Research On Bearing Cover Classification Method Based On Machine Vision

The car engine bearing cover is mounted on the engine block to hold and support the crankshaft.Several bearing covers are required for any type of engine.Different engines require different bearing covers.The bearing convers in different positions on the same engine are also different.In the process of multi-variety production,it is necessary to classify the engine bearing covers before delivery.At present,the classification of bearing covers is mainly completed manually,with slow speed,high cost and poor stability.This paper studies the online and real-time classification method of bearing covers based on machine vision,which is mainly based on the features of end face geometry,top digital character,top geometric symbol and local plane shape of bearing cover.These features basically cover all the classification identification features of bearing covers that have been adopted at present.In order to realize the automatic classification of machine vision on bearing covers,this paper first designs a set of online image acquisition system,adopts different lighting schemes according to different classification identification features,and highlights the classification identification features to be identified.In the section of image processing,this paper briefly describes the relevant image processing algorithms,divides various classification and recognition features from the original image,realizes the identification and location of the identification features,and extracts the contour elements of various classification and identification features.For the end shape image,the end shape may be flipped horizontally along the image in different acquisition images,and in this paper,an image feature extraction method with invariability of it is proposed.For digital character image and local plane shape image,the image feature of Freeman direction code histogram is adopted in this paper,which overcomes the shortcoming of fixed image size and large computation required by Histogram Orientated Gradient image feature.With image features of the identification features of various parts extracted by the above methods and their category information,this paper adopts supervised classification method of Support Vector Machine.In the case that only a small number of category information of parts images are known,a semi-supervised spectral clustering method based on pairwise constraints is adopted,where pairwise constraints are searched by active learning algorithm.For the random selection of initial sample of the active learning algorithm Min-Max and the disadvantage of the selection range of samples as the whole dataset,and active learning algorithm based on density coefficient is proposed in this paper.This paper also carries on the Support Vector Machine supervised classification experiment and semi-supervised clustering experiment based on active learning for the part images.In terms of supervised classification,the classification accuracy of test set of bearing cover end face shape images,concave character images and local shape images is 99.8%,100% and 100% respectively.In terms of semi-supervised clustering,this paper carries out semi-supervised spectral clustering experiments based on FFQS algorithm,Min-Max algorithm and the active learning algorithm proposed in this paper respectively on the bearing cover image data set.The experimental results show that the active learning algorithm based on density coefficient proposed in this paper has a better guiding effect on the clustering process in the dataset of bearing cap end face shape images.In the classification test based on semi-supervised clustering results,the classification accuracy of the end shape image test set is 96.1%.