An efficient fitness function in genetic algorithm classifier for Landuse recognition on satellite images.

Area of Science:

• Computer Science
• Remote Sensing
• Geographic Information Systems

Background:

• Genetic algorithms (GA) are effective optimization tools for unsupervised image classification.
• Improving classification accuracy often involves developing better fitness functions for GA classifiers.
• Existing indices like DBI and FCMI are commonly used but can be further enhanced.

Purpose of the Study:

• To propose a new index, DBFCMI, by integrating DBI and FCMI for GA-based unsupervised classification.
• To enhance the accuracy and robustness of land-use classification using satellite imagery.
• To evaluate the performance of DBFCMI against established indices.

Main Methods:

• A novel index, DBFCMI, was developed by combining the Davies-Bouldin index (DBI) and Fuzzy C-Means index (FCMI).
• The DBFCMI index was integrated into a genetic algorithm (GA) classifier.
• Performance was evaluated using a SPOT-5 satellite image for land-use classification, comparing DBFCMI with DBI, FCMI, and PASI.

Main Results:

• The proposed DBFCMI index achieved higher overall accuracy compared to DBI, FCMI, and PASI.
• DBFCMI demonstrated superior robustness in unsupervised classification tasks.
• Land-use classification results from the Shihmen reservoir watershed showed the effectiveness of DBFCMI.

Conclusions:

• DBFCMI offers improved accuracy and robustness for unsupervised land-use classification within a GA framework.
• The integration of DBI and FCMI into DBFCMI provides a more effective fitness function for GA classifiers.
• This research contributes a valuable tool for remote sensing image analysis and land-use mapping.