CN213779856U - Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection - Google Patents
Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection Download PDFInfo
- Publication number
- CN213779856U CN213779856U CN202020138594.3U CN202020138594U CN213779856U CN 213779856 U CN213779856 U CN 213779856U CN 202020138594 U CN202020138594 U CN 202020138594U CN 213779856 U CN213779856 U CN 213779856U
- Authority
- CN
- China
- Prior art keywords
- chinese medicine
- hyperspectral
- traditional chinese
- hyperspectral image
- medicine injection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The utility model relates to an application field of hyperspectral technique, in particular to many indexes of traditional chinese medicine injection detecting system based on hyperspectrum. The device comprises a traditional Chinese medicine injection ampoule bottle conveying module, a hyperspectral image acquisition module and a hyperspectral image processing module; the ampoule bottle conveying module conveys the to-be-detected traditional Chinese medicine injection to a hyperspectral image acquisition area, the hyperspectral image acquisition module acquires a hyperspectral image of the injection, and the hyperspectral image processing module performs data processing on the acquired hyperspectral image and analyzes and outputs a chemical component prediction result, an activity prediction result, a chromaticity detection result and a visible foreign matter detection result in the traditional Chinese medicine injection. The high spectrum image is formed only by scanning the traditional Chinese medicine injection bottle on the bottle conveying module, and the extracted spectrum information can predict a plurality of chemical indexes and a plurality of activity indexes and determine the chromaticity; the extracted image information can judge whether the injection has visible foreign matters; greatly simplifying the detection mode and the detection equipment, and saving time and labor.
Description
Technical Field
The utility model relates to an application of high spectrum technique field, in particular to a detecting system who is arranged in a plurality of indexes of simultaneous determination traditional chinese medicine injection.
Background
The traditional Chinese medicine injection is a product combining traditional Chinese medicines and modern preparations, is widely applied to the aspect of treating a series of indications such as cardiovascular and cerebrovascular diseases, viral infection, inflammation and the like, the quality of the traditional Chinese medicine injection needs to be strictly controlled in order to avoid adverse drug reactions during the use of the medicine, the quality detection mode of the traditional Chinese medicine injection at present generally adopts off-line sampling detection, and the detection indexes comprise chemical indexes, visible foreign matters, chromaticity and the like.
For the detection of chemical indexes of injection products, the detection methods generally include thin-layer chromatography identification, high performance liquid chromatography and the like, and the main detection method is to detect the index components of the traditional Chinese medicine injection. Although these methods are stable and reliable, they have the disadvantages of slow detection speed, sample destruction, environmental pollution caused by generated chemical waste liquid, and the like.
Other nondestructive chemical index detection methods such as a method for detecting chemical components in liquid by applying near infrared spectrum and ultraviolet spectrum have a series of reports, for example, a near infrared spectrum analysis method based on a one-dimensional convolutional neural network is disclosed in a patent of China, the application number of which is CN201710780270.2, and the patent generally comprises the following steps: collecting and preprocessing the spectral data of a training set sample; establishing a spectrum correction model by utilizing the preprocessed training set data; collecting spectral data of a test set sample, and preprocessing; and substituting the preprocessed test set data into the spectrum correction model to obtain a prediction result. The establishment of the quantitative correction modeling method generally needs to be carried out through spectrum preprocessing and characteristic wave band extraction. Although the analysis method simplifies the detection process to a certain extent, the following problems still exist: the acquired spectral data is first subjected to spectral preprocessing, which results in unnecessary computational burden. Moreover, the preprocessing method is different with different environmental factors (temperature, humidity and ambient light) and subjective factors of operators, and the popularization performance of the model is limited. Erroneous use of the preprocessing mode may also distort the spectral signal, causing the model accuracy to drop. Although the algorithm for extracting the characteristic bands selects the key bands in the spectrum, the algorithm may cause loss of effective information. Neural Network (NN) can theoretically achieve the purpose of completing modeling without preprocessing and extracting characteristic bands, but because parameter variables in a Neural Network are too many, the training speed is slow, and meanwhile, the risk of overfitting is caused.
