Research On Abnormal Identification And Processing System Of Automatic Dispensing Of Traditional Chinese Medicine

With the development of traditional Chinese medicine industry,the traditional Chinese medicine slice industry has realized mechanized and automatic production,but the prescription dispensing of TCM pharmacy still depends on the pharmacist’s "hand weighing",which is inefficient and difficult to ensure the accuracy of weighing.The automation and intelligence of TCM pharmacy has become the inevitable trend of the development of modern medical system.The vigorous market demand has prompted the research and development and launch of a bath of automatic dispensing equipment for traditional Chinese medicine,but most of its design ideas follow the automatic dispensing equipment of Western medicine,which is difficult to meet the essence of TCM syndrome differentiation,and the prescription verification after dispensing has not been considered.Medicine is a special commodity,which has an important impact on patients’ health and life safety.Errors in prescription dispensing will lead to serious medical accidents.Therefore,the recheck of dispensing results of traditional Chinese medicine decoction pieces automatic dispensing equipment and the handling of abnormal dispensing will effectively improve the safety of the automatic dispensing system,ensure the safety of patients’ medication,and further promote the construction of automatic TCM pharmacies.In view of the defects of the automatic dispensing system of traditional Chinese medicine,this thesis designs the abnormal identification and processing system of automatic dispensing of traditional Chinese medicine based on the image recognition algorithm,realizes the prescription check of automatic dispensing of traditional Chinese medicine,and shunts the abnormal prescription to the manual processing station to make up for the defects of the automatic dispensing system.Firstly,according to the structural characteristics and working mode of the automatic dispensing equipment,the functional requirements of the anomaly identification and processing system are determined,and the hardware equipment and working process of anomaly identification and processing are determined respectively,so as to determine the overall design scheme.Secondly,according to the functional requirements of the anomaly recognition system,calculate and determine the basic parameters of the required industrial camera,and determine the installation position of the camera and the image acquisition scheme.The collected images are classified by image classifier to eliminate the redundant images that do not contain decoction pieces,so as to reduce the demand for storage space and computing power of host computer.The image data set of traditional Chinese medicine decoction pieces with 80 kinds of more than 60000 images is collected personally,combined with web crawler and data expansion technology.Then,the deep learning technology is used to recognize the collected images of traditional Chinese medicine decoction pieces,so as to obtain the dispensing results.Based on the consideration of system performance and the requirement of recognition accuracy,the Efficientnet neural network model is adopted,and the parameter finetuning based on Image Net pretraining is used to accelerate the convergence speed of model training and avoid over fitting.At the same time,macro accuracy rate,macro recall rate and macro F1 are used to evaluate the training results of the model,which shows the recognition effect of the model.Finally,the accuracy of the model is 94.3%,which has achieved an ideal training effect.Finally,the prototype of automatic dispensing equipment for traditional Chinese medicine pieces is improved and the function test prototype is built.The visual interface of the automatic dispensing exception identification and processing system is designed to realize the functions of prescription information acquisition and distribution,dispensing equipment control,exception identification and exception processing.The results show that the functional modules of the system have achieved the expected design objectives.This thesis has 64 pictures,18 tables and 97 references.