An Attention Convolutional Neural Network for Forest Fire Smoke Recognition

In order to find forest fire in time and accurately, the identification of forest fire smoke based on computer vision has become an important research direction. In this paper, a convolutional neural network model based on the attention mechanism is designed for forest fire smoke recognition. By focusing on the regions with obvious discrimination in the image, more precise local features are extracted for fire smoke identification with the auxiliary of backbone network. The performance of network on the unbalanced forest fire dataset is improved by optimizing the cross-entropy loss function with weights. The experimental results show that attention convolutional neural network improves the accuracy of the model which reached 89.3% while reducing false positives and false negatives.