The detection of visible foreign matters is a test item specified in general regulations of injections, and the visible foreign matters can cause adverse reactions such as clinical thrombus, inflammation and the like, thereby affecting the medication safety. Generally, the manual light inspection mode is adopted for detection, the efficiency is low, the subjectivity is strong, a light inspection machine is also used for detection, and unqualified samples containing visible foreign matters on a production line can be removed quickly and efficiently. However, the lamp inspection machine market in China is monopolized by three groups of Japan sanitary wares, Italy Bewit and Germany Segen and has very high price, so that most of the Chinese medicine enterprises cannot use the lamp inspection machine to detect products, and an inefficient subjective manual inspection method is adopted.
The chromaticity check of the traditional Chinese medicine injection is one of the physical indexes for controlling the consistency among batches by enterprises, different enterprise chromaticity standards may be different, and the chromaticity difference of the traditional Chinese medicine injection is closely related to the strict degree of the control of technological process parameters, so the chromaticity check of the traditional Chinese medicine injection is also an important physical index for controlling the quality of the traditional Chinese medicine injection.
The two indexes of controlling the chromaticity and the visible foreign matters are beneficial to ensuring the medication consistency and the safety.
In addition, the in vitro activity detection of the traditional Chinese medicine injection is carried out, the activity index of the traditional Chinese medicine injection is often related to the drug effect of the injection, at present, although the traditional Chinese medicine injection is not listed as a necessary inspection item, the detection of the in vitro activity index of the traditional Chinese medicine injection is more and more important in the aspect of guaranteeing the effectiveness of the traditional Chinese medicine along with the stricter and more standard control of the quality of the traditional Chinese medicine injection, and the quality of the traditional Chinese medicine injection evaluated by adopting an activity detection method can more guarantee the effectiveness of the traditional Chinese medicine injection. However, the conventional methods (such as anticoagulant activity determination) also have the defects of slow detection speed, sample damage, waste liquid generation, environmental pollution and the like.
In the actual production of the traditional Chinese medicine injection, the quality and the process parameters of the raw materials need to be strictly controlled, the difference of the raw materials of the traditional Chinese medicine and the fluctuation of the process parameters can cause certain difference of the quality among the batches of the product, and even the product in the same batch can cause slight difference of the quality of the product due to the difference of different filling time, different sterilization positions and the like. However, the quality detection of the traditional Chinese medicine injection at present basically adopts a sampling detection mode, so that the possibility of missed detection exists, and the use safety is influenced.
In short, in the prior art, chemical, activity and physical indexes of a traditional Chinese medicine injection are determined by dividing the traditional Chinese medicine injection into different detection systems, and off-line sampling detection is generally adopted, so that a detected sample is broken and damaged after opening a bottle, which causes the defects of time-consuming detection process, large workload, low efficiency, missed detection risk, numerous instruments, expensive part of detection equipment and the like.
SUMMERY OF THE UTILITY MODEL
To many indexes testing process among the prior art consuming time, work load is big, inefficiency, partial instrument is expensive, selective examination easily leads to lou examining the risk scheduling problem, the utility model provides a detecting system of a plurality of indexes of medicinal injection in survey simultaneously based on hyperspectrum.
A hyperspectral-based multi-index detection system for a traditional Chinese medicine injection comprises a traditional Chinese medicine injection ampoule bottle conveying module, a hyperspectral image acquisition module and a hyperspectral image processing module; the ampoule bottle conveying module conveys the to-be-detected traditional Chinese medicine injection to a hyperspectral image acquisition area, the hyperspectral image acquisition module acquires a hyperspectral image of the injection, and the hyperspectral image processing module performs data processing on the acquired hyperspectral image and analyzes and outputs a chemical component prediction result, an activity prediction result, a chromaticity detection result and a visible foreign matter detection result in the traditional Chinese medicine injection.
Furthermore, the hyperspectral image processing module comprises a chemical/activity index processing unit, a chromaticity index processing unit and a visible foreign matter index processing unit, the chemical/activity index processing unit analyzes to obtain a prediction result of chemical components and activity in the traditional Chinese medicine injection, the chromaticity index processing unit analyzes to obtain a chromaticity detection result, and the visible foreign matter index processing unit analyzes to obtain a visible foreign matter detection result.
