基于改进YOLOv5s的火灾烟雾检测算法研究

The fire smoke detection algorithm based on improved YOLOv5s research

为了解决火灾烟雾检测算法中存在的错检、漏检以及实时性差等问题,提出了一种基于YOLOv5s的火灾烟雾检测模型。首先,使用Ghost Convolution模块代替原YOLOv5s网络结构中的常规卷积模块,在保持相同性能的基础上,降低检测模型的计算成本、减少模型参数;其次,在原YOLOv5s模型骨干网络中加入Vision Transformer结构,减少对卷积神经网络的依赖性,同时提高获取全局和局部特征的能力;最后,引入Coordinate Attention注意力机制,有效地提取特征信息,进一步提高检测的准确率。实验结果表明,所提出的火灾烟雾检测模型参数减少17%,准确率提高0.73%,检测速度提升22.5%,可以满足实际场景下的火灾烟雾检测。

In order to solve the problems of false detection,missed detection and poor real-time performance in fire smoke detection algorithm,a fire smoke detection model based on yolov5s is proposed.Firstly,ghost revolution module is used to replace the conventional convolution module in the original yolov5s network structure to reduce the calculation cost and model parameters of the detection model on the basis of maintaining the same performance;Secondly,the vision transformer structure is added to the backbone network of the original yolov5s model to reduce the dependence on convolutional neural network and improve the ability to obtain global and local features;Finally,the coordinated attention mechanism is introduced to effectively extract feature information and further improve the accuracy of detection.The experimental results show that the parameters of the proposed fire smoke detection model are reduced by 18%,the accuracy is improved by 3.12%,and the detection speed is improved by 20%,which can meet the fire smoke detection in the actual scene.