CN-121299004-B - Method and related apparatus for identifying pharmaceutical compositions

CN121299004BCN 121299004 BCN121299004 BCN 121299004BCN-121299004-B

Abstract

The embodiment of the application provides a method and related equipment for identifying medicinal components, and belongs to the technical field of medicament identification. The method comprises the steps of obtaining preset medicine sample data, including positive medicine sample data and negative medicine sample data, wherein the positive medicine sample data comprise a first thin-layer chromatographic spectrum, the negative medicine sample data comprise a second thin-layer chromatographic spectrum, constructing a model training spectrum set through the first thin-layer chromatographic spectrum to train a preset twin neural network model to obtain a first twin neural network model, constructing a model testing spectrum set through the first thin-layer chromatographic spectrum and the second thin-layer chromatographic spectrum to optimize parameters of the model by combining with the model training spectrum set to obtain a target neural network model, and inputting the thin-layer chromatographic spectrum to be identified into the target neural network model to perform component identification analysis to obtain a component identification result. The embodiment of the application can avoid the use of reference substances and reference medicinal materials as much as possible in the drug identification, thereby achieving the purposes of economy, green and environmental protection.

Inventors

Jiang Junhuang

Assignees

暨南大学

Dates

Publication Date: 20260508
Application Date: 20250928

Claims (7)

1. A method of identifying a pharmaceutical composition, the method comprising the steps of: The method comprises the steps of obtaining preset medicine sample data, wherein the preset medicine sample data comprise positive medicine sample data and negative medicine sample data, the positive medicine sample data comprise a first thin-layer chromatographic spectrum of a medicine sample to be identified, and the negative medicine sample data comprise a second thin-layer chromatographic spectrum of a non-identified medicine sample; constructing a model training atlas through the first thin-layer chromatography atlas; Training a preset twin neural network model according to the model training atlas set to obtain a first twin neural network model; constructing a model test spectrum set through the first thin-layer chromatography spectrum and the second thin-layer chromatography spectrum; Performing parameter optimization on the first twin neural network model according to the model test atlas set and the model training atlas set to obtain a target neural network model; Inputting the thin-layer chromatographic pattern to be identified into the target neural network model for component identification analysis to obtain a component identification result; wherein, after the acquiring the preset drug sample data, the method further comprises: Cutting each map in the preset medicine sample data according to a preset display area to obtain cut maps, wherein the preset medicine sample data is obtained through a preset medicine thin-layer chromatography set; adjusting the cutting map according to a preset image size to obtain target drug sample data; Training a preset twin neural network model according to the model training atlas set to obtain a first twin neural network model, including: randomly dividing the model training atlas set according to a preset dividing proportion to obtain a training data set and a verification data set; Inputting the training data set into the preset twin neural network model, and further carrying out feature extraction through a double-branch convolution feature extraction module in the preset twin neural network model to obtain target feature data, wherein the target feature data comprises a first feature vector and a second feature vector; performing contrast loss calculation according to the first feature vector and the second feature vector, and optimizing model parameters through a counter propagation and gradient descent mechanism according to the calculated contrast loss data to obtain a second twin neural network model; inputting the verification data set into the second twin neural network model for verification analysis to obtain verification loss data; When the verification loss data meets the preset early-stop braking condition, carrying out parameter updating on the second twin neural network model according to preset model parameters to obtain the first twin neural network model, wherein the preset model parameters comprise model parameters corresponding to the verification loss data in a minimum state; The parameter optimization is performed on the first twin neural network model according to the model test atlas set and the model training atlas set to obtain a target neural network model, and the method comprises the following steps: inputting the model test atlas set and the model training atlas set into the first twin neural network model to determine a sample prediction category through a minimum distance matching algorithm; and carrying out credibility analysis through a preset threshold mechanism according to the sample prediction category, and further adjusting a distance threshold parameter of the first twin neural network model according to an analysis result to obtain the target neural network model.
2. The method according to claim 1, wherein inputting the training data set into the preset twin neural network model, and further performing feature extraction by a dual-branch convolution feature extraction module in the preset twin neural network model, to obtain target feature data, includes: performing feature extraction on the first sample data of the training data set through a first convolution feature extraction sub-module to obtain third feature data; Performing feature extraction on second sample data of the training data set through a second convolution feature extraction submodule to obtain fourth feature data, wherein the first convolution feature extraction submodule and the second convolution feature extraction submodule comprise a first convolution layer, a second convolution layer, a third convolution layer, a batch normalization layer, an activation function layer and a pooling layer; and respectively performing reduction and normalization processing on the third characteristic data and the fourth characteristic data through a characteristic compression and normalization sub-module to obtain the target characteristic data.
3. The method according to claim 1, wherein after performing the inputting the thin-layer chromatography to be identified into the target neural network model for component identification analysis, the method further comprises: Comparing and analyzing the components according to the component identification result through preset component standard parameters to obtain a medicine effectiveness analysis result; And generating a medicine component analysis report according to the medicine effectiveness analysis result and the thin-layer chromatographic map to be identified, and uploading the medicine component analysis report to a preset management platform.
4. A pharmaceutical composition identification device, characterized in that it is applied to the method according to any one of claims 1 to 3, comprising: The device comprises a first module, a second module and a third module, wherein the first module is used for acquiring preset medicine sample data, the preset medicine sample data comprises positive medicine sample data and negative medicine sample data, the positive medicine sample data comprises a first thin-layer chromatographic spectrum of a medicine sample to be identified, and the negative medicine sample data comprises a second thin-layer chromatographic spectrum of a non-identified medicine sample; the second module is used for constructing a model training spectrum set through the first thin-layer chromatographic spectrum; the third module is used for training a preset twin neural network model according to the model training atlas set to obtain a first twin neural network model; A fourth module for constructing a model test pattern set from the first thin layer chromatography pattern and the second thin layer chromatography pattern; a fifth module, configured to perform parameter optimization on the first twin neural network model according to the model test atlas set and the model training atlas set, so as to obtain a target neural network model; And a sixth module, configured to input the thin-layer chromatographic spectrum to be identified into the target neural network model for component identification analysis, so as to obtain a component identification result.
5. An electronic device, comprising: at least one processor; at least one memory for storing at least one program; The at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1 to 3.
6. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 3.
7. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 3.

