Drug Surface Defect Detection Based On Deep Learning

Surface defects of tablets are an important factor affecting the quality of tablets.Surface defect detection has become an indispensable technical link in the production process of pharmaceuticals.At present,most domestic manufacturers still use traditional manual detection methods,and there are generally problems such as low detection efficiency and strong subjective factors.The use of advanced machine vision technology to study the automatic detection system of tablet quality defects and realize automated tablet defect detection has become an inevitable trend in the development of the pharmaceutical industry.According to the actual needs of the tablet manufacturers,the research team carried out the design and development of the machine vision-based bulk defect surface detection technology system.The research work in this paper mainly focuses on the design and construction of the image acquisition system for bulk tablets,as well as the research and implementation of the surface defect detection algorithm for bulk tablets.The main tasks that have been completed include:An image acquisition system for bulk tablets was designed and developed.According to the actual requirements of the enterprise,the overall design,hardware selection,platform assembly and acquisition testing of the image acquisition system were completed.The system adopts a ring-shaped LED light source,and adopts the Daheng MER-132-30GC/M industrial camera and the OPT-C2514-2M lens to take a frontal illumination to obtain the surface image of the bulk tablets.And through testing,the appropriate image acquisition scheme was determined.In addition,the design of related devices such as sorting mechanism,rejection mechanism and turning mechanism was initially completed.The collection,processing and database construction of the sample images of the tablets.The image collection work of 1200 tablets of bulk tablets provided by the company was completed,including 300 tablets of qualified tablets,defective tablets,scratched tablets and stained tablets;and 1200 original image databases of bulk pills were constructed.In order to eliminate the redundant information in the tablets and reduce the amount of data,the original image of the bulk tablets was filtered,edge-detected,etc.,and 1200 processed batch edge image databases of bulk tablets were constructed.In order to improve the accuracy and generalization performance of the model detection,the tablet edge image database was expanded by simple data enhancement methods such as rotation and flipping.The enhanced database contains 1200 sheets of various types of tablets,totaling 4,800 sheets.The optimized CNN model is used to test and verify the performance of image edge processing and the feasibility of simple data enhancement methods.Aiming at the problems of low detection accuracy,small application range and cumbersome process of traditional pattern recognition methods,a deep defect-based surface defect detection algorithm for bulk tablets was proposed.Based on the VGG model,the basic convolutional neural network model was built for the surface defect detection task of the tablet,and the network model was gradually trained by 75% data in the enhanced database.Aiming at the over-fitting problem of the network model in the training process,the batch normalization layer and the Dropout layer are added in the model,and the early stop strategy is used in the model training phase,which effectively suppresses the model over-fitting.The model structure is gradually optimized according to the training results,and the optimized Mini VGG network model is finally obtained.After the actual test,the detection accuracy of the optimization model can reach 98%,which is significantly higher than the detection algorithm based on pattern recognition.Finally,based on the above research work,this paper proposes a further research direction of the defect detection algorithm based on convolutional neural network.