Title: Scratch detection system of the inner surface of super long gas cylinder based on VGG-16 neural networks

Authors: Baocheng He; Linguang Li; Shijie Ren; Dianwei Qian

Addresses: School of Control and Computer Engineering, North China Electric Power University, Changping District, Beijing, 102206, China ' China ENFI Engineering Corporation, #205, No. 5 Building, 12 Fuxing Avenue, Beijing 100038, China ' Beijing ZhongHangKunWu Technology Co., Ltd., Haidian District, Beijing, 100085, China ' School of Control and Computer Engineering, North China Electric Power University, Changping District, Beijing, 102206, China

Abstract: Aiming at the problems of difficulty and low detection accuracy in manual detection of scratches on the inner surface of ultra-long energy storage gas cylinders in the aerospace industry, a detection system for scratches on the inner surface of ultra-long gas cylinders based on VGG-16 neural network is designed. In this paper, VGG-16 recognition model is innovatively proposed to apply to the application of scratch detection. Compared with the ordinary detection model to detect the required target from the complex background, this article first processes the image acquired by the image acquisition module into a single binary image with a scratched area and a non-scratched area. The VGG-16 recognition model learns the characteristics of the scratches, so as to recognise the scratches under the ordinary background, and achieve the purpose of scratch detection. The results show that the accuracy rate of the scratch detection on the inner surface of gas cylinders reaches 98.5%, which greatly improves the accuracy rate of the scratch detection compared with the previous manual detection methods.

Keywords: scratch test; image processing; neural network; linear array camera.

DOI: 10.1504/IJAMECHS.2022.123140

International Journal of Advanced Mechatronic Systems, 2022 Vol.9 No.4, pp.193 - 202

Received: 12 Jul 2021
Accepted: 16 Nov 2021
Published online: 31 May 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article