AI visual detection method and system based on deep learning, and readable storage medium

Abstract

The invention provides an AI visual inspection method and system based on deep learning, and the method comprises the steps: obtaining an appearance surface image of a housing structure through an appearance surface visual inspection device, a first side visual inspection device and a second side visual inspection device of AI visual inspection equipment, obtaining a structural surface image of the shell structure through a structural surface visual detection device of the AI visual equipment; determining an appearance defect detection result of the shell structure according to the appearance surface image and an appearance defect detection model, and determining a structure defect detection result of the shell structure according to the structure surface image and a structure defect detection model; determining a quality detection result of the shell structure according to the appearance defect detection result and the structure defect detection result; wherein the appearance surface image at least comprises appearance surface images of the top surface of the shell structure and a plurality of side surfaces of the shell structure. According to the AI visual inspection method provided by the invention, the omnibearing inspection of the product is realized, and the outgoing quality of the product is improved.

Description

technical field [0001] The present invention relates to the field of visual detection, in particular to an AI visual detection method, device and system based on deep learning. Background technique [0002] Visual inspection refers to measuring and judging by the machine instead of the human eye. In practical application, the surface defect detection of the production shell structure can be realized by visual inspection, and with the development of many years, this technology has become very mature. [0003] In the process of intelligent manufacturing, the image of the objective object is taken by the camera with the aid of the lens and light source, and then the image is collected by the capture card and stored in the computer or server hard disk. The image processing software uses algorithms and rules to cut the image into slices. After dividing and calculating, a certain result is obtained, and the result is used for the detection, measurement and control of the shell str...

AI Technical Summary

Problems solved by technology

[0005] 1) The cost of automation is high, and the rotating structure used for rotating products in existing testing equipment is relatively complicated (such as including multiple rotating shafts for rotation), which correspondingly increases the production cost and operating cost of the testing equipment, and this type of testing There are problems of poor stability and accuracy in the actual operation of the equipment, and if higher stability and accuracy requirements are required for the testing equipment, the automation cost of the equipment will be further increased;

[0006] 2) The beat of the production line is slow. Due to the need to rotate a single product, it takes a long time to shoot and detect a single product, and a single station needs to perform more operations (for example, take multiple photos with a single camera), which is difficult Meet the production line beat requirements. In addition, the invention patent with the patent announcement number of CN113670923A is mainly aimed at the defect detection of the shell structure of the notebook machine. Low, for example, a single detection of a notebook machine takes only ten or twenty seconds, and the corresponding detection equipment takes a long time for a single station. However, the detection of the notebook shell (not the whole machine structure) Therefore, the existing testing equipment often cannot meet the production line rhythm requirements of the notebook shell structure in actual work;

[0007] 3) Poor compatibility. The shell structure produced by raw material manufacturers usually has many types of defects. For some special types of defects, or defects located in specific positions, only the side image or appearance image of the shell structure may not be clearly identified corresponding defects

For example, only acquiring the side image or appearance surface image of the shell may not be able to identify the defects at the corners of the shell (that is, the intersection of two adjacent sides of the shell), so it is necessary to The angle is taken separately, which involves adjusting the rotation angle of the rotating mechanism / module, which will further increase the structural complexity of the rotating mechanism / module and affect the stability of the equipment; that is to say, the detection equipment has a single function and is compatible with Poor performance (poor generalization ability), it can only be tested for the shell structure of the current model. If you need to test shell structures with different materials, types, colors, and characteristics, you will need to test the light source, lens, camera, and acquisition card. , image processing software, controller, communication unit, etc. are replaced or modified, and the mold change time is long;

[0008] 4) The demand for worker participation is relatively high, the efficiency of worker detection is limited, and when workers are prone to fatigue after long-term repetitive work, the false detection rate will also increase

Images