Combining YOLO V5 and transfer learning for smoke-based wildfire detection on boreal forests

Abstract

Wildfires present severe threats to various aspects of ecosystems, human settlements, and the environment. Early detection plays a critical role in minimizing the destructive consequences of wildfires. This study introduces an innovative approach for smoke-based wildfire detection in Boreal forests by combining the YOLO V5 algorithm and transfer learning. YOLO V5 is renowned for its real-time performance and accuracy in object detection. Given the scarcity of labelled smoke images specific to wildfire scenes, transfer learning techniques are employed to address this limitation. Initially, the generalisability of smoke as an object is examined by utilising wildfire data collected from diverse environments for fine-tuning and testing purposes in Boreal forest scenarios. Subsequently, Boreal forest fire data is employed for training and fine-tuning to achieve high detection accuracy and explore benchmarks for effective local training data. This approach minimises extensive manual labelling efforts while enhancing the accuracy of smoke-based wildfire detection in Boreal forest environments. Experimental results validate the efficacy of the proposed approach. The combined YOLO V5 and transfer learning framework demonstrates a high detection accuracy, making it a promising solution for automated wildfire detection systems. Implementing this methodology can potentially enhance early detection and response to wildfires in Boreal forest regions, thereby contributing to improved disaster management and mitigation