基于全局注意力机制的变电设备红外图像识别方法

Infrared Image Recognition Method for Substation Equipment Based on Global Attention Mechanism

常规的变电设备红外图像识别特征提取以目标标点提取为主,识别速度慢,容易导致红外图像的缺陷误识率升高。为此,提出对基于全局注意力机制的变电设备红外图像识别方法的设计与验证分析。根据当前识别需求,先采集红外图像数据,通过多尺度的方式,提高识别速度,进行多尺度特征提取。以此为基础,设计全局注意力机制变电设备红外图像识别模型,采用阈值辅助判别的方式来实现图像识别。测试结果表明:在选定的3个阶段中,对比的2种辅助方法对红外图像的缺陷误识率均高达15%以上,而所设计的全局注意力机制变电设备红外图像识别组误识率被较好地控制在10%以下,说明此次在全局注意力机制的辅助下,设计的图像识别方法针对性更强,识别效率高,更为高效。

The conventional infrared image recognition feature extraction of substation equipment mainly focuses on target punctuation extraction,which has a slow recognition speed and can easily lead to an increase in defect misidentification rate of infrared images.In view of this a design and validation analysis of an infrared image recognition method for substation equipment based on global attention mechanism is performed.Based on the current recognition requirements,the method first collects infrared image data,improves recognition speed through multi-scale methods,and performs multi-scale feature extraction.Then a global attention mechanism for infrared image recognition of substation equipment is designed,and threshold assisted discrimination is used to achieve image recognition.The test results show that in the selected three stages,the two auxiliary methods compared have a defect misidentification rate of over 15%for infrared images.However,the misidentification rate of the infrared image recognition group of the substation equipment designed in this paper is well controlled below 10%,indicating that the global attention mechanism facilitates more targeted and efficient recognition of images.