Research on Fire Detection of Cotton Picker Based on Improved Algorithm.

Area of Science:

• Agricultural Engineering
• Sensor Technology
• Artificial Intelligence

Background:

• Cotton pickers operate in complex environments where hidden fires pose a significant risk.
• Traditional fire detection methods are inadequate for the dynamic conditions of cotton harvesting.
• The physical and operational characteristics of cotton increase the likelihood of combustion during picking.

Purpose of the Study:

• To design an improved multi-sensor data fusion algorithm for enhanced fire detection in cotton pickers.
• To develop a robust fire detection system utilizing infrared temperature and CO sensors.
• To propose and validate an optimized Backpropagation (BP) neural network model for accurate fire prediction.

Main Methods:

• Development of a cotton picker fire detection system integrating infrared temperature and CO sensors with an upper computer.
• Implementation of a novel BP neural network model optimized using a hybrid Gray Wolf Optimizer and Particle Swarm Optimization (MGWO-PSO) algorithm.
• The MGWO-PSO algorithm incorporates a mutation operator for improved searchability and PSO principles for efficient optimization of the BP network.

Main Results:

• The MGWO-PSO optimized BP neural network achieved a correlation coefficient (R) of 0.96929.
• The system demonstrated a prediction accuracy rate of 96.10% with a low prediction error rate of 3.9%.
• Accurate early warning rate reached 96.07%, with false alarm and omission rates below 5%.

Conclusions:

• The proposed multi-sensor data fusion and optimized BP neural network provide an effective method for real-time cotton picker fire detection.
• The system offers timely warnings, significantly improving safety during field operations.
• This research presents a novel approach to accurately detect fires in cotton pickers, mitigating risks associated with combustion.