Dehazing Model of Extra-High Voltage Converter Station Based on Two-Stage Attention

Fire detection is an important technology to reliably guarantee the safety of power transmission scenarios. An effective fire detection model can help fire-fighting robots be deployed in a timely and accurate manner. However, under hazy conditions, the fire information in the image will appear blurred, resulting in a decrease in the accuracy of the fire detection model. In response to this problem, this paper designs a hazy fire detection enhancement model based on two-stage attention and fusion loss. First, the visual multi-head self-attention model is used to extract fire features in the first stage, and then through the channel dimension and spatial dimension attention as the reconstruction augmentation module for the second stage. The color loss is computed by color space conversion and combined with structural information restored loss. The dehazing and fire detection experiments are carried out on the fire dataset. The fire image restoration evaluation metrics and fire detection evaluation metrics are calculated. The experimental results verified the promotion and strengthening effect of the proposed dehazing model on the fire detection task.