【摘要】 针对无人机在线喷涂绝缘子RTV涂层的问题，提出一种基于计算机视觉和深度学习模型的喷涂质量评价方法。构建语义分割模型提取图像中绝缘子RTV喷涂区域，将提取的区域进行网格划分生成子图像块，再将每个图像块送入神经网络分类模型进行缺陷检测和分类，最后结合模糊评价手段，按照各类子图像块所占的面积比例来生成评定分数，实现喷涂质量的评价。经实验验证，所提方法能准确有效地检测喷涂缺陷，生成的评价结果符合运检标准，满足实际生产需要。更多还原

【Abstract】 An automatic evaluation method based on computer vision and depth learning model was proposed to evaluate the insulator RTV spraying quality. This method first constructs a semantic segmentation model to extract the insulator RTV coating area from the image background. Then, the extracted area is divided into rectangle blocks which will be classified into different defect types through a neural network classification model. Finally, the fuzzy evaluation method is combined to generate a rating score to evaluate the RTV spraying quality according to the area proportion of the defect blocks in the whole image. Experiments show the proposed method is accurate and effective in that the evaluation results are consistent with the operation and inspection standards, which can meet the actual production needs.更多还原