Research On Medical Bottle Package Defects Recognition Based On BP Neural Network

Medical-used infusion is one of the five important preparations of the pharmaceutical industries in China.They are frequently-used in medical institutions and playing a very important role in modern clinic.However,during the process of packaging,it appears easily sealed and lax circumstance for the bottle and cap because of packaging technology.It will not only affect the quality of drugs seriously,threat to patients,but impact on the next process.In order to enhance the quality of drugs,it is necessary to design and realize the automatic classification method for defects recognition,and remove the defective goods.According to the characteristic of images with package defect,the paper used the technologies of image processing and pattern recognition,and designed an automatic defect recognition algorithm based on BP neural network,overcame the shortcomings of traditional manual testing.The main research contents of this paper are as follows:(1)In order to obtain high quality binary image,the image processing algorithm was studied.Firstly,gray scale processing of the image was carried out.Second,on the basis of gray image,a variety of algorithms for image filtering,image enhancement,image segmentation and morphological operations were used in the experiment.Last,selected a feasible and advanced image processing operation flow after compared and analyzed.(2)Bottle image features were extracted and collected.Used characteristics of different types of defects,took the principle of following feature extract,shape features and texture features were extracted,which paved the way for further analysis.(3)Inspect package flaws by taking pattern recognition measures.First of all,the structure,learning algorithm and design method of BP neural network were introduced in detail,after that the improved momentum BP method was used to identify the defects of the package because of the drawbacks that the traditional BP neural network algorithm had.And the recognition results were verified by experiments.Secondly,the paper analyzed system error,proposed instruction of decreasing error for future research.At last,designed this system's GUI,achieved a good human-machine interaction interface.