Flame detection method and system based on BiHR-Net and YOLOv3-head

The invention discloses a flame detection method and system based on BiHR-Net and YOLOv3-head, and the method comprises the steps: firstly, constructing a flame detection model, enabling a backbone network in the model to obtain the small-resolution features of an input image in a two-time down-sampling manner, and reducing the parameter quantity of the detection model; then, providing a feature extraction and feature fusion method BiHR-Net, carrying out up-sampling and down-sampling on a feature map obtained by carrying out two times of convolution on an input feature map to generate two sub-routes to obtain additional features, and fusing features of three scales at the last of the BiHR-Net to achieve the purpose of improving the detection effect; inputting the features obtained by the BiHR-Net into a YOLO-head network for prediction and regression; and finally, training the flame detection model by using the data set to obtain a detector for flame detection. The flame detection speed reaches 112 frame/s, the flame can be rapidly detected and early warned, and the model has good robustness.