Sample image generation method, sample image generation device, circuit board defect detection method and circuit board defect detection device

Abstract

The invention provides a sample image generation method, a sample image generation device, a circuit board defect detection method and a circuit board defect detection device. The sample image generation method comprises the steps: obtaining an original circuit board sample image set as a sample set; respectively determining the distribution proportion of the circuit board sample images in each circuit board sample image category in the sample set; for each circuit board sample image category of which the distribution ratio is smaller than a first ratio threshold, performing oversampling processing on the circuit board sample images in the circuit board sample image category; and / or, for each circuit board sample image category of which the distribution proportion is greater than a secondproportion threshold, performing undersampling processing on the circuit board sample images in the circuit board sample image category, wherein the first proportion threshold is smaller than the second proportion threshold. By the adoption of the scheme, multiple types of sample images can be balanced. The circuit board defect detection model obtained through training can learn more balanced image features. Therefore, the accuracy of circuit board defect detection is improved.

Description

technical field [0001] The present application relates to the technical field of circuit board detection, in particular, to a sample image generation method and device, and a circuit board defect detection method and device. Background technique [0002] As a carrier for the electrical connection of electronic components, a printed circuit board (PCB) realizes specific circuit functions through multiple components mounted on it and the preset logical circuit connection relationship between these components. Considering that during the production, processing or assembly process of PCB boards, components may be missing or over-installed due to various situations, which will cause the PCB boards to fail to work or run normally. As the PCB board is the carrier of electronic products, these PCB boards that cannot work or run normally can only be detected to ensure the performance of electronic products. [0003] A method for detecting PCB board defects based on deep learning is ...

AI Technical Summary

Problems solved by technology

[0004] Although the data volume of the above sample PCB board images is particularly large, the number of sample PCB board images of a certain type is significantly less than the number of other sample PCB board images, and this phenomenon of sample data imbalance is often unavoidable

The above sample data imbalance phenomenon is necessarily related to data acquisition methods and actual application scenarios, which will cause small category information (probably useful information) to be covered up by large category information at the two levels of sample structure and feature dimension, making Model training is often difficult to learn small category information, resulting in poor accuracy of model training

Images