基于改进深度学习的风机油污识别

Identification of Fan Oil Contamination Based on Improved Deep Learning

针对风机设备油液渗漏影响风机正常运行亟需解决的对风机设备油污的识别问题,提出了一种基于改进深度学习的风机油污检测方法;基于深度学习在目标检测中的应用特点,对目标检测网络YOLOv5n(You Only Look Once v5n)进行改进,将原网络中的非极大抑制(NMS,non maximum suppression)替换为Soft-NMS,降低了网络的误检率,添加CA(Coordinate Attention)注意力机制,增强了模型对目标的定位能力,改进原网络损失函数为α-IoU(Alpha-Intersection over Union)损失函数,提高了边界框检测的准确度;实验结果表明:模型平均精度提升了8.1%,查全率提高了19.1%,网络推理速度提高了28.6%;改进后的模型能准确检测风机油污,有效解决了风机实际运行中油液渗漏所带来的问题。

Aiming at the identification problem of fan oil pollution,it needs to be solved urgently when the oil leakage of fan equipment affects its normal operation,an oil pollution detection method of fan equipment based on the improved deep learning is proposed.The method is based on the application characteristics of deep learning in object detection,improves the object detection network YOLOv5n(You Only Look Once v5n),replaces the non-maximum suppression(NMS)in the original network by the Soft-NMS,reduces the false detection rate of the network,adds the coordinate attention(CA)mechanism,and enhances the positioning ability of the model to the target,and improves the original network loss function to the alpha intersection over union(α-IoU)loss function,and improves the accuracy of the bounding box detection.Experimental results show that the average accuracy of the model is improved by 8.1%,the totality rate by 19.1%,and the network inference speed by 28.6%.The improved model can accurately detect the oil pollution of the fan,and it effectively solve the problem of oil leakage in the actual operation of the fan.