Image Fire Quantification Method Though Flame Imaging Target Detection

With the continuous development of urbanization and the increase of population, fire as a catastrophic event poses a great threat to the safety of life and property. However, traditional fire monitoring methods mainly rely on devices such as sensors and detectors, which are difficult to reflect the fire condition and development trend at the scene and are not conducive to rescue preparation. To improve the accuracy of fire condition analysis, this study proposes an image fire quantification method based on flame imaging target detection. The method firstly uses the YOLOv8 network model with integrated Receptive-Field Attention convolutional operation to pre-locate to the area where the fire occurs and performs feature extraction for each flame frame; then, the correlation coefficients between each flame feature and the pre-set fire potential values are calculated and the flame features with higher correlation are screened out; finally, a quantitative assessment model is developed based on five flame characteristics, including flame area and flame profile perimeter, to quantitatively map the fire potential to the [0, 1] value range. A pilot study was carried out on an outdoor platform and the effectiveness of the proposed method was evaluated using the obtained flame images. The results show that the model mean-square error is maintained within 0.001 throughout the training process; under different testing conditions, there is a high degree of consistency between the predicted values and the true values of the test set, which proves the effectiveness of the proposed fire quantification method. Compared to the original YOLOv8 model, the improved model achieves significant performance enhancement in terms of prediction rate.