Lightweight factory fire detection method based on improved yolov8

Abstract

The invention provides a lightweight factory fire detection method based on improved yolov8, which effectively solves the defects of limited feature extraction capability and high model complexity of a common YOLOv8 model through an improved attention mechanism-channel priority expansion attention module and a mixed classification detection head. In addition, an N-CIoU loss function is introduced to improve the positioning precision and convergence speed of the model and the detection precision of small target flames. Then, in the Neck part, a lightweight convolution module is applied to reduce parameters while keeping the accuracy of the model. And finally, a knowledge distillation strategy is used in the training stage to enhance the generalization ability of the model and reduce false detection. The method has the advantages that mAP0.5 of 90.3% is achieved on a customized factory fire data set, the mAP0.5 is 4.4% higher than that of YOLOv8, FPS is 9.7% higher, high detection precision is guaranteed, light weight of the model is achieved, and the method is more suitable for being deployed on low-power-consumption equipment and suitable for factory fire detection scenes.

Images