| Summary: | 碩士 === 國立雲林科技大學 === 環境與安全工程系碩士班 === 97 === Ground-based air quality observations were set up traditionally in accordance-populated area. It could only sample at single location with operation and maintenance. Nowadays, remote sensing technology has been widely applied to monitor air environment, this research is intended to develop a portable hyperspectral camera (HyCAM-I) to monitor air Pollutant Standard Index (PSI) remothly and flexibly. With the establishment of air quality prediction model from the hyperspectral data and PSIhr, we can measure a regional air quality promptly by mobile HyCAM-I rathen than the traditional site-specific monitoring data.
For building up the air quality prediction model, this study adopted Support Vector Regression (SVR) and Multivariate Linear Regression (MLR) models to analysis the measured hyperspectral data and air quality index. Three bands of 500nm, 550nm and 600nm were used as the input variables to estimate output of the PSIhr. The learning results have shown that the RMSE (Root Mean Square Error) the samples were 6 and 10 respectively for SVR and MLR, and the MAPE (Mean Absolute Percentage Error) were 8% and 14%. The uncertainty analysis with random cross- validation (CV) indicated that the estimation of the average RMSE of validations were 12 and 10, and the average MAPE of validations were 20% and 17%, respectively. In general, the results have shown that the SVR model have better estimation than that of over-learning MLR. However, the validation results did not prove the whole advantage of SVR model whereas the MLR model has better and more stable validations. Conclusionly, The prediction errors of the MLR model were acceptable but need to be verified by improving hyperspectral device and by expanding sample numbers to enhance the reliability.
|
|---|
