Study On Small Target Recognition Method Of Artemisinin Purification Under Fuzzy Noise

Artemisinin is one of the main drugs for malaria treatment in the world medical community.It is the main medical product to promote the internationalization of the traditional Chinese medicine industry in China.At present,the industrial preparation of artemisinin mainly adopts the traditional process mode,using organic solvent to extract,then through column chromatography,and finally through crystallization and recrystallization to refine,and finally obtain artemisinin.Many procedures in the production workshop involve labor,which greatly hinders the industrial production efficiency of artemisinin and increases the industrial production cost of artemisinin.Therefore,this paper designs an intelligent scheme for the column chromatography process in the artemisinin industrial production process,simulates the manual monitoring of artemisinin content in the eluent,and realizes the impersonal monitoring operation in the column chromatography process.In this paper,the method of extracting candidate detection regions from salient features is used to optimize the input of small target detection network to track and identify artemisinin white crystal particles separated from the glass panel after eluent volatilization in a fuzzy noise environment,so as to reduce the false detection rate of artemisinin white crystals.Based on the cavity convolution,the density counting network is designed to count the white artemisinin crystals,so as to quantify the artemisinin crystals separated from the eluent.Finally,it will replace the manual monitoring of artemisinin content in the eluent in the chromatographic column process of artemisinin industrial preparation process,reduce the manual intervention in the production and preparation process of artemisinin,improve the productivity of artemisinin in industrial preparation engineering,and improve the purity of artemisinin products.The main research results of this paper are as follows:(1)Collect the image data of artemisinin precipitated white crystal;Based on the image denoising method,data preprocessing is completed to reduce the influence of imaging factors such as impurities and Reflections on small target detection in the original image data;Based on the data enhancement method,the image data is enhanced without changing the distribution of data features.Make the data meet the needs of deep learning training,verification and testing.(2)The method of extracting detection candidate regions based on salient features is studied to optimize the input of small target detection network for deep learning,improve the speed of small target detection network in the reasoning process,and reduce the false detection rate of small target detection network;Comparative experiments are designed to verify the effectiveness of extracting detection candidate regions based on salient features to reduce the false detection rate of small target detection models through popular small target detection networks,such as yolo-v5,SSD,hrdnet and reasoning RCNN.(3)Based on the cavity convolution,the target counting method in the deep learning is studied,the target counting network is designed,the white crystal density map of artemisinin is generated,and the white crystal particles of artemisinin are quantified,so as to achieve the purpose of monitoring the artemisinin content in the eluent of column chromatography;A comparative experiment is designed and compared with a more effective target counting network to prove the effectiveness of the target counting network designed in this paper.(4)Based on Jetson TX2,intelligent hardware equipment is built to realize the timed extraction and volatilization of the extract from the industrial preparation of artemisinin;Accelerate the deep learning network based on tensorrt framework,deploy the algorithm on intelligent hardware devices,and realize the intelligent identification and quantification of artemisinin white crystals;Finally,through the actual debugging,it verifies the effectiveness of the algorithm designed in this paper and building intelligent devices.