Further, the chemical/activity index processing unit establishes a prediction model by using a convolutional neural network (namely CNN network) method, and processes the acquired hyperspectral image by using the CNN network method; the CNN network comprises an input layer, three convolution pooling layers, a full-connection layer and N parallel output modules, wherein each output module comprises two full-connection layers and an output layer; the input layer adopts an original spectrum of an input sample, and the output layer simultaneously outputs N chemical component quantitative results and activity prediction results.
Further, in the chemical/activity index processing unit, selecting a pixel point with the maximum RGB value in the full hyperspectral image, and defining a rectangular area on the bottle body, wherein the RGB value of the pixel point in the rectangular area is more than or equal to 70% of the maximum value, and the rectangular area is different from other areas, represents the bottle body of the traditional Chinese medicine injection and serves as a hyperspectral sampling processing area; and uniformly selecting a plurality of sampling processing points of each sample in a matrix form within the width and height range of the bottle body sampling processing area.
Further, the average spectrum of the pixel block covered by the sampling processing point, the volume of the sample liquid covered by the sampling processing point and the optical path (the length x width x optical path depth of the sampling point region) are calculated, the obtained average spectrum of the pixel block is input as a quantitative correction model, and the value obtained by multiplying the volume by the chemical component content and the activity data of the sample measured by a conventional method is output as the quantitative correction model.
Further, the chromaticity index processing unit calculates an average spectrum of the hyperspectral image in the hyperspectral sampling processing area, selects light intensities of the spectrum at 700nm (red), 546.1nm (green) and 435.8nm (blue), and converts the light intensities into an H value in an HSV color space, namely chromaticity.
Further, the visible foreign matter index processing unit selects a hyperspectral image collected at the bottom of the traditional Chinese medicine injection as an image processing area; the bottle body conveying module collects the conveyed visible foreign matters of the traditional Chinese medicine injection to the bottom edge of the injection bottle body and conveys the visible foreign matters to the hyperspectral image acquisition area, and the visible foreign matter index processing unit extracts the bottom area outline of the acquired hyperspectral image and detects large-scale and small-scale particles precipitated in the injection bottle.
Further, aiming at the outline of the bottom area of the injection bottle body, the visible foreign matter index processing unit extracts the bottom edge of one side containing the visible foreign matter by using a chain code technology to be used as a lower curve; carrying out large-scale detection on the lower curve, detecting points with curvature mutation on the lower curve, connecting adjacent corner points by using line segments to determine possible foreign matter areas, judging whether the foreign matter areas are false alarms or not according to the root-mean-square standard deviation of the spectrums in the foreign matter areas, and judging the false alarms if the root-mean-square standard deviation is less than 0.2; and (3) continuously carrying out small-scale inspection on the lower curve part where the foreign matters cannot be detected by the large-scale detection method, directly inspecting the curvature of the lower curve by the small-scale inspection, and judging that small-particle foreign matters exist in the sample when the curvature of pixel points on the curve is more than 0.3.
Further, the bottle conveying module comprises an inclined vibration conveying platform, an inclined flat conveying platform and a fixing support for enabling the bottles to be vertically conveyed in order, the inclined flat conveying platform and the inclined vibration conveying platform are in an inclined state, the inclination angle is 5-15 degrees, and the injection bottles are conveyed forwards sequentially from the inclined vibration conveying platform and the inclined flat conveying platform.
Furthermore, the hyperspectral image acquisition module comprises a halogen lamp light source, a hyperspectral camera and a light shielding cover, the halogen lamp light source and the hyperspectral camera are respectively arranged at two sides of the bottle body transmission module, wherein the hyperspectral camera acquires all wavelength range combinations within the wavelength range of 400-1700nm, the wavelength resolution is 5nm or less, and the hyperspectral camera scans the traditional Chinese medicine injection bottle body on the inclined flat conveying platform to acquire a hyperspectral image.
