FOREST WILDFIRE DETECTION IN A MACHINE VISION COURSE EXPERIMENT USING VGG19 AND DEEP LEARNING

Abstract

AI and digital picture processing techniques are combined in the
interdisciplinary field of machine vision. In this work, we design a large-scale experiment to
detect wildfires in forests using a method that naturally combines deep learning, machine
learning, with virtual image processing. Many of the studies in wildfire detection research
are not student-friendly, despite notable advancements in the field. Moreover, attaining very
accurate detection remains a significant challenge. This article addresses two components of
the forest detection challenge: the classification of wildfire photos and the localization of
wildfire regions. We provide two methods: one is an optimized convolutional neural network
(CNN) that takes into account both geographical and temporal data for area identification,
and the other is a unique technique to wildfire photo categorization using Reduce-VGGnet.
According to the experimental findings, the best CNN model, which takes into account both
temporal and spatial data, attains an accuracy of 97.35%; however, the suggested Reduce-
VGGNet model may obtain 91.20%. We provide a new framework for the integration of
education and research. It may accomplish outstanding detection performance, foster the
development of machine vision ability, and act as a comprehensive experiment for the
computer science course, among other things.