Publication Date
1993
Description
A neural network was applied to the extraction of various types of grasslands using Landsat Thematic Mapper (TM) data. Training fields contained 12 classes (water, paddy field, farmland, sands and rocks, urban area, coniferous forest, deciduous forest, golf course, Sasatype grassland, Miscanthus type grassland, meadow before culling and meadow after cutting). Classification performance using the neural network was 99.4%, which was 2.4% higher than that obtained using the maximum likelihood method. For all types of grasslands, classification performance was 99.8%. The results of the classification area obtained using the neural network and the maximum likelihood method resembled each other. A neural network can be regarded as a non-linear and non-parametric multivariate analysis, and it is useful for the prediction of grassland production.
Citation
Tsuiki, Mikinori; Takahashi, Shigeo; and Oku, Toshiki, "Application of Neural Networks to the Extraction of Various Types of Grasslands in Japan using Landsat Thematic Mapper Data" (2024). IGC Proceedings (1993-2023). 5.
https://uknowledge.uky.edu/igc/1993/session19/5
Included in
Agricultural Science Commons, Agronomy and Crop Sciences Commons, Plant Biology Commons, Plant Pathology Commons, Soil Science Commons, Weed Science Commons
Application of Neural Networks to the Extraction of Various Types of Grasslands in Japan using Landsat Thematic Mapper Data
A neural network was applied to the extraction of various types of grasslands using Landsat Thematic Mapper (TM) data. Training fields contained 12 classes (water, paddy field, farmland, sands and rocks, urban area, coniferous forest, deciduous forest, golf course, Sasatype grassland, Miscanthus type grassland, meadow before culling and meadow after cutting). Classification performance using the neural network was 99.4%, which was 2.4% higher than that obtained using the maximum likelihood method. For all types of grasslands, classification performance was 99.8%. The results of the classification area obtained using the neural network and the maximum likelihood method resembled each other. A neural network can be regarded as a non-linear and non-parametric multivariate analysis, and it is useful for the prediction of grassland production.
