A detection method for unbalanced defect samples based on convolutional neural network

Abstract

The invention discloses a method for detecting unbalanced defect samples based on a convolutional neural network, which includes the following steps: (1) collecting early data: collecting two types of data, one is the original defect data of the product, and making a training Set the original image; the second is the defect-free data produced in the production environment, called defect-free data, forming a defect-free image; (2) making the label image of the original defect data; (3) deep neural network training: ( 4) Defect reasoning: For a new image in the actual production environment, multiple non-defective images are randomly selected, and each non-defective image is cascaded with the new image to form a 6-channel input, which is sent to the network to obtain an output, and voted out the last defective pixel. The present invention collects defect data and non-defect data, that is, adds a large number of non-defect samples to form new data input, thereby improving detection performance and being able to deal with some new types of defects.

Description

technical field [0001] The invention relates to the technical field of defect detection and recognition, in particular to a method for detecting unbalanced defect samples based on a convolutional neural network. Background technique [0002] Convolutional Neural Networks (CNN) is a type of Feedforward Neural Networks (Feedforward Neural Networks) that includes convolution calculations and has a deep structure. It is one of the representative algorithms for deep learning. Convolutional neural network has the ability of representation learning, and can perform shift-invariant classification on input information according to its hierarchical structure, so it is also called "Shift-Invariant Artificial Neural Networks". , SIANN). The research on convolutional neural networks began in the 1980s and 1990s. Time-delay networks and LeNet-5 were the earliest convolutional neural networks. After the 21st century, with the development of deep learning theory Proposed and improved numer...

AI Technical Summary

Problems solved by technology

[0004] However, the above method has the following disadvantages: (1) In some actual production environments, the types of defects are very complex and changeable, and some defects are easy to be collected, and the amount of data is also large, while others are very complex. Rarely, the proportion of these small number of defects in the overall data set is relatively small, and it is precisely because of their small proportion that they cannot be trained well by the model, resulting in this type of defect may not be detected ; (2) In actual production, the proportion of good products is usually high, while the proportion of defective products is low. Therefore, the defect data that can be collected is also relatively small. time to collect defect data (such as the method of CN109993094A), but this method is too time-consuming, and it takes too long and costs too much for enterprises

Images