基于改进SSD的航空发动机目标缺陷检测

Aeroengine Target Defect Detection Based on Improved SSD

针对航空发动机的维修检测存在结构复杂、难度大、目标小等问题,提出了改进的单激发多盒探测器(single short multibox detector, SSD)模型,用于检测航空发动机内部凸台缺陷。首先,介绍了实验选用的数据集以及对数据集的处理。然后,分析了SSD模型的基本原理和检测流程,根据凸台缺陷的特点对SSD模型进行了调整。一是对实验数据集采用聚类分析算法来计算模型默认框大小;二是采用模型更底层的卷积层所输出的特征图来进行凸台缺陷的特征提取。最后,通过MATLAB软件对数据集进行扩充。改进后的SSD模型识别凸台缺陷的准确率从从2%提高到了19.6%,但是对实际应用来讲还有很大的提升空间。

Aiming at the problems of complex structure, high difficulty and small targets in the maintenance and detection of aeroengines, an improved single short multibox detector(SSD) model is proposed in this paper to detect the internal boss defects of aeroengines. Firstly, the data set selected in the experiment and the processing of the data set are introduced. Then, the basic principle and detection process of SSD model are analyzed, and SSD model is adjusted according to the features of boss defects. First, cluster analysis algorithm is used to calculate the default frame size of the model for the experimental data set. Secondly, the feature map output by the convolution layer at the lower level of the model is used to extract the features of boss defects.Finally, the data set is expanded by MATLAB software. The accuracy of the improved SSD model to identify boss defects increases from 2% to 19.6%, but there is still much room for improvement in practical application.