Research And System Design Of Sanitary Ceramic Defect Detection Based On Deep Learning

In recent years,as the sanitary ceramic industry has continuously increased the requirements for the intelligence of defect detection systems,the demand for microdefects on irregular surfaces has become greater and greater.Traditional detection methods for regular spheres or planes with diffuse reflection properties are difficult to meet the requirements of use,and deep learning-based defect detection technology still has the characteristics of high accuracy and strong generalization on complex surfaces.An effective way to efficiently obtain surface defects.In this paper,the bathroom ceramics after glazing and firing are taken as the research object,and the combination of deep learning and computer vision is used to study the detection of surface micro-defects.The fired sanitary ceramic surface is smooth and easy to reflect,and has an irregular outer contour.The contrast between the micro-defects existing on the surface of the workpiece and the background is not obvious,which increases the difficulty of locating the defect area by the model.In addition,different types of defects usually have similar external contours,and small differences between classes increase the difficulty of distinguishing the types of defects in the model.In view of the above difficulties,this article conducts systematic in-depth research on two aspects of model selection and improvement of the defect detection framework.The main contents and innovations of the article are as follows:Firstly,taking the micro-defects on the surface of bathroom ceramics as the research object,the development process of defect detection at home and abroad is expounded,and the defect detection model based on deep learning is selected as the main solution,and further research is done on the solution.Secondly,through the design of comparative experiments,the research focuses on the detection performance of the surface micro-defects of the model,and compares the more commonly used Faster R-CNN,YOLO-V3(You Only Look Once,YOLO),and SSD(Single Shot Multi Box Detector,SSD)3 kinds of target detection framework.From the experimental results,it is known that Faster R-CNN has higher accuracy in identifying and locating micro-defects than YOLO-V3 and SSD.Thirdly,an optimization method for detecting micro-defects on the surface of sanitary ceramics is proposed.First,according to the morphological characteristics of surface defects,the K-Means algorithm with BWP(Between-Within Proportion,BWP)as the evaluation index was introduced to improve the initial bounding box generation mechanism to improve the positioning accuracy of the model;secondly,in the VGG16 network A feature fusion mechanism is introduced in the model to improve the recognition accuracy of small defects by fusing feature matrices with different receptive fields.The experimental results show that the improved model has the best detection performance when the number of cluster centers is 3,the base size for generating the initial bounding box is 8,and the two layers of features after the feature extraction network are fused.Compared with the standard Faster R-CNN,the comprehensive detection performance F-score of the improved model is improved by 2%.Finally,this article designs a set of sanitary ceramic surface defect detection system,and introduces in detail the hardware platform,system operation logic and functional flow of each module required to run the software.The results of software functional verification show that the bathroom ceramic defect detection system can complete the set task goals according to the inspection process.