Training Data Sets Construction from Large Data Set for PCB Character Recognition

Abstract

Deep learning has become increasingly popular in both academic and industrial areas nowadays. Various domains including pattern recognition, Computer vision have witnessed the great power of deep neural networks. However, current studies on deep learning mainly focus on quality data sets with balanced class labels, while training on bad and imbalanced data set have been providing great challenges for classification tasks. We propose in this paper a method of data analysis-based data reduction techniques for selecting good and diversity data samples from a large dataset for a deep learning model. Furthermore, data sampling techniques could be applied to decrease the large size of raw data by retrieving its useful knowledge as representatives. Therefore, instead of dealing with large size of raw data, we can use some data reduction techniques to sample data without losing important information. We group PCB characters in classes and train deep learning on the ResNet56 v2 and SENet model in order to improve the classification performance of optical character recognition (OCR) character classifier.

Character Recognition of Components Mounted on Printed Circuit Board Using Deep Learning

“…We have identified several points in the previous study [3,4], and the learning data in the previous study has been subdivided by case, and the test data is as shown in Table 1. -Amount of data from the 1st factory, 2nd factory: 10% of the total amount -Amount of other factories data: 90% of the total amount -Collect at a different period…”

“…Moreover, when labeling separated character images, there was an error of specifying a different label. Figure 3 shows a problem that cannot be properly divided because of the short distance between characters [3][4][5]. All data were automatically labeled using a pattern matching software that was previously used for PCB inspection during the collection process.…”

“…That data must satisfy both at the same time quantity and diversity requirements. Even when a good deep learning model is trained with a large dataset, there are many differences in performance and accuracy according to the quality of the dataset used for training [1][2][3][4]. When applying deep learning, which can yield high accuracy in general classification problems, it can show good results through collecting, preprocessing, and learning on a large amount of data.…”

“…Figure 1 shows an example of a component attached to a PCB, with letters printed on the tops of the components. These components may be visually indistinguishable because each one varies in size, errors that occur when printing and attaching other components, therefore, the inspection can be carried out using images acquired with high-resolution magnifying cameras [3][4][5]. In the past, a pattern matching method of an operator machine was used, which when performing an inspection can save a pattern in advance and compare it with an image.…”

Brief Paper: Augmentation of Hidden Markov Chain for Complex Sequential Data in Context

“…Although the HMMs in the network were designed, they were first trained together in a network and learned the component boundaries simultaneously. With a small training set, this, however, didn't turn out well [7]. So the model has been primed by prior samples, about 10% of the training set selected at random from the training set.…”