Enhancing the Efficiency of Wildfire Detection Using Google’s Teachable Machine Supapit Khwanyoo1, Suabsakul Gururatana1,*

Abstract

The objective of this study is to fine-tune hyperparameters for optimal learning performance from Google’s Teachable Machine in wildfire detection and to apply in the development of a wildfire detection system. The procedure begins with researching data and theories related to wildfires and Google’s Teachable Machine. Then, the data is prepared, categorized into three types: Wildfires, Smoke, No Wildfires. The data is divided into two sets: the first set, used for training, consists of 1,000 images per category and is used to train Google’s Teachable Machine to find the optimal hyperparameters. The second set, used for testing, consists of 290 images per category, which have not been used in training, to measure the accuracy resulting from the hyperparameter adjustments. Using Google’s Teachable Machine, the optimal hyperparameters were identified to achieve high accuracy: Epoch = 60, Batch Size = 16, and Learning Rate = 0.001. These settings resulted in an average accuracy of 97%, indicating that these hyperparameters are appropriate for use. This demonstrates that the technology can accurately detect wildfires, suggesting its potential for developing high-efficiency wildfire detection devices in the future.