Effective training method based on PCB noise annotation data

Abstract

The invention discloses an effective training method based on PCB noise annotation data, and the method comprises the following steps: S1, collecting images, and building a PCB image data set containing noise annotation and a PCB image data set after data cleaning; S2, respectively extracting image features of each image in the two data sets through an integrated Effective net model; S3, performing dimension reduction on the extracted image features through a deep convolutional network; S4, training a classifier model by adopting a cosine loss function; and S5, alternately training the two data sets by adopting transfer learning, and carrying out fine adjustment on a relatively high-resolution image in the PCB image data set containing the noise annotations to finally realize data training. The invention is mainly used for solving the problem of realizing image retrieval when the training set contains a large number of noise annotations. Efficient speed and accuracy are realized in theretrieval process.

Description

technical field [0001] The invention relates to the technical field of image retrieval, in particular to an effective training method based on PCB noise label data. Background technique [0002] The PCB board is the basic component of many modern information-based and automated electronic products, and it plays a connecting role in the development of the entire electronics industry. In the past few years, due to political and economic influences, the consumption of major electronic products such as smartphones and personal computers was relatively small, and the growth rate of the electronics industry was slow. Later, through the promotion and popularization of emerging technologies in the global market, the international demand for PCB output is increasing. Large, the market capacity is constantly expanding, and the market is huge. In order to meet the requirements of the times, it is necessary to improve the quality of PCB products. Due to the complexity of the PCB manufa...

AI Technical Summary

Problems solved by technology

In order to meet the requirements of the times, it is necessary to improve the quality of PCB products. Due to the complexity of the PCB manufacturing process, various factors will cause defects in the entire production process.

Before machine vision technology was widely used, the main method of PCB defect detection was manual detection, and the detection process was cumbersome.

As the design of PCB boards becomes more and more complex, the size of components soldered on the surface is getting smaller and more densely distributed, making manual inspection more difficult, time-consuming, and taking up too many human resources, which leads to an increase in production costs. At the same time, manual The detection has great damage to the eyesight of workers, and there are also problems of detection consistency and low accuracy

In order to improve productivity, reduce production costs, and improve detection efficiency and accuracy, an efficient PCB defect detection method based on machine vision technology has been developed. Noise labeling, there is no good algorithm to achieve effective training on these data, so it is of great practical significance to propose an effective training method based on PCB noise labeling data

Images