研究目的
To introduce a framework that solves regression problems based on high-dimensional and small datasets by combining unsupervised with supervised learning, specifically for estimating soil moisture values from hyperspectral data.
研究成果
The SOM framework outperforms RF and SVR in regression tasks based on high-dimensional data with basic tuning techniques. It offers a model-independent regression without any model biases. Further research is proposed to evaluate its prospects and limitations in the context of high-dimensional remote sensing data.
研究不足
The tuning of the SOM framework is time-consuming and can be extended to all hyperparameters. The framework's performance is expected to increase after applying intensive tuning techniques.
1：Experimental Design and Method Selection:
The framework involves two self-organizing maps (SOM) for regression, combining unsupervised and supervised learning.
2：Sample Selection and Data Sources:
A measured dataset from a field campaign is used, consisting of hyperspectral images and soil moisture values.
3：List of Experimental Equipment and Materials:
Hyperspectral snapshot camera (Cubert UHD285), TDR sensors for soil moisture measurement.
4：Experimental Procedures and Operational Workflow:
The dataset is split into training and test subsets. The SOM framework is trained and compared with Random Forest (RF) and Support Vector Regressor (SVR).
5：Data Analysis Methods:
The performance is evaluated using the coefficient of determination (R2) and root mean square error (RMSE).
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
