基于深度主动学习的路面裂纹检测

Road Crack Detection Based on Deep Active Learning

道路裂纹检测是道路日常管理中的重要工作,基于机器视觉的缺陷检测方法已广泛应用于道路裂纹检测。基于图像处理的方法需要人工提取裂纹特征,导致这类方法的通用性不强和准确性不高。基于深度学习的方法需要大量标注数据训练模型,标注工作费时费力,训练集类别数据不平衡会造成深度学习模型的训练效果差。针对上述问题,笔者提出一种基于基尼指数的深度主动学习方法并用于路面裂纹检测。该方法使用未标注的路面图像通过深度神经网络计算其基尼系数值,用基尼指数值作为主动学习判断依据,从大量未标注数据集中选择信息量丰富的样本进行标注,再用卷积神经网络训练主动学习选择并标注的样本。实验结果表明,该方法在道路裂纹检测任务中的准确率达到99.45%,且能够减少81.08%的数据标注工作。

Road crack detection is an important task in the daily management of roads.Defect detection methods based on machine vision have been widely used in road crack detection.Image processing-based methods require manual extraction of crack features,which leads to poor versatility and accuracy of such methods.The method based on deep learning requires a large amount of labeled data to train the model,which results in poor training effect of the deep learning model.In response to the above problems,the author proposes a deep active learning method based on the Gini index.This method uses the Gini coefficient calculated by the unlabeled sample through the deep neural network as the basis for active learning judgment,uses the active learning method based on the convolutional neural network to select the informative samples in a large number of unlabeled data sets for labeling,and then uses the convolutional neural network.Network training actively learns selected and labeled samples.Experimental results show that the accuracy of this method in the task of road crack detection reaches 99.45%,and it can reduce data labeling work by 81.08%.