| In the field of industrial electronic products production,the dispensing machine is usually used to glue the parts of the product to prevent the parts from falling off.Dispensing machine in the completion of rapid dispensing at the same time,it will inevitably have a little glue defects.The past practice is the manual dispensing quality inspection,not only low efficiency and easy to produce misjudgment and miss judgment,and rely on a large number of human demand.In this paper,a deep learning-based dispensing defect detection system is designed for dispensing detection in this scenario.The specific work is as follows:(1)In this paper,through the combination of image preprocessing technology,2D image processing is done on the data taken by 3D camera to transform it into grayscale image.Since 3D cameras rely on laser reflection for composition,it is inevitable that the diffuse reflection of the laser will cause the pixel at the specified position to fail to receive the accurate value,which is manifested in noise and bright spots on the gray image.In view of these two phenomena,Gaussian filtering and morphological open operation were used to remove noise and bright spots.(2)As the focus of this paper is on the defects of the glue line,and the picture taken is the whole picture,the glue line is relatively small,so in order to improve the detection efficiency and accuracy,it is necessary to use template matching to carry out accurate image matting on the glue line.Considering that there are transformation relations of displacement,Angle and scale between the template image and the image to be matched,this paper uses the template matching based on the frequency domain to make accurate image matching by using the transformation relations of the image frequency domain.Finally,according to the data definition of the defect category,the classification of the stranded diagram is judged.Because the accuracy and time complexity of defect classification determination based on image processing data can not meet the needs of industrial scenes,this paper proposes to use deep learning method to classify and discriminate the micelle maps.Data enhancement based on Lenet-5,transfer learning based on VGG-16 and model structure improvement based on Lenet-5 are designed.(3)Finally,the dispensing defect detection software system has done a detailed demand analysis and index analysis,and then the overall design of the software module and the detailed design of each module,including image acquisition and preprocessing module,template library module and glue line defect detection module.Finally,the whole software system is realized through programming,and the system basically meets the design requirements.Through experimental verification,the scheme using data enhancement on the basis of the improved Lenet-5 has the best index,and is superior to the data calculation method of image processing in both time complexity and accuracy.Finally,this scheme is determined as the final deep learning model for the classification of dissolving defects in this paper.The experimental results show that the dispensing defect detection system based on deep learning proposed in this paper can complete the dispensing defect detection task with higher accuracy and faster efficiency. |
