Design and experimental evaluation of a triple-band infrared flame detection system using elastic net regression

This paper proposes a flame detection system integrating a triple-band infrared sensor array and an Elastic Net regression model, aiming to mitigate issues such as slow response and high false alarm rates associated with conventional flame detection methods. The developed hardware utilizes pyroelectric infrared sensors operating at wavelengths of 3.8 µm, 4.3 µm, and 5.3 µm, specifically chosen to detect flame radiation and discriminate against common interference sources. The Elastic Net algorithm is implemented to effectively select significant features and enhance flame classification robustness. Experimental results conducted under controlled laboratory conditions with representative flame sources (butane and alcohol flames) and typical interference sources (hot air gun, artificial heat sources, incandescent lamps, and infrared lamps) demonstrate that the system achieves a detection accuracy of 98.2%, an AUC of 0.998, and a response time of 4.2 seconds. Although the validation is limited to simplified scenarios, the proposed system outperforms conventional methods such as LASSO regression, BP neural networks, and threshold-based techniques. This research provides a foundation for future development and validation of reliable flame detection technologies applicable to more complex and realistic industrial environments.