Research On Visual Recognition Method Of Gear Shaft Parts Based On Image Processing And Deep Learning

Grinding is a post-finishing process in the manufacturing process of gear shafts.Correctly grinding the gear shaft is the last link to ensure the smooth completion of the entire production.However,with the continuous development of mixed production methods with multiple varieties and small batches,the traditional method of grinding parts by workers recognizing the types of numerous parts and then retrieving and calling corresponding processing programs has problems of low automation and difficulty in ensuring recognition accuracy.In addition,if the worker does not clamp the gear shaft in the correct way,the processing of the gear shaft to be ground will be scrapped,and even the machine tool equipment will be damaged.The target recognition and clamping direction recognition of gear shaft parts are performed online based on vision,and the corresponding processing program is called according to the recognized parts,which can improve the grinding efficiency and accuracy of gear shafts.Therefore,this paper conducts research on the visual recognition method of gear shaft parts.The main research contents are as follows:(1)Build a visual recognition and image acquisition platform.According to the actual processing environment,the industrial camera,lens and light source are selected,and the visual recognition hardware platform is built.Aiming at the color deviation and distortion of industrial cameras,the dynamic threshold white balance algorithm and camera calibration are used for correction.An image acquisition software platform is built on the hardware platform of visual recognition,which provides an image acquisition tool for the visual recognition research in this paper.(2)Research gear shaft image enhancement algorithm.Aiming at the problem of uneven illumination and blurring of gear shaft images collected in the processing environment,several classical image enhancement algorithms are used for testing and analysis.Based on the characteristics of the classical algorithm and the shortcomings of the processing effect,a gear shaft image enhancement algorithm based on Gaussian pyramid multi-scale illumination decomposition is proposed,which provides image preprocessing function for gear shaft target recognition.The effectiveness of the algorithm in uniform image brightness distribution,improving image clarity and enriching image features is verified by image quality evaluation experiments.(3)Research on the target recognition method of gear shaft parts.Aiming at the problems of similar shape,single texture feature and complex background of gear shaft,the convolution attention mechanism module is introduced into the YOLOv5 algorithm model,and the target recognition method of gear shaft parts is designed.The target recognition of gear shaft parts is realized by using the collected image data set for model training.Combined with the image enhancement algorithm for experiment and analysis,the effectiveness of the image enhancement algorithm and target recognition method studied in this paper in improving the recognition accuracy of gear shaft parts is verified.(4)The recognition method of clamping direction of gear shaft parts is studied.Aiming at the problem of low segmentation accuracy caused by the small difference between the background region and the target part,the influence of the background region on the image segmentation accuracy is eliminated by locating the target gear shaft.The Res-UNet model was built and trained to realize the segmentation of gear shaft images.By extracting the edge contour of the gear shaft,the center of gravity of the contour and the center point coordinates of the circumscribed rectangle are detected,and the clamping direction of the gear shaft parts is identified.The accuracy of this method in identifying the clamping direction of gear shaft is verified by experiments.(5)A visual recognition software system for gear shaft parts is designed and tested.In order to complete the visual recognition task of gear shaft parts,based on the built visual recognition hardware platform,the image enhancement and visual recognition algorithms studied in this paper are integrated,and the visual recognition software system of gear shaft parts is designed and developed.Through the functional test of the software system and the performance test of the visual recognition algorithm,the feasibility of the software system in function and time performance is verified.