Deep Learning Based Syringe Defect Detection Research

Product quality inspection is the key to industrial development.With the continuous improvement of product quality requirements,the profits it brings also increase.Therefore,the inspection of surface defects has become an integral part of the production process when producing high-quality products.In recent years,the CONVID-19 epidemic has spread around the world,and the use of syringes also skyrocketed.With the increase in syringes production,the demand for syringe surface defect detection has also increased.A medical syringe defect not only affects the appearance of the syringe but may further endanger the patients.Therefore,surface defect detection is a top priority in syringes production.Presently,methods commonly in the domain of syringe surface defect detection include manual inspection,traditional machine vision-based defect detection,and deep learning-based defect detection methods.Among them,manual inspection takes a lot of time and laborintensive,which is hard to satisfy the requirement of industrial defect detection.With the rapid growth of syringe production capacity,its defect detection scenarios are more flexible and changeable,and deep learning methods have also been successfully applied in the field of defect detection.Due to the powerful feature learning ability and complex structure fitting ability of the deep convolutional neural network,it can automatically learn defect features directly from input data,to better detect complex defects.This paper is devoted to the research on the surface defect detection system of medical syringes based on deep learning.The specific research contents include the following aspects:1.The design of medical syringes surfaces defect detection scheme.According to the characteristics of the syringes,this paper designs a defect detection scheme to detect syringe surface defects.This scheme includes two parts,that is,defect detection equipment and the corresponding algorithms.In the defect equipment part,this paper designs its structure and selects the model of components,such as cameras,lenses,and light sources.In the algorithm part,this paper describes the algorithm framework and provides the concrete realization way of the algorithm.2.Research on syringe defect segmentation network.This paper proposes a defect segmentation network with multiple side-branch structures for syringes.To better train the segmentation network,a joint loss function is proposed to release the influence of data imbalance.To further analyze the proposed network,the paper implements the ablation experiments to verify the importance of each component of the network and the joint loss function.In addition,this paper proposes a preprocessing scheme for the medical syringe after analyzing the efficient preprocessing technologies.3.Research on defect classification network.Aiming to improve the defect detection accuracy on syringes,this paper utilizes a classification network to determine whether the defect exists in the segmentation result.4.Syringe surface defect detection method validation.This paper implements experiments to verify the proposed method with a syringe dataset.The experiment result demonstrates that the accuracy and stability of the proposed method can satisfy the requirements of industrial production.