研究目的
Addressing the problem of the spatial-spectral classification of very high-resolution optical images using a kernel- and region-based approach.
研究成果
The proposed method demonstrated high accuracies on a challenging VHR data set, confirming the relevance of GEOBIA approaches to VHR data interpretation. The integration of region- and kernel-based analysis proved effective, with potential for further generalization to non-white Gaussian processes to account for intra-region correlations.
研究不足
The method assumes white Gaussian processes for pixel intensities inside homogeneous regions, neglecting intra-region correlations. The use of estimated covariance matrices may limit applicability when some regions include a few samples.
1：Experimental Design and Method Selection:
The method integrates region-based or object-based information into a kernel machine, using a Gaussian process model to characterize each segment in a segmentation map and define a region-based admissible kernel. This kernel is combined with a marker-controlled watershed segmentation that incorporates scale adaptivity.
2：Sample Selection and Data Sources:
A multispectral IKONOS image (4-m resolution; 1999 × 1501 pixels; red, green, and near-infrared bands) collected over the Itaipu area (Brazil/Paraguay border) was used for experiments.
3：List of Experimental Equipment and Materials:
IKONOS data.
4：Experimental Procedures and Operational Workflow:
The method involves segmentation using the EMF+ method, defining a region-based kernel, and combining it with an SVM classifier for classification.
5：Data Analysis Methods:
The classification accuracy is evaluated using overall accuracy (OA), average accuracy (AA), producer and user accuracies (PA and UA), and the McNemar's statistics.
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
