Study on wildfire detection method for transmission corridor based on multi-scale adaptive feature fusion

With the global climate change and the frequent occurrence of forest fires, early detection and realtime monitoring of wildfires are crucial. To address the limitations of traditional fire detection methods in complex scenarios, this paper proposes a new method for fire image recognition using a Convolutional Neural Network with Adaptive Feature Fusion (AFF-CNN). This method introduces a multi-scale feature extraction mechanism and incorporates a feature adaptive weight module into the network structure to enhance the recognition capability of key features such as flames and smoke in different scenarios. Experimental results show that the AFF-CNN model significantly improves the recognition accuracy of fire images compared to traditional CNNs, especially in complex environments with lighting changes and smoke obstruction, demonstrating stronger robustness. The experimental results indicate that the recognition accuracy of AFF-CNN in complex environments is significantly higher than that of traditional CNNs, particularly in scenarios with lighting changes and smoke obstruction, with significant improvements in accuracy, precision, recall, and F1-Score.