Detection of forest fires with image processing

From camera footage of events or fire/flames, it's important for early fire suppression systems. In this way, fires can be intervened early and fires can be extinguished before they grow. State-of-the-art display hardware and machine learning early fire warning systems have attracted an interest in entertainment and have been published in use with the subject. This is basic basic color space based image segmentation applications. In the images given, a different color is stretched and flame/fire zones are determined by color partitioning. The main disadvantage of this type is the adequacy of the parameter. It is staged in network segmentation applications, together with their deep learning. In this study, a new flame/fire segmentation method using deep mesh is proposed. The method in the path is a Seg AttentionNet structure in which the Gate Module (GM) is integrated. GMs can be easily integrated into standard Convolutional Neural Networks (CNN) architecture with minimal computing and increased by model precision and prediction. In use, dice (dice), tversky and focal tversky structural use for performance evaluation of deep network intent. Experimental set. In accordance with a 5-fold cross-validation criterion containing 500 images and according to the obtained performance mean and Jaccard similarity criteria, the mean Dice and Jaccard values calculated as criteria are 0.8755 and 0.7870. These results were compared with the results available in the literature and more successful results could be obtained with successful methods.