Patch-mounting anomaly detection method fusing CMA-ES algorithm and sequential extreme learning machine

Abstract

The invention discloses a patch-mounting anomaly detection method fusing a CMA-ES algorithm and a sequential extreme learning machine. The patch-mounting anomaly detection method comprises the following steps: 1, obtaining a sample training set from a normal welding PCB database; 2, performing image enhancement on the training set to obtain an enhanced training set; 3, extracting training set features through Haar transformation; 4, constructing a sequential extreme learning machine model through a single hidden layer feedforward neural network; 5, completing initialization training of the model; 6, obtaining an optimal parameter of the sequential extreme learning machine by using a CMA-ES algorithm; 7, designing an anomaly detection algorithm after pasting based on the sequential extreme learning machine model; 8, performing online training of the sequential extreme learning machine model; and 9, detecting whether a patch is abnormal by using the model. The method can effectively detect the abnormity after patch mounting, has good precision and real-time performance, does not need additional auxiliary information, is suitable for abnormity detection after SMT surface mounting, can be widely applied to an SMT production line, and has a wide application prospect.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and specifically relates to a post-pattern abnormality detection method which integrates a CMA-ES algorithm and a sequential extreme learning machine. Background technique [0002] With the development of the semiconductor industry, embedded systems based on PCB circuits have been more and more widely used, and the post-mount abnormality detection method that can ensure the normal operation of the entire embedded system has great research value; the post-mount abnormality detection method The purpose is to detect whether there are missing stickers, flying materials, skewing, etc. before the PCB is completed; [0003] The main idea of ​​abnormal detection after placement is to realize real-time detection of the placement of the circuit board and then to understand its placement situation. The existing visual solution to realize this technology is mainly divided into two modules: feature ve...

AI Technical Summary

Problems solved by technology

[0003] The main idea of ​​abnormal detection after placement is to realize real-time detection of the placement of the circuit board and then to understand its placement situation. The existing visual solution to realize this technology is mainly divided into two modules: feature vector acquisition and classifier design The features applied to abnormal detection after patching mainly include: ①Haar_like feature ②HOG feature. Commonly used classification learning methods include Adaboost ensemble learning and deep convolutional neural network (Convolutional Neural Networks, CNN) These manually designed shallow features are targeted , but also has the disadvantages of limitation and low robustness

Its generalization ability is often weak. Once the model is formed, it cannot be modified and optimized. At the same time, the calculation speed is slow, and as the network progresses layer by layer, some significant features will be lost.

There are obvious disadvantages in the patch anomaly detection system that has high requirements for accuracy and detection speed.

Images