基于深度学习的无人机巡检图像质量问题识别与实践

无人机在巡检时受到电噪声和振动噪声干扰，产生图像非线性畸变和振动偏移，造成巡检过程中电力仪表指针数值识别失败。为了解决上述问题，研究基于深度学习的无人机巡检图像质量问题识别与实践。引入深度学习训练提取噪声畸变点信息，利用最小二乘法计算噪声畸变点检测数据，分析检测距离，根据无人机自动读数机制的角度数据计算圆弧数值，解析仪表读数，完成深入分析。根据三维直方图像内部的灰度信息得到维轴距概率，按照深度学习模式提取相应的噪点影响参数，通过计算权值参数求得非局部均值。实践结果表明，所提方法能有效提高电力仪表指针图像识别的清晰度，有效滤除噪声，加强图像识别效果。

Identification and practice of common problems for image quality of UAV inspection based on deep learning

During unmanned aerial vehicle(UAV)inspections,non-linear image distortion and vibration offset may occur due to electrical and vibration noise,causing failure in recognizing the power meter pointer during the inspection.This paper studies the identification and practice of UAV inspection image quality based on deep learning to solve the above problems.Deep learning training is introduced to extract the information of noise distortion points.The least squares method calculates noise distortion point detection data and analyzes the detection distance.The arc value is calculated based on the angle data of the UAV automatic reading mechanism,and the instrument reading is analyzed to complete an in-depth analysis.According to the grayscale information of the three-dimensional histogram,the probability of the dimensional axis base is obtained,and the corresponding noise impact parameters are extracted according to the deep learning mode.The non-local mean is obtained by calculating the weight parameters.The practical results show that the proposed method can effectively improve the clarity of the image recognition of the power meter pointer,filter out the noise and enhance the image recognition effect.