Visual Inspection Methods Of Automobile Instrument Based On Prior Information Extraction

With the increasing demand for intelligent production in industry,the research on image segmentation and localization methods in automated optical inspection systems has been paid more and more attention.Automobile instrument,as an essential component of automobile,is in great demand in automobile production.The research on automobile instrument detection technology is very important,which is directly related to the improvement of product quality.However,the diversity of automobile instrument products and the complexity of the industrial environment challenge the development of related inspection methods.This paper designs prior information extraction-based visual detection methods to solve the above challenges.Two models of target segmentation in standardized products are proposed and used to solve the problem of image segmentation in automobile instrument inspection.This paper comprehensively analyzes the design of automobile instrument inspection system and the development of detection algorithms,aiming at enhancing the system’s detection performance.The research works included in this paper are as follows:To address the problems of insufficient data and the big variances across nonhomologous data in industrial inspection,multiscale feature correlation perception network(MFCP-Net),a semantic segmentation model with variable topology,is proposed to segment images with prior background.MFCP-Net uses a Siamese structure with shared weights to extract the multiscale features of the template and test images.A feature correlation perception block is designed to measure the correlation of latent representation between the template and test images at multiscales,using background information more effectively.In addition,MFCP-Net provides different topological structures for the training and testing phases.A background auto-correlation branch is included during the testing phase to eliminate false positives observed in the background.MFCP-Net was compared with the state-of-the-art models on the proposed automobile instrument detection system,demonstrating that it offers superior accuracy and generalizability,especially in the case of insufficient data and big variances across data.To enhance the detection capability of deep neural networks for non-homologous data,a Siamese Swin-Transformer network with difference perception is proposed to segment images with prior background.The proposed network first constructs an encoder with a dual-branch structure based on the Swin-Transformer module,such that the extracted features contain long-range dependencies,improving the feature representation capability of the encoder.On the basis of the spatial-channel attention mechanism,a difference feature extraction module is then proposed,which detects the difference between the output features extracted from the two branches of the encoder and increases attention to the region of interest.Finally,the decoder predicts segmentation using difference perception features and deep semantic of the test image.Experimental validation on two image segmentation datasets demonstrates that the proposed network outperforms state-ofthe-art models and significantly improves the detection performance on non-homogenous data.Since there are many kinds of automobile instruments and the detection environment is usually complex,automatic instrument detection during automobile production test becomes a challenging task.Therefore,an automobile pointer instrument detection method based on prior information and fuzzy sets is proposed.The proposed method consists of two frameworks built around a pointer meter prior information model(PMPIM).The first one targets PMPIM construction to obtain the required prior information.With this purpose,a pointer-free template is obtained from a template generation algorithm and pointer positions are mapped into an energy function space for optimization,using an energy function-based pointer positioning algorithm.The energy function is defined based on the distance between the crisp and fuzzy sets.The second framework targets PMPIM utilization to detect pointer meters during production test.A fuzzy-based image enhancement method is proposed to enhance test images and the template simultaneously.Finally,PMPIM is introduced into the pointer positioning of the test image,and the parameters of the pointer symmetry axis are calculated.Experimental results show that the proposed method achieves better generalization and robustness than existing state-of-the-art methods.Considering that a manufacturing line is necessary to detect various types of instruments,a unified analog detection framework for the indicated values of pointer instruments is proposed.The proposed framework innovatively gives the corresponding solutions in view of the existing problems for each subtask.First,a template image is established by pixelwise background modeling.Then,the two-stage similarity measurement based on fuzzy theory and image reconstruction is designed to enhance images and emphasize pointers,boosting the generalization of the detection framework.For the pointer positioning,the edges of pointers are fit by the set-to-set distances introducing the directionality and intensity of pointers to improve the robustness.For the scale marks’ positioning,the spatial periodicity of scale marks is creatively introduced to address the challenges caused by noises and the connected scale marks.Finally,the indicator value is calculated according to the obtained positions of the pointer and scale marks.Experimental results show that the proposed detection framework can reliably detect various types of pointer meters,even under complex illumination conditions.In conclusion,this paper addresses the challenging issues encountered by visual inspection technology in automobile instrument inspection systems.It proposes a comprehensive solution that encompasses fundamental image segmentation methods,positioning techniques,and identification measurement methods,thereby forming a holistic approach that spans from basic tasks to specific objectives.The experimental results show that the proposed method effectively solves complex problems related to visual inspection tasks of automobile instruments.