Dynamic Evolution Analysis of Desertification Images Based on BP Neural Network.

Area of Science:

• Environmental Science
• Remote Sensing
• Artificial Intelligence

Background:

• Land desertification is a critical global environmental issue requiring effective dynamic monitoring.
• Deep neural networks are increasingly utilized in image recognition research, particularly with advancements in artificial intelligence.

Purpose of the Study:

• To develop and evaluate an improved model for classifying and monitoring land desertification using remote sensing images.
• To compare the performance of a novel hybrid model (GA-PSO-BP) against traditional methods for remote sensing image classification.

Main Methods:

• Construction of a GA-PSO-BP (Genetic Algorithm-Particle Swarm Optimization-Backpropagation) analysis model integrating BP neural network, genetic algorithm, and particle swarm optimization.
• Comparative analysis of classification training accuracies for BP, GA-BP, PSO-BP, and GA-PSO-BP models.
• Application of the selected GA-PSO-BP model for dynamic desertification analysis on remote sensing images of the Horqin area.

Main Results:

• The GA-PSO-BP model demonstrated superior classification accuracy for remote sensing images compared to traditional methods and simpler neural network models.
• Analysis of the Horqin area revealed an increase in desertified land by 1.56 km² from 2010-2015, followed by a decrease of 1.131 km² from 2015-2020.
• The study observed a trend of increasing then decreasing desertified land area between 2010 and 2020, aligning with actual conditions.

Conclusions:

• The GA-PSO-BP model offers a simple, effective, and accurate approach for remote sensing image classification and desertification monitoring.
• The model exhibits good performance portability, making it suitable for dynamic analysis of environmental changes.
• Accurate dynamic monitoring of land desertification is crucial for addressing this significant environmental challenge.