Battery piece EL defect detection method based on improved SSD algorithm

Abstract

The invention provides a battery piece EL defect detection method based on an improved SSD algorithm, battery piece (EL) defect detection is one of important application directions of machine vision,and traditional visual detection based on manpower is low in precision and long in consumed time. In order to improve the recognition rate of multiple defects of a battery piece, the invention provides a deep learning algorithm based on SSD. An EL defect data set is established, and the data set is composed of four types of defects of 870 1536 * 1536 pixel images. A data set is expanded and improved through rotation and denoising, and the SSD is modified, trained and tested based on the data set. The result shows that the SSD algorithm is simpler, quicker and more accurate in EL defect detection, and shows the superiority of the method in EL defect detection.

Description

technical field [0001] The invention belongs to the field of defect detection and machine vision, and in particular relates to a cell EL defect detection method based on SSD (Single ShotMultiBox Detector) algorithm. Background technique [0002] With the generation of renewable new energy, among them, the use of solar power has the fastest development. Semiconductor devices that directly convert solar energy into electrical energy are called solar cells. At present, solar power generation based on monocrystalline silicon wafers develops rapidly and is the main source of solar cells. The production process of solar cells is relatively complicated, and the inevitable defects in the production and installation process directly affect the conversion efficiency and service life of solar cells. Therefore, defect detection is an indispensable link in the production process of solar cells. At present, the detection method of artificial vision is mainly used in the industry, which ...

AI Technical Summary

Problems solved by technology

Ross Girshick proposed a fast R-CNN target detection method by training the VGG19 network. Although this can also be applied to EL detection, the detection accuracy is too low; Rui Huang et al. improved Faster R-CNN to further improve the Wheel Hub. Detection accuracy, although the accuracy has been improved, but the accuracy of small-scale defects in EL is low; Yue Pang et al. proposed a multi-spectral convolutional neural network to detect the surface of solar cells, although this method can also detect the surface of the battery chip detection, but can not meet the requirements for EL intelligent detection

Therefore, the existing detection technology on EL cannot achieve the combination of accuracy and efficiency.

Images