Estimation for a Sample Size of Deep Learning Used in Hyperspectral Data Application
| Autor: | Tomomi Takeda, Shinya Odagawa |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
010504 meteorology & atmospheric sciences
Computer science business.industry Deep learning 0211 other engineering and technologies Hyperspectral imaging Pattern recognition 02 engineering and technology 01 natural sciences Regularization (mathematics) Data modeling Tree (data structure) Sample size determination Artificial intelligence business 021101 geological & geomatics engineering 0105 earth and related environmental sciences |
| Zdroj: | IGARSS |
| DOI: | 10.1109/igarss.2018.8517564 |
| Popis: | This paper describes an estimation for a sample size of deep learning used in tree species classification by using an airborne hyperspectral sensor CASI-3. Sample size is a number of reference data needed for an estimation model building. As a result of parameter search of deep learning, the estimation model using hyperbolic tangent function, 5 hidden layers, 200 node size with non-dropout and regularization was the most accuracy and stable. In the result of estimating a sample size increasing based on the number of bands of the hyperspectral sensor, the applicative estimation model was built by 4 times bands. This study achieved to estimate a sample size of deep learning model for tree classification using hyperspectral data, as well as this study is considered to suggest an estimation method of a sample size of deep learning model. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|