基于图像特征融合的粉末床缺陷检测方法

针对单一特征对粉末床缺陷表达不明确导致检测效果不佳的问题，提出了一种基于特征融合的增材制造过程粉末床缺陷视觉检测方法。该算法分别使用SIFT方法、灰度共生矩阵和Hu不变矩提取尺度空间特征、纹理特征和几何特征，借助词袋模型对每张图像构建3组视觉单词直方图，通过串行融合3组视觉单词直方图得到新的特征矩阵，采用特征选择对融合后特征矩阵进行降维，并传入随机森林分类器中进行训练。实验结果表明，不同特征对粉末床不同类型缺陷检测具有不同的贡献，优化特征融合参数后，算法平均准确率达到97.46%，缺陷检测效果明显提升。

Defect detection method of powder bed based on image feature fusion

A visual detection method for powder bed defects in the additive manufacturing process based on feature fusion is proposed to improve the poor detection effect due to unclear expression of single features for powder bed defects. The algorithm firstly extracts the scale-space features, texture features, and geometric features from each powder bed image using methods of SIFT, gray-level co-occurrence matrix and Hu invariant moments; then, these images are translated to the feature matrix by serially fusing the three visual word histograms constructed from the three kinds of features by the bag-of-words model; finally, the feature matrix of each image is reduced by feature selection, and passed to the random forest classifier for model training. Experimental results show that different features have different contributions to the detection of different types of defects; through optimizing the feature fusion parameters, the average accuracy of the proposed algorithm reaches 97.46%, significantly improving the defect detection effect.