The Study and Development of Artificial Intelligence (AI) Prototype for Forest Fire Detection (FFD) System using Image Data from Unmanned Aerial Vehicle (UAV)

Abstract

The research project aimed 1) to monitor the specific studied area of its wildfire, using Unmanned Aerial Vehicle (UAV) 2) to study and to develop the Artificial Intelligence (AI) system in purpose of detecting wildfire in surveillance location where may occur repeated fire, using automatic steering ship. There both operations in an actual location and applied theoretical knowledge used, were tested. Those of which done by the results from fire forest images data of actual locations using UAV, were used to build an Artificial Intelligence prototype in detecting forest fire in the actual location. The method of the research was made the specific 655 images of fire figure included within and those of 3,881 images of fire figure not included. The data was divided into 3 groups; the first group was used to train, the second group was used to validation, and the last group was used to test the model. The data then was divided into 2 parts for testing, which are composed of 840 images of train model, 208 images of speed and accuracy test on model, and 208 images of performance test on model. The research was applied 4 measurement tools; Accuracy, Sensitivity, Specificity, and Matthews correlation coefficient. Those of which were specifically tested through direct validation. Aside, to do the test, the research employed 6 deep learning algorithms, which included Restnet18, Restnet34, Alexnet. Vgg11, Densenet121, and Vgg16. Separated images were also used in the research to teach models, and deep models were taught using Densenet121 on the Kaggle system. Ten ready-models for effective teaching were shown in the results. As can be observed, both the quantity of training and the accuracy percentage increased from 44.79 to 68.42, respectively.