Forest fire monitoring via uncrewed aerial vehicle image processing based on a modified machine learning algorithm

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Forest fires pose a significant threat to ecosystems, impacting biodiversity, climate, and geochemical cycles.
• Effective forest fire prevention is crucial for sustainable development and ecological balance.
• Early detection and intervention are key to minimizing forest fire damage.

Purpose of the Study:

• To develop an improved forest fire risk identification algorithm using deep learning.
• To accurately identify forest fire risk in complex natural environments.
• To enhance forest protection strategies through advanced monitoring technology.

Main Methods:

• Image enhancement and morphological preprocessing of forest fire risk images.
• Segmentation of suspected forest fire areas using HAF and MCC methods, followed by feature extraction.
• Development and comparison of classification methods, including an improved backpropagation (BP) neural network and support vector machine (SVM) classifier, using a dataset of 1,450 images.

Main Results:

• Image enhancement techniques improved image clarity and useful information.
• The improved algorithm achieved a high accuracy rate of 92.73% in forest fire risk identification.
• The combined BP neural network and SVM classifier demonstrated effective risk recognition based on feature extraction.

Conclusions:

• The developed algorithm accurately identifies forest fire risk in natural environments.
• The improved deep learning approach contributes significantly to forest protection efforts.
• Advanced image processing and machine learning are vital for effective forest fire management.