Investigation on Perceptron Learning for Water Region Estimation Using Large-Scale Multispectral Images

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Earth Observation

Background:

• Land cover classification and temporal change analysis are key remote sensing applications.
• Accurate water/non-water region estimation is fundamental but challenging with traditional methods.
• Existing approaches struggle with global adaptability and challenging conditions like shadows and sunglint.

Purpose of the Study:

• To develop an automated water/non-water region estimation formula using perceptron neural networks.
• To introduce the Perceptron-Derived Water Formula (PDWF) for enhanced water mapping.
• To improve the accuracy and global adaptability of water detection in remote sensing.

Main Methods:

• Utilized Landsat-8 imagery for developing the Perceptron-Derived Water Formula (PDWF).
• Employed perceptron neural networks with automatically derived tuning parameters.
• Implemented a sunglint correction for improved water/non-water estimation.
• Compared PDWF against Modified Normalized Difference Water Index (MNDWI), Automatic Water Extraction Index (AWEI), and Deep Convolutional Neural Networks.

Main Results:

• The PDWF demonstrated superior performance in water/non-water region estimation compared to MNDWI, AWEI, and Deep Convolutional Neural Networks.
• PDWF showed consistent accuracy improvements, particularly in challenging scenarios like hill shadows, building shadows, and dark soils.
• The integrated sunglint correction enhanced the reliability of water detection in affected areas.

Conclusions:

• The Perceptron-Derived Water Formula (PDWF) offers a significant advancement in water/non-water region estimation.
• PDWF provides a robust and adaptable solution for global water mapping using remote sensing data.
• The method's effectiveness in challenging conditions highlights its potential for diverse hydrological and environmental monitoring applications.