Penicillin Crystallization State Detection And Analysis Based On Machine Vision

The industrial crystallization process of penicillin is divided into three stages: the earlier stage,evaporation with the increasing temperature.The middle stage,crystal growth and crystallization at the constant temperature.The later stage,evaporation with the rapidly rising temperature.Whether the crystallization state can be accurately detected in the middle stage is the key for penicillin crystallization production.At present,the detection method of crystallization state completely relies on artificial way,which causes some problems,such as individual subjective differences,inefficiency and high labor costs,etc.Combined the actual demand of crystallization state detection for penicillin pharmaceutical manufacturer with the application research based on machine vision in industrial production,penicillin crystallization state detection system based on machine vision is designed.Many of the drawbacks of artificial detection can be removed.The purpose of replacing artificial detection can be achieved.The main research contents are as follows:According to the investigation from the crystallization workshop of the pharmaceutical manufacturer,design principles and functional requirements are determined,the design of the whole framework of the detection system is completed.Analyzing the image characteristics of penicillin crystallization reaction and the factors that affecting crystallization state detection,the detect criteria is explicated and the detection and identification research is carried out.Firstly,in order to effectively reduce the interference factors,such as noise,light,fog,turbidity,stirring shaft and flushing pipeline,which causes the interference to the collection of images.A series of image pre-processing operations are carried out before crystallization state detection algorithm is designed.Image pre-processing operations includes image basic processing(image scaling,image filtering,white balance processing),image blur detection and extraction of detection object.Setting the criteria for specular blur insure effectively washing the mirror and detectability for image.In order to extract the detection object,an improved K-means algorithm on the basis of the V component of HSV color space is presented.Combined the edge detection with the mathematical morphology,the improved K-means algorithm can remove out the stirring shaft and flushing pipeline shadow area and obtain detection object.Secondly,the algorithm of crystallization state detection based on the image pre-processing result is studied in depth.The crystallization status detection and recognition algorithm based on entropy-weighted support vector machine is designed.The color and texture characteristics of crystallization status images is extracted.Considering the different influences of each feature vector,the entropy-weighted feature vector group is established,therefore the weighted support vector machine detection model is set up.By comparing the simulation results of single feature,multi-feature models and weighted combined feature model,the results show that the weighted feature detection model has more advantages than the the first two-models.Finally,the detection is carried out on the platform built by raspberry PI embedded development environment.By comparing the results of the detection system and artificial detection,it is shown that the effect of penicillin crystallization state detection system and artificial detection is consistent.The moisture measurement results are also qualified and proved that the detection system can be used on instead of artificial to detection and can accurately detect the crystallization state.