Remote Sensing Image Classification Using the Spectral-Spatial Distance Based on Information Content.

Area of Science:

• Remote Sensing
• Image Classification
• Geospatial Analysis

Background:

• Traditional remote sensing image classification methods often struggle with accuracy.
• Integrating spatial information with spectral data, known as spectral-spatial classification, improves performance.
• Spectral-spatial distance is a common technique for combining spectral and spatial information.

Purpose of the Study:

• To propose a novel spectral-spatial distance measure method to enhance remote sensing image classification.
• To improve the performance of spectral-spatial classification by introducing a new way to measure spatial relationships between pixels.
• To validate the effectiveness of the proposed method in contextual classifiers.

Main Methods:

• Introduced Information Content (IC) to measure the spatial relation between pixels based on shared information.
• Computed IC from a hierarchical tree generated by statistical region merging (SRM) segmentation.
• Applied the novel distance measure to k-nearest neighbors-statistical region merging (k-NN-SRM) and optimum-path forest-statistical region merging (OPF-SRM) classifiers, creating k-NN-SRM-IC and OPF-SRM-IC.

Main Results:

• The proposed spectral-spatial distance measure was implemented in four land cover images.
• The novel contextual classifiers (k-NN-SRM-IC and OPF-SRM-IC) demonstrated superior performance.
• Classification results using the new distance measure outperformed other competitive contextual classifiers.

Conclusions:

• The proposed information content-based spectral-spatial distance measure is effective for improving remote sensing image classification.
• The novel method enhances the accuracy of distance-based contextual classifiers.
• This approach offers a valid strategy for leveraging spatial information in remote sensing classification.