基于机器视觉深度学习的绝缘子前端定位方法

Insulator Front-end Location Based on Computer Vision Deep Learning

将尚未应用于电力领域的SSD(single shot multibox detector)深度学习算法和现场可编程门阵列(field-programmable gate array,FPGA)硬件加速方法应用于电力巡检无人机,设计了一种基于目标图像识别技术的绝缘子前端定位系统。首先,计算目标物在图像像素坐标系中的位置坐标,并分别计算该位置在X方向、Y方向与图像中心的位移偏差值;然后,根据坐标转换关系,将位移偏差值转化为2个偏转方向上的偏转角值;最后,将偏转角信息作为控制参数反馈给云台,驱动电动机完成摄像机的转动,使拍摄的目标物保持在摄像机视野的中心位置,利用经过图像数据库训练的SSD深度学习算法实现绝缘子定位。检测结果显示,该系统能够准确地对绝缘子进行定位,数据处理速度较快,提高了巡线效率。

The single shot multibox detector(SSD)deep learning algorithm and the field-programmable gate array(FPGA)hardware acceleration method that have not yet been used in power field are applied in electric power inspection UAV and an insulator front-end location system based on target image recognition technology is designed.Firstly,the position coordinates of the target in the pixel coordinate system is calculated and displacement deviation values between the location in the X direction and the Y direction and the image center are worked out respectively.Then according to coordinate transformation relationship,the displacement values are transformed into deflection angle values in two deflection directions.Finally,the information of deflection angles is fed back to the cloud platform as the control parameter to drive the motor complete rotation of the camera and keep the target in the central position of the camera field view.Meanwhile,the SSD deep learning algorithm trained by the image database is used to realize insulator location.The detection result indicates this system could accurately locate the insulate and is fast in data process,which has improved line inspection efficiency.