Compared with the prior art, the utility model has the advantages of it is following and beneficial effect:
(1) the utility model provides a many indexes detecting system of traditional chinese medicine injection based on hyperspectral, its hyperspectral image collection module only need to go up traditional chinese medicine injection bottle scanning formation hyperspectral image to bottle conveying module, hyperspectral image processing module carries out data processing to the hyperspectral image who gathers, and the spectral information who draws can predict many chemical index and many activity index and survey the colourity; the extracted image information can judge whether the injection has visible foreign matters; after the modeling is accomplished, need not any sample processing, the harmless many indexes of realizing fast detect simultaneously, have greatly simplified test mode and check out test set, labour saving and time saving adopts the utility model provides a system has improved the detection efficiency of traditional chinese medicine injection quality index.
(2) The utility model discloses can be applied to the on-line measuring on the production line, detect all injection products high-flux ground, realize that the product examines entirely, overcome the problem of examining because of the hourglass that the selective examination leads to, guarantee product quality better, effectively promoted the safety, the effectiveness of using medicine of traditional chinese medicine injection, and need not to damage injection sample just can the prediction index result, clean environmental protection.
(3) The utility model discloses a chemistry/activity index processing unit uses convolution neural network (be CNN network) method to establish prediction model, owing to adopted the structure of degree of depth neural network and convolution layer, makes the variable number that needs training reduce greatly, under the prerequisite that reduces the calculated amount and guarantee the model precision, has increased the scalability of model; in addition, CNN model herein adopts the convolutional layer module of a sharing to N index, what draw promptly is public spectral feature, the utility model discloses a CNN model establishes a model can predict N index, only needs to spend 1/N time, need not spectrum preliminary treatment and draws the manual regulation of characteristic wave band and model parameter, is an end-to-end quantitative correction model, and the commonality is strong.
(4) The utility model discloses a visible foreign matter detection method of injection based on high spectrum is different from the detection principle of traditional lamp inspection machine. The visible foreign matters are gathered at the bottom of the bottle, so that the large-scale and small-scale detection is carried out on the bottom of the bottle, and the detection precision is higher; the device adopted by the method is simple in structure and convenient and fast to operate, and the detection cost can be greatly reduced.
(5) The system can be widely applied to the field of quality detection of traditional Chinese medicine injections, corresponding modeling adjustment is only needed to be carried out according to specific chemical components and activity indexes of different preparations, and chromaticity and visible foreign matters are general inspection items of the water injections and can be directly applied.
Drawings
Fig. 1 is a simple flowchart of the detection system provided by the present invention.
Fig. 2 is a schematic diagram of the detection system provided by the present invention.
Fig. 3 is a detection flow chart of the embodiment of the present invention.
Fig. 4 is a diagram of a convolutional neural network structure according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of a sampling unit of a biochemical indicator processing unit according to an embodiment of the present invention.
Fig. 6 is a schematic diagram of a visible foreign object index processing unit according to an embodiment of the present invention.
Fig. 7 is a chromaticity distribution diagram of 7 manufacturers according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clear, the present invention is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the invention.
Referring to fig. 1, the embodiment provides a hyperspectral-based multi-index detection system for a traditional Chinese medicine injection, wherein the traditional Chinese medicine injection is in a cylindrical bottle shape and comprises a traditional Chinese medicine injection ampoule bottle conveying module, a hyperspectral image acquisition module and a hyperspectral image processing module; the ampoule bottle conveying module conveys the to-be-detected traditional Chinese medicine injection to a hyperspectral image acquisition area, the hyperspectral image acquisition module acquires a hyperspectral image of the injection, the hyperspectral image processing module performs data processing on the acquired hyperspectral image, the hyperspectral image processing module comprises a chemistry/activity index processing unit, a chromaticity index processing unit and a visible foreign matter index processing unit, the chemistry/activity index processing unit analyzes and obtains chemical components and activity prediction results in the traditional Chinese medicine injection, the chromaticity index processing unit analyzes and obtains chromaticity detection results, and the visible foreign matter index processing unit analyzes and obtains visible foreign matter detection results.
As different substances have different performances under different wave band spectrum signals, the hyperspectral imaging technology is based on a plurality of narrow wave band image data technologies, combines the imaging technology and the spectrum technology, detects two-dimensional geometric space and spectrum information of a target, and acquires high-resolution continuous and narrow wave band image data. For the utility model discloses implement N indexs of detection and provide the possibility.
