Interactive Training of Advanced Classifiers for Remote Sensing Image Analysis

Incorporating supplemental GIS information and human expert knowledge into digital image processing has been acknowledged as a necessity for improving remote sensing image analysis. However, no commercial image analysis system can currently translateexpert knowledge automatically to a computer-usable format. The objective of this project is to develop a system for seamless and transparent integration of machine learning and a rule-based classifier. We are proposing to use decision tree-basedclassifiers to create rules for hierarchical scene analysis. Images will be modeled in pixel, region and scene levels. Spectral values, ancillary GIS layers and other image measurements like textural features will be used in decision tree learning. Thelearning process will include algorithms for handling missing data as well as methods for rule portability. These decision trees will then be automatically converted into rules. Easy-to-use interfaces will support user relevance feedback and enable rulegeneration and editing with a small amount of training data. The system will be evaluated using quantitative and qualitative criteria on independent training and test data. The final product will be a module that integrates the new decision tree classifierand rule generator, our powerful statistical analysis product S-PLUS, and the image analysis capabilities of ERDAS Imagine. The proposed research will significantly benefit the remote sensing community by providing an easy access to advanced statisticaldata analysis resources of our S-PLUS product from ERDAS Imagine image analysis software. Analysts will be able to train models interactively and see the results of the classification without going through the time consuming process of transferring databetween software packages. We will also extend the application domain of our classification software to biomedical image analysis systems by providing tools for automatically transfering expert knowledge to medical knowledge bases.