基于超分辨率改进Faster R-CNN的点阵结构内部缺陷判识方法

利用增材制造技术制备金属三维点阵结构件过程中，结构内部经常会出黏连、断裂等多种细小缺陷，导致样件结构功能下降。根据缺陷结构与正常结构之间的特征区别，提出了一种针对金属点阵结构内部出现的细小缺陷自动判识的方法。利用X-射线微聚焦CT扫描金属点阵结构获得原始输入图片，在Faster R-CNN(Faster region-based convolutional neural networks)框架的基础上，改进原有特征提取网络，开发图像超分辨率重建模块。通过对工业CT图片的局部细节特征增强，实现了快速有效地识别细小缺陷的类型，以及缺陷位置信息的标注。试验证明，改进Faster R-CNN模型对金属点阵结构样件内部的两种典型细小缺陷识别的平均正确率高达93.5%。研究结果表明，通过超分辨率网络对图像进行放大，可以提高细小缺陷的特征提取，通过加深网络加强特征学习，从而实现了点阵结构内部细小缺陷的自动判识。

A Method of Improved Faster R-CNN for Detecting Internal Defect of Metal Lattice Structure Based on Super-resolution Reconstruction

In the process of fabricating metal 3D lattice structure with additive manufacturing technology, many small defects such as adhesion and fracture often occur in the structure, which leads to the structural function decline of the sample. According to the characteristic difference between defective structure and normal structure, an automatic identification method for small defects in metal lattice structure is proposed. The original input images are obtained by scanning metal lattice structures with X-ray microfocusing CT, and the original feature extraction network is improved based on the Faster R-CNN(Faster region-based convolutional neural networks) framework. The image super-resolution reconstruction module is developed. By enhancing the local detail features of industrial CT images, it can quickly and effectively identify the types of small defects and mark the location information of defects. Experimental results show that the improved Faster R-CNN model has an average correct rate of 93.5% for identifying two typical small defects in metal lattice structure samples. The results show that the feature extraction of small defects can be improved by using super-resolution network to enlarge the image, and the feature learning can be enhanced by deepening the network, thus realizing the automatic identification of small defects in the lattice structure.