Jingyi LI, BAOJI UNIVERSITY OF ARTS AND SCIENCES, China
Baiyangdian wetland is rich in resources and diverse in types of aquatic plants. It is difficult to obtain high-precision wetland classification results using traditional classification methods. The research studied the wetland decision tree classification method which is the combination of texture features and spectral features based on feature optimization. Took the Baiyangdian wetland dyke as the research area, the Sentinel-2 image of ESA was used as the image texture image based on the gray level co-occurrence matrix. Calculating the band index, red band, green band, blue band, and near red band of the features in the image to classify the wetland remote sensing image using random forest algorithm. The research results showed that the method can effectively use the spectral and texture features of the ground objects in remote sensing images, and have a high classification accuracy. The overall classification accuracy could reach 92.549%, and the overall Kappa coefficient is 91.27%. Compared with traditional remote sensing image classification methods based on unsupervised classification algorithms and geospatial resolution, this method effectively extracts the spectral and texture features of different features in wetlands, which improves the accuracy of wetland classification and realizes rapid dynamic monitoring of large areas of wetlands.
Mots clés : Baiyangdian wetland|random forest decision tree|feature selection|classification
A102787MZ