Description

Method and related apparatus for identifying pharmaceutical compositions Technical Field The application relates to the technical field of medicine identification, in particular to a medicine component identification method and related equipment. Background The Chinese patent medicine is a compound preparation prepared by taking Chinese medicinal materials as raw materials and processing the Chinese medicinal materials by a modern process, and has the characteristics of complex prescription, various components, wide sources and the like. The quality control of the Chinese medicinal materials is difficult due to the differences of the growth environment, harvesting season, processing method and the like of the Chinese medicinal materials and the influence of the technological conditions in the production process of the Chinese patent medicines. In the related art, the analysis and identification method for the medicine components is often high in cost and not green and environment-friendly. In summary, the technical problems in the related art are to be improved. Disclosure of Invention The embodiment of the application mainly aims to provide a medicine component identification method and related equipment, which can avoid the use of reference substances and reference medicinal materials as much as possible in medicine component identification, thereby achieving the purposes of economy, green and environmental protection. To achieve the above object, an aspect of an embodiment of the present application provides a method for identifying a pharmaceutical ingredient, the method including: The method comprises the steps of obtaining preset medicine sample data, wherein the preset medicine sample data comprise positive medicine sample data and negative medicine sample data, the positive medicine sample data comprise a first thin-layer chromatographic spectrum of a medicine sample to be identified, and the negative medicine sample data comprise a second thin-layer chromatographic spectrum of a non-identified medicine sample; constructing a model training atlas through the first thin-layer chromatography atlas; Training a preset twin neural network model according to the model training atlas set to obtain a first twin neural network model; constructing a model test spectrum set through the first thin-layer chromatography spectrum and the second thin-layer chromatography spectrum; Performing parameter optimization on the first twin neural network model according to the model test atlas set and the model training atlas set to obtain a target neural network model; inputting the thin-layer chromatographic pattern to be identified into the target neural network model for component identification analysis to obtain a component identification result. In some embodiments, after the acquiring the preset drug sample data, the method further comprises: Cutting each map in the preset medicine sample data according to a preset display area to obtain cut maps, wherein the preset medicine sample data is obtained through a preset medicine thin-layer chromatography set; and adjusting the cutting map according to a preset image size to obtain target medicine sample data. In some embodiments, training the preset twin neural network model according to the model training atlas set to obtain a first twin neural network model includes: randomly dividing the model training atlas set according to a preset dividing proportion to obtain a training data set and a verification data set; Inputting the training data set into the preset twin neural network model, and further carrying out feature extraction through a double-branch convolution feature extraction module in the preset twin neural network model to obtain target feature data, wherein the target feature data comprises a first feature vector and a second feature vector; performing contrast loss calculation according to the first feature vector and the second feature vector, and optimizing model parameters through a counter propagation and gradient descent mechanism according to the calculated contrast loss data to obtain a second twin neural network model; inputting the verification data set into the second twin neural network model for verification analysis to obtain verification loss data; And when the verification loss data is determined to meet the preset early-stop braking condition, carrying out parameter updating on the second twin neural network model according to preset model parameters to obtain the first twin neural network model, wherein the preset model parameters comprise model parameters corresponding to the verification loss data in a minimum state. In some embodiments, inputting the training data set into the preset twin neural network model, and further performing feature extraction by a dual-branch convolution feature extraction module in the preset twin neural network model to obtain target feature data, where the method includes: performing feature extraction on the