一种基于知识引导CNN的小样本相似磨粒辨识方法,根据磨粒生成机理,以二值图形式标记磨粒高度图的关键特征;以此为基础,构建VGG16模型的U-net网络来自动提取磨粒的典型特征;通过加权方式,将U-Net网络输出与全卷积CNN网络的卷积层融合,引导全卷积CNN网络训练,使其能够快速定位相似磨粒的区别性特征;所构建网络模型采用Focal loss损失和二分类交叉嫡损失的加权和作为整体损失函数﹐以SGD优化算法进行参数训练,获得最终的相似磨粒分类模型,实现典型相似磨粒的辨识;本发明有效地将磨粒知识经验与CNN网络相结合,解决了目前磨粒分析领域相似磨粒样本数量少、识别准确率低的问题。
A knowledge-guided CNN-based method for the identification of small-sample similar abrasive grains was developed. The network model is constructed using the weighted sum of Focal loss and binary cross-first loss as the overall loss function, and the SGD optimization algorithm is used for parameter training to obtain the final similar grain classification model and achieve the identification of typical similar grains, The invention effectively combines the knowledge and experience of abrasive grains with CNN network to solve the problem of low number of similar abrasive grain samples and low recognition accuracy in the field of abrasive grain analysis.