Research On Sealing Detection Method Of Gas Storage Tank Based On Machine Vision

Gas storage tank is an indispensable storage tool in people's production and life.Because of the particularity of its application occasion,the requirement of safety qualification of gas storage tank is very important.Sealability is an important method to check whether the gas storage tank is safe and qualified in industrial production.The traditional gas tightness detection is the bubble method of soaking and smearing.It is inefficient and misjudged to detect whether there are bubbles or not.Machine vision is a rapidly developing direction in the field of artificial intelligence,and it is also a hotspot in the research of scholars at home and abroad.It has a strong comprehensive direction of disciplines and a combination of machine learning and in-depth learning.The hotspot direction of machine vision in industry is detection and measurement.Its principle is to convert the product to be tested into image signal through machine vision products,and then transmit it to image processing system to get the digital signal information of the target,and then make corresponding processing.This paper aims to improve the traditional air tightness detection method based on machine vision.In this paper,the popular portable oxygen tank produced by an enterprise is taken as the research and testing object,and its air tightness is tested on the basis of traditional methods and machine vision in new directions.This paper mainly studies the current situation and development trend of airtightness testing methods for sealed gas storage tanks in China,and discusses the advantages and disadvantages of several different testing methods in this paper.The characteristics of the light source system in machine vision,the main parameters affecting the light source system,the experience of selecting the light source and the principles to follow are discussed.How to transform from an object point in space to a point,a pixel plane coordinate system,an image physical coordinate system,a camera coordinate system,a world coordinate system and a coordinate system in a camera image is discussed.Target detection methods include image-based and inter-frame difference detection methods,optical flow-based motion detection methods,and background subtraction-based detection methods.For these methods,experimental conditions are tested respectively.Combined with improved inter-frame difference method,morphological operation and threshold processing,detection is realized on the basis of traditional single-Gaussian model.The contour detection algorithm is discussed.The area of bubbles is calculated by Freman chain code and image moments.The central coordinates of bubbles are calculated by correlation calculation of moments,and the qualitative calculation of bubbles is realized.The necessary mechanical parts are designed,and the experimental platform is built for experimental verification.