基于YOLOv4算法的低压动力配电箱表面缺陷检测方法研究

YOLOv4-based Surface Defect Detection of Low-voltage Distribution Box

目前常规的低压动力配电箱表面缺陷检测方法主要采用红外传感设备对表面进行扫描,从而实现缺陷检测,但缺乏对缺陷特征的融合处理,导致检测精度较差。对此,提出基于YOLOv4算法的低压动力配电箱表面缺陷检测方法。首先对原始图像进行灰度化处理,并采用高斯滤波算法降低图像中的噪声部分。然后通过构建CNN网络结构,对图像中缺陷的边缘特征进行分离,并通过对特征提取结果进行拟合处理实现多特征融合。最后通过构建YOLOv4网络结构,对激活函数与注意力机制进行设计,实现配电箱表面缺陷的定位与检测。对该方法进行缺陷检测性能的检验,结果表明垂直深度检测误差值较低,检测精度较理想。

The current conventional detection method of surface defects for low-voltage distribution boxes mainly uses infrared sensing equipment to scan the surface to achieve defect detection,which leads to poor detection accuracy due to the lack of fusion processing of defect features.In this regard,a YOLOv4 algorithm-based detection method for low-voltage distribution box surface defects is proposed.First the original image is grayed out and the Gaussian filtering algorithm is used to reduce noise part in the image.Then by constructing a CNN network structure,defect edge features in the image are separated,and the multi-feature fusion is realized by fitting the feature extraction results.Finally,by constructing the YOLOv4 network structure,the activation function and attention mechanism are designed to localize and detect surface defects.The proposed method is verified by performance test to have low vertical depth detection error and higher accuracy.