Research On Online Detection And Recognition Method Of Bright Parts Based On Multi-scale Features

Parts are the essential components in the process of industrial production and manufacturing.Whether in the aerospace equipment manufacturing or precision instrument manufacturing,parts play an important role that cannot be ignored.Machine vision detection and recognition technology are used on various automatic equipment to accomplish the intelligent recognition and grasping of parts,which has a great role in promoting the intelligent development of industrial production and the manufacturing field.Object detection and recognition based on deep learning is easy to extract object features.In different application scenarios,the method of image augmentation is used to expand the training image dataset to enhance the generalization ability of the model,and fusion of multi-scale feature network to improve the real-time accuracy of detection.In this dissertation,different kinds of bright metal parts are taken as research objects to carry out in-depth research,and a method of online detection and recognition of bright parts based on machine vision is proposed.The main work of this thesis is as follows:First of all,according to the characteristics of the research object in this thesis,the machine vision online detection and recognition system is designed,the key hardware equipment of this system is selected,and the machine vision detection platform is built.With this system,the image dataset of bright metal parts is collected and established.Secondly,many traditional image data augmentation methods are studied,and the limitations of traditional image data augmentation methods are verified by experiments.Combined with the characteristics of bright metal parts,an improved method of metal parts image dataset augmentation based on the Deep Convolutional Generative Adversarial Network(DCGAN)is proposed.The feasibility of this method in image dataset augmentation and the superiority of improving image quality are verified by comparing experiments with public datasets.Finally,based on the comparison of various object detection methods based on deep learning and the characteristics of the multi-scale Feature Pyramid Network(FPN),an improved method based on the multi-scale feature online detection and recognition for metal parts is proposed.The effectiveness of this method in the detection and recognition of bright metal parts is verified by experiments with different object detection methods.The results show that the average detection accuracy of this method can reach 99.2%,which can meet the accuracy of the industrial detection requirements.The average detection speed of this method is 0.093 seconds,also can meet the realtime requirements of industrial detection.To sum up,this thesis takes the bright metal parts as the research object and uses the improved deep convolutional generative adversarial network to expand the dataset of bright metal parts.The improved multi-scale feature pyramid network is combined with the object detection algorithm to accomplish the online detection and recognition of metal parts based on machine vision.