance, and methods of analysis used include Logistic Regression and Genetic Algorithm of Operation Tree, GAOT, with multispectral reflectance of the sensors taken as variables. (The multispectral bands of SPOT-2 consist of green bands, red bands, and near infrared bands, whereas SPOT-5 offers the additional shortwave infrared bands.) The data obtained from the sampled locations are taken as actual values, and the status of land is categorized as either vegetation or non-vegetation. A statistical forecasting model is established for each time period by ways of Logistic Regression and GAOT. An error matrix analysis is performed to compare not only the accuracy of image interpretations but also the differences as well. The result indicates that the forecasting model established via GAOT leads to higher accuracy in its predicted values.