基于自注意力的排水管道缺陷检测方法

A Method for Defect Detection of Drainage Pipes Based on Self-attention

排水管网是城市的生命线。人工检测的方法,费时费力,效率低下。传统计算机视觉算法和机器学习算法仅分析了少量缺陷特征,无法满足排水管道检测准确度要求。针对以上问题,本文提出了一种基于自注意力的排水管道缺陷检测方法。该方法采用自注意力机制代替了卷积神经网络作为特征提取器,通过多层感知机和Softmax函数为缺陷图像预测分类标签,提高了排水管道缺陷检测的准确度。对比经典的卷积神经网络算法,本模型的准确度最高。以上研究证明了自注意力算法在排水管道缺陷图像分类领域的可行性,提高了检测的准确度,为排水管道检测提供了一种新思路。

Drainage pipelines are the lifeblood of cities.Manual detection method is time-consuming,laborious and inefficient.Traditional computer vision algorithm and machine learning algorithm only analyze a few defect features,which can not meet the requirements of accuracy of drainage pipe detection.Aiming at the above problems,a method for defect detection of drainage pipes based on self-attention is proposed.In this method,self-attention mechanism is used instead of convolutional neural network as feature extractor,multi-layer perceptron and Softmax function are used to predict and classify the defect image,and the accuracy of defect detection of drainage pipeline is improved.Compared with the classical convolutional neural network algorithm and the traditional algorithm,this model has the highest accuracy.The above research proves that self-attention algorithm is feasible in the field of drainage pipe defect image classification,improves the accuracy of detection,and provides a new idea for drainage pipe detection.