Research And System Design Of Machine Vision Detection Algorithm For Drone Rotor Pad Printing Defects

As the drone market grows,the demand for drone related accessories rises in tandem,with rotor in particular in great demand as consumables.The drone’s paddle wings are printed with paddle shadow patterns,and various defects may occur when the patterns are transferred.The current method of manual visual inspection of defects has problems such as low inspection efficiency and poor consistency of inspection quality.Although machine vision inspection systems can achieve automated detection,unified detection quality,but the existing machine vision algorithm detection standards are too single,it is difficult to reproduce the intelligent judgment and comprehensive decision-making of manual inspection,that is,according to the different defects customer acceptance to take different detection standards to obtain the best match between product quality control and production cost control.To address the above issues,this paper delves into the integrated application of traditional vision algorithms and deep learning algorithms to design and develop a drone paddle shadow pad printing defect detection and identification system.The main research work of the paper includes the following points.In-depth understanding of the process of paddle shadow pad printing and the causes and types of defects,combined with the production site environment and the requirements of enterprise quality inspection,research and design of a set of machine vision-based pad printing defect detection and identification system solutions,including lighting scheme design,image acquisition hardware selection.The application of grey value and shape matching algorithms in template matching algorithms for defect detection is investigated in depth.Firstly,the two algorithms are investigated separately,and it is experimentally demonstrated that the rate of missing qualified products is lower when the two algorithms are used in combination,while ensuring that there are zero errors in qualified products.Secondly,the matching algorithm is optimised using image pyramids to reduce the algorithm running time.Finally,a comparative analysis was carried out to identify the reasons why some of the qualified products could not be detected by the matching algorithm.The application of deep learning to image classification is investigated in order to solve the problem of partially qualified products that cannot be detected by the template matching algorithm and the classification of defects.A deep learning pre-training model using three network structures,AlexNet,ResNet-50 and MobileNetV2,was used for classification training and the best model was selected.Using this as a means of subsequent optimisation of the template matching algorithm has greatly reduced the rate of missed qualified products,while also enabling defect classification and providing a basis for targeted solutions to defect problems.The detection software was developed using C#in conjunction with Halcon programming,and the integrated application of template matching algorithms and deep learning classification models was implemented and experimentally tested,showing that the software functions and defect detection and identification have met the initial design objectives.In this paper,a drone paddle shadow pad printing defect detection and recognition system is designed and developed by combining traditional vision algorithms with deep learning.The system not only solves the problems of low manual detection efficiency and poor quality consistency,but also has the advantage of intelligent recognition of manual detection.