Defect Detection Of The End Face Of Micro-precision Glass Encapsulated Electrical Connectors Based On Multi-scale Feature Fusion SSD

Micro-precision glass encapsulated electrical connector(MPGE electrical connector)is made of glass powder and metal wire sintered by a special process,which is an important component to ensure the stable operation of the system.Due to its complex background and high precision requirements,traditional image processing methods are difficult to accurately locate the end face defects of MPGE electrical connector and cannot achieve the detection standards.The defect detection technology based on deep learning is able to extract defect features autonomously and has the advantages of high accuracy and speed.In this thesis,an algorithm for detecting end face defects of MPGE electrical connector based on multi-scale feature fusion SSD is proposd.The main studies are as follows:(1)Selecting the appropriate equipment according to the hardware requirements for image acquisition of MPGE electrical connector,to complete the image acquisition.According to the national standard,the defects are classified and the Label Img tool is used to complete the defect calibration to prepare for the subsequent defect detection.(2)In order to precisely locate the defects of MPGE electrical connector,different target detection algorithms are studied and an algorithm for detecting defects of MPGE electrical connector connectors based on multi-scale feature fusion SSD is proposd.The SSD algorithm model with fused depth residual structure is constructed.Then,a top-down multiscale feature fusion method is proposed to further optimize the model and introducing a lightweight channel attention module.The feature extraction capability of the algorithm in complex background and weak target features scenarios is improved.The experimental results show that the defect detection accuracy of the algorithm reaches 91.28% in the end-face defect detection.(3)The concentricity detection of MPGE electrical connector is realized.An image segmentation algorithm based on HSV color space is used to remove the background of micro-precision glass-encapsulated electrical connectors.Two different improved random circle detection algorithms are proposed to detect the inner circle and outer circle parameters respectively.The tilt angle of the kovar alloy column is obtained by calculating the offset and length of the kovar alloy column.There are 72 figures,7 tables,and 61 references.