Overview / Usage

We trained our model on a dataset consisting of approximately 60 images depicting three categories, namely fire, no fire, and start a fire, mostly in the forest or forest-like surroundings. Images labelled with "fire" exhibit visible flames, while those marked "start the fire" depict smoke indicating the onset of a fire. On the other hand, "No Fires" images show the forest environment without any fire present. Augmenting of the dataset During our experimentation, we discovered that our network encountered difficulties when classifying images depicting fire outbreaks. To rectify this issue, we included additional images of this category in the dataset by extracting frames from capturing the onset of a fire. To enhance the network's ability to generalize to new images, we utilized Keras' data augmentation function to perform random transformations such as zooming, panning, cropping, and rotating the images before inputting them into the network. LANDSAT: - Approximately every two weeks, the entire Earth's surface at 30-meter resolution, including multispectral and thermal data. MODIS: - Since 1999, includes daily imagery, 16-day BRDF adjusted surface reflectance, and derived products such as vegetation index and snow cover. High-Resolution Imagery: - The National Agricultural Imaging Program (NAIP) provides aerial imagery of the United States at 1-meter resolution.

Methodology / Approach

Till now I have had to clean the satellite image by modifying and cropping it into a 100x100 size image then all the images have been applied in Conventional Neural Networks and different machine learning algorithms that integrate it with a web application. to see the output whether the image contained fire in it or not.

Technologies Used

Python, Pandas, Scikit-learn, (HTML,CSS JS)