Fused Feature Vector based DenseNet201 with Deep Convolutional Generative Adversarial Network for Forest Fire Prediction

The rise in temperature on the earth is one of the main sources for the cause of fire in the forest area. The dry vegetation of the forest leads to sudden fire in the forest. Various technologies have been introduced to capture the aerial views of the forest to monitor the ongoing actions. The different existing methods failed to predict the occurrence of the fire in the forest area due to lack of feature dimensionalities handling and noises generated during the training process that reduced the performance of appropriate forest fire prediction. To overcome the existing problems the Fused Feature Vector Based DenseNet201 (FFV-DenseNet201) is proposed with the Deep Convolutional Generative Adversarial Network (DCGAN). The min-max normalization was used in the pre-processing stage to rescale the image pixel ranges. The FFV-DenseNet201 is used to extract the significant features from the images. The DCGAN algorithm was used to predict the fire and non-fire forest images. The developed FFV-DenseNet 201 and DCGAN has obtained better results of $\mathbf{9 8. 3 4 \%}$ accuracy, $\mathbf{9 9. 0 1 \%}$ of precision, $\mathbf{9 8. 2 9 \%}$ recall and $\mathbf{9 8. 6 4 \%}$ f1-score respectively for the FLAME dataset compared to existing YOLOv5 model.