Improvement of Land Cover Classification Accuracy by using High Resolution Multi Sensor Fused Image.
| Project/Area Number |
09680513
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Environmental dynamic analysis
|
|---|
| Research Institution | Kanazawa Institute of Technology |
|---|
Principal Investigator |
HOSOMURA Tsukasa Kanazawa Institute of Technology, Department of Engineering, Professor, 工学部, 教授 (80124999)
|
|---|
| Project Period (FY) |
1997 – 1998
|
| Project Status |
Completed (Fiscal Year 1998)
|
| Budget Amount *help |
¥2,800,000 (Direct Cost: ¥2,800,000)
Fiscal Year 1998: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 1997: ¥2,200,000 (Direct Cost: ¥2,200,000)
|
|---|
| Keywords | High resolution satellite image / Image fusion / Land cover classification / Classification category / Classification accuracy / Test site / Classification algorithm / Texture analysis / テクス4p解析 |
|---|
| Research Abstract |
Preparation high resolution satellite image of the simulation image using the aerial photograph is the situation which can not be obtained by postponements and failure of the satellite launch, etc. in the March 1999. In this study, the simulation image was made from aerial photograph, and various examination was carried out on the basis of this.
The clustering was applied for examination fusion image on the classification item, and the examination on the classification item number was carried out. Since it changes by degree of separation of each class, as this result, it became clear that it greatly depended on what kind of land cover is included for the object image, for the classification item number.
Test site data of ground resolution 1m was made in order to verify renewal classification accuracy of the test site data.
By doing the schooling in the training data in the maximum likelihood method used for examination land cover classification of sorting algorithm the best, the improvement in the classification accuracy was attempted by correcting so that the training data may approximate to the normal distribution, and carrying out the classification.
As a result of verification classification of the classification result, the drastic improvement in the classification accuracy could not be recognized. Then, it became clear to be one of the reasons why the classification accuracy lowers mixing pixel classification accuracy.
|
Report
(3 results)
Research Products
(30 results)
