Automatic fire pixel detection using image processing: a comparative analysis of rule-based and machine learning-based methods

摘要: This paper presents a comparative analysis of state-of-the art image processing-based fire color detection rules and methods in the context geometrical characteristics measurement wildland fires. Two new two using an intelligent combination are presented, their performances compared with counterparts. The benchmark is performed on approximately hundred million pixels seven non-fire extracted from five images under diverse imaging conditions. categorized according to existence smoke; meanwhile, average intensity corresponding image. characterization allows analyze performance each rule by category. It shown that existing literature category dependent, none them able perform equally well all categories. Meanwhile, proposed method based machine learning techniques as features outperforms state-of-the-art performing almost different Thus, this method, promises very interesting developments for future metrologic tools unstructured environments.