【摘要】 针对现有的石英坩埚气泡检测准确度低和速度慢的问题，提出一种基于YOLOv5n的轻量化气泡检测算法。首先，在主干网络采用C3Ghost模块替换C3模块，然后使用GSConv替换颈部网络的Conv,同时使用VoVGSCSP模块替换颈部网络的C3。实验结果表明，在自建的石英坩埚透明层气泡数据集上，改进后算法mAP为98.55%,计算量为2.7 G,权重大小为2.54 MB,帧率为137 fps,后期方便部署到嵌入式设备。更多还原

【Abstract】 To address the issues of low accuracy and slow speed in existing quartz crucible bubble detection, a lightweight bubble detection algorithm based on YOLOv5n is proposed.Firstly, the C3 module in the backbone network is replaced with the C3Ghost module.Then, GSConv is used to replace the Conv in the neck network, and the VoVGSCSP module is used to replace the C3 in the neck network.Experimental results show that on a self-built dataset of transparent layer bubbles in quartz crucibles, the improved algorithm achieves mAP of 98.55 %,a calculated amount of 2.7 G,a weight size of 2.54 MB,and a frame rate of 137 fps, making it suitable for deployment on embedded devices in the future.更多还原