20200311_Water-Quality classification of Inland Lakes Using Landsat8 Images by Convolutional Neural Networks

 

1. Title, Journal and Authors

Title: Water-Quality classification of Inland Lakes Using Landsat8 Images by Convolutional Neural Networks

Journal: Remote sensing

Authors: Fangling Pu*, Chujiang Ding, Zeyi Chao, Yue Yu and Xiu Xu

 

2. Summary

In this paper, water quality prediction was conducted with a machine learning algorithm with water parameters and optical images. Although most of the water parameters have no relations with optical images, they were correlated with optical active parameters such as Chl-a, TSS, and CDOM used as labels for satellite images. CNN and transfer-learning were introduced for the deficiency of data.

 

CNN showed higher prediction accuracy in comparison of SVM and RF. When using transfer-learning had higher accuracy because of decreasing the effects of overfitting. There were some interferences such as sampling intervals and water inflow materials.

 

3. Originality & Creativity

- Almost the water quality estimation model focuses on the estimation of a single parameter, but remote-sensing reflectance may be influenced by changes in other water-quality variables.

>> It would be available to find out how the water quality estimation method was derived.

- The multispectral data acquired by the satellite sensors are influence by atmospheric absorption and scattering, sensor target illumination geometry, and the influences of climate changes.

>> Its preprocessing was done by radiometric calibration and atmospheric correction. And, FLAASH can eliminate the effect caused by the atmosphere and convert spectral radiance to water-surface reflectance.

- CNN model has no requirements that input vectors must be one-dimensions.

 

4. Contact

Dae Seong Jeong / Integrated Ph.D. program

 

Environmental Systems Engineering Lab.

School of Environmental Science & Engineering

Gwangju Institute of Science and Technology

1 Oryong-dong Buk-gu Gwangju, 500-712, Korea

 

Phone : +82-10-2003-7860

E-mail : jeongds92@gist.ac.kr