Color Models Aware Dynamic Feature Extraction for Forest Fire Detection Using Machine Learning Classifiers

• Published in: Automatic control and computer sciences Vol. 57; no. 6; pp. 627 - 637
• Main Authors: Avudaiammal, R., Rajangam, Vijayarajan, Durai Raji V., Senthil Kumar S.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.12.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Classifiers; Color; Computer Science; Control Structures and Microprogramming; Ecological effects; False alarms; Feature extraction; Forest & brush fires; Forest fire detection; Machine learning; Support vector machines; forest fire; precision; accuracy; KNN; color models; SVM; machine learning
• ISSN: 0146-4116; 1558-108X
• DOI: 10.3103/S0146411623060020

The earth’s ecology is well balanced and protected by forests. On the other hand, forest fires affect forest resources, thus causing both economical and ecological losses. Hence, preserving forest resources from fires is very essential to reduce environmental disasters. Controlling forest fire at an early stage is necessary to control their spread. This requirement enforces the necessity of fast and reliable fire detection algorithms. In this paper, a color models aware dynamic feature extraction for forest fire detection using machine learning classifiers is proposed to achieve early detection of fire and reduced false alarm rate. The proposed algorithm extracts fire detection index, wavelet energy, and gray level co-occurrence matrix features from RGB, L*a*b*, and YC b C r color models respectively to train the machine learning classifiers. The performance of the proposed model is analysed using various machine learning algorithms and the standard classification metrics. The proposed color-aware feature extraction gives precision, recall, F1-score, and accuracy of 99, 95, 94, and 97% respectively for the K-nearest neighbourhood model. The support vector machine model delivers 98, 95, 93, and 96.5% respectively. The accuracy of the proposed model is improved by a minimum of 3%, and a maximum of 11% than other color models. Similarly, the false rate reduction is a minimum of 5% and a maximum of 17% than other models.