基于改进Yolov5s的无人机火灾图像检测算法

UAV fire image detection algorithm based on improved Yolov5s

摘要:

为了解决现有火灾检测算法模型复杂，实时性差，难以部署在无人机平台的问题，通过改进Yolov5s算法对无人机火灾图像目标检测进行分析研究。利用搭载高清摄像头的无人机设备获取的火灾图像、公开数据集、互联网航拍视频自主建立无人机火灾图像数据集；采用轻量化模型Yolov5s为基础模型，MobileNetV3作为特征提取主干网络，降低模型参数和计算量，解决实时性差和模型部署的问题；模型颈部引入注意力模块CBAM，综合了通道和空间信息，加强网络对高层次语义信息的传递；修改模型检测头部结构，增强小目标检测能力。通过消融试验对比分析各个模块对模型的影响，与常见火灾模型进行对比分析，分析本文算法的优劣。算法在自建数据上的平均精度达到78.2%，模型大小为6.7M，单帧（640×640）图像处理时间为15.2ms。实验结果表明，本文算法模型简单、实时性好，为火灾检测算法部署在无人机平台奠定技术基础。

Abstract:

In order to solve the problem that the existing fire detection algorithm model is complex, the real-time performance is poor, and it is difficult to deploy on the UAV platform, the UAV fire image target detection is analyzed and studied by improving yolov5s algorithm. Use the fire image, public data set and Internet aerial video obtained by the UAV equipment equipped with high-definition camera to independently establish the UAV fire image data set; The lightweight model yolov5s is used as the basic model and mobilenetv3 is used as the feature extraction backbone network to reduce the model parameters and computation, and solve the problems of poor real-time performance and model deployment; The attention module CBAM is introduced into the neck of the model, which integrates channel and spatial information to strengthen the transmission of high-level semantic information; Modify the head structure of the model to enhance the ability of small target detection. Through ablation test, the influence of each module on the model is compared and analyzed with common fire models, and the advantages and disadvantages of this algorithm are analyzed. The Average accuracy of the algorithm on the self built data is 78.2%, the model size is 6.7m, and the single frame is 640 × 640) the image processing time is 15ms. The experimental results show that the algorithm model in this paper is simple and has good real-time performance, which lays a technical foundation for the deployment of fire detection algorithm in UAV platform.