A transfer learning-based method for identifying surface image defects of injection molded products

Abstract

The invention discloses a method for identifying surface image defects of injection molded products based on migration learning. The surface defect images of non-injection-molded products are used as the source domain data set, and the surface defect images of injection-molded products are used as the target domain data set. The source domain data set is marked with defect categories, all images are marked with domain information, and a CNN model is established. The two data sets Input the CNN model for training, the CNN model extracts the first feature map of the sample through several convolutional layers, and then outputs the predicted defect classification results through the fully connected layer; establishes a migration learning model, constructs a migration loss function, and converts the source domain according to the migration loss function The data set is transferred to the CNN model as knowledge, and the target CNN model is obtained by optimization and iteration; the injection molded product image with surface defects is collected, and the injection molded product image is input into the target CNN model for testing to obtain the predicted defect classification result. The invention has high identification accuracy and solves the problem of lack of samples for identification of surface defects of injection molded products.

Description

Technical field [0001] The present invention relates to the field of computer vision and industrial automation, and in particular, there is a method of surface image defect recognition method based on migration learning based on migration learning. Background technique [0002] The surface of the injection molded article is affected by many factors such as mold and raw material, and is closely related to processing environment, product cooling time, and post treatment process, which is easy to produce various defects, and severely affects the quality of the product; its characteristics can reflect the quality of injection molded products. The surface defect formation mechanism of injection molded products is complex, and the form is diverse, difficult to quantify, and surface defect recognition has always been a problem. Since the surface defect detection technology based on machine vision is an intuitive, non-contact quality detection method, it is possible to efficient and auto...

AI Technical Summary

Problems solved by technology

[0003] At present, the methods for identifying surface defects of injection molded products are mainly divided into two categories: one is to perform operations such as noise reduction filtering, edge extraction, and feature matching on images through digital image processing technology, and to identify defects according to the feature description of images. The advantage of this method is that it is convenient to calculate and easy to implement. The disadvantage is that using simple features to describe the image will lose the details of the image, making it difficult for the classification accuracy of defects to reach a high standard; the second is to use a large amount of training data and GPU computing power. , use the method of deep learning to recognize the surface image of injection molded products. The advantage of this method is that it has high classification accuracy and many types of defects that can be processed. The disadvantage is that it requires high hardware and relies heavily on a large amount of data for training. set, otherwise it is easy to cause overfitting phenomenon

Images