基于RetinaNet算法的输电线路耐张线夹压接缺陷图像检测方法

Image detection method for compression defects in tension line clamps of transmission lines based on RetinaNet algorithm

对输电线路耐张线夹进行射线检测时,图像特征投影过程忽略了图像清晰度对检测结果的影响,会导致检测结果AP值较低。因此,提出基于RetinaNet算法的输电线路耐张线夹压接缺陷图像检测方法。该方法通过判别耐张线夹压接图像的灰度值等级,计算最优分割阈值并保留检测目标轮廓,补偿该轮廓的倾角差值以提高其清晰度,然后通过二维傅里叶逆变换重构缺陷图像,引入RetinaNet算法融合提取图像特征,解构耐张线夹压接部位,计算缺陷特征比例得出缺陷检测结果。试验结果表明,所提方法得出的缺陷检测结果的AP值较高,检测精度较高,满足了输电线路耐张线夹压接质量检测的需求。

When conducting radiographic testing on tension clamps of transmission lines,the projection process of image features ignores the impact of image clarity on the detection results,which can lead to lower AP values in the detection results.Therefore,a method for detecting compression defects in tension line clamps of transmission lines based on RetinaNet algorithm was proposed.This method determined the grayscale level of the compressed image of the tension wire clamp,calculated the optimal segmentation threshold to retain the detection target contour,compensated for the difference in inclination angle of the contour to improve its clarity,and reconstructed the defect image through two-dimensional inverse Fourier transform.The RetinaNet algorithm was introduced to fuse and extract image features,deconstructed the compressed part of the tension wire clamp,and calculated the proportion of defect features to obtain the defect detection result.The experimental results showed that the defect detection results obtained by applying the proposed method had a high AP value and detection accuracy,which met the practical needs of quality inspection for tension line clamps in transmission lines.