Paticle Defect Detection Based On Machine Vision

Particulate matter is the basic form of raw materials for people's daily necessities.With the development of economy and the improvement of people's quality of life,there is an urgent need for particulate matter with higher quality.At present,the methods of manual inspection and sample sampling commonly used in the quality management of particulate matter production process have some problems,such as low accuracy,long time-consuming and lack of data informatization.To solve these problems,this paper proposes a method based on machine vision for particle detection and recognition.In this paper,plastic particles in industrial production are taken as research examples.The design of particle detection platform is divided into three parts: target detection,target tracking,target defect classification.The specific work mainly includes the following three aspects:(1)In order to accurately locate the particle targets in the detection field,the target detection network SSD based on deep learning is proposed to complete the detection of particles.The model parameters and training parameters of SSD network are analyzed and designed in detail.The detection performance of 99.4%mAP is obtained through training and testing,and it has good robustness to particle adhesion and partial occlusion.(2)In order to accurately count the number of particles in the detected field of vision,a tracking-by-detection framework is proposed,which first calculates the sparse optical flow characteristics of adjacent frame images,and then correlates the targets of the adjacent frame images based on the optical flow characteristics.(3)Particulate matter defects in this paper are mainly divided into blob,poor shape and color difference.For blob,a method based on local threshold is proposed,which is suitable for blob defect segmentation.It can effectively segment the connecting area of blob defect in particulate matter.For shape and color difference defect,the idea of template matching is used,and the standard particle feature template is established and compared.Specifically,for the defect of poor shape of particulate matter,statistical detection of particles is carried out.Among the four geometric features of particle size,area,circumference,aspect ratio and rectangularity,defective particles and standard template have obvious distinction.By setting appropriate thresholds for the four characteristics,the particulate matter with bad shape can be distinguished.For color defects,this paper presents a histogram statistics of particulate matter's HSV colorspace and calculates histogram intersection.The method can effectively distinguish the color defective particles from the standard template.The experimental results show that the algorithm and detection process used in this paper are more feasible and effective,and can accurately detect and count the particles and defects in real time,which satisfies the relevant detection requirements of the actual production line.