Referring to fig. 2, in the system, the bottle body conveying module comprises an inclined vibration conveying platform 4, an inclined flat conveying platform 3 and a fixed support 2 for enabling bottle bodies to be vertically and obliquely conveyed in order, and the inclined angles of the inclined vibration conveying platform 4 and the inclined flat conveying platform 3 are 5-15 degrees, preferably 5 degrees, so that better balance between conveying and foreign particle deposition is achieved. The hyperspectral image acquisition module comprises a halogen lamp light source 1, a hyperspectral camera 7 and a light-shielding cover (not shown in the figure), wherein the hyperspectral camera 7 is a visible-shortwave near-infrared hyperspectral camera/near-infrared hyperspectral camera, the hyperspectral camera 7 acquires all wavelength range combinations in a wavelength range of 400 plus 1700nm, the wavelength resolution is 5nm or less, and the hyperspectral camera 7 scans the traditional Chinese medicine injection bottle body on the inclined flat conveying platform 3 to acquire a hyperspectral image; the hyperspectral image processing module is a workstation 6.
The halogen lamp light source 1 is a light intensity-adjustable area light source, the fixing support 2 is used for preventing an injection 5 from toppling, a traditional Chinese medicine injection bottle body is conveyed to pass through the hyperspectral camera lens forwards through the inclined vibration conveying platform 4 and the inclined flat conveying platform 3 in sequence, the vibration conveying platform 4 and the inclined flat conveying platform 3 are used for gathering visible foreign matters on the bottom edge of the injection, the hyperspectral VNIR camera 7 acquires a hyperspectral image of the injection, and the workstation 6 is used for performing subsequent data processing on the acquired image. The hyperspectral image acquisition module is arranged in a light avoiding cover or a dark box to operate, so that the interference of ambient light is prevented.
The utility model discloses in, take shuxuening injection as an example, shuxuening injection is the sterilization aqueous solution that ginkgo leaf was made through processing, has the dilatation blood vessel, improves microcirculation effect, and clinical mainly used ischemia cardiovascular and cerebrovascular disease, coronary heart disease, angina pectoris cerebral embolism, cerebral vasospasm. Is one of the large varieties of Chinese medicine injections. The quality standard stipulates quantitative detection of the contents of two major chemical components of total flavonol glycosides and ginkgolides, so 2 quantitative indexes are selected as chemical indexes; in addition, from the aspect of the indication, the anticoagulant activity and the antioxidant activity index of the medicine are closely related to the medicine effect, so that the 2 indexes are selected as the activity indexes. Meanwhile, the chromaticity and visible foreign matters are selected as physical detection indexes, and 6 quality indexes are selected in total.
At present, the chemical and activity indexes of injection products are detected at home and abroad mainly by adopting a sampling mode to perform off-line detection respectively. As shown in FIG. 3, the total flavonol glycoside content was measured by HPLC-UV method, the bilobalide content was measured by HPLC-ELSD method, the anticoagulation activity (thrombin activity inhibition) was measured by fluorometry, and the antioxidative activity (DPPH radical clearance) was measured by visible light photometry. Although these methods are stable and reliable, they have the disadvantages of slow detection speed, sample destruction, environmental pollution caused by generated chemical and biological reagents, and the like. In the prior art, chemical, activity and physical indexes of the traditional Chinese medicine injection are determined by dividing the traditional Chinese medicine injection into different detection systems, so that detection equipment is complex, the detection process is complicated and time-consuming, the workload is large, and the efficiency is low. The prior art has a series of reports of applying near infrared spectrum and ultraviolet spectrum to the detection of chemical components in liquid, but the traditional spectroscopy has two main defects: (1) the traditional spectrum technology only acquires spectrum information but does not have image information, so that the problem of spatial heterogeneity of the cylindrical ampoule bottle cannot be solved; (2) spatial information is needed for detecting visible foreign matters, and the spectroscopy cannot be realized. In the prior art, for the detection of visible foreign matters, expensive lamp detectors are mostly used, so that the overall detection cost is very high.
When the Shuxuening injection is applied to the system, the hyperspectral image is formed by scanning the traditional Chinese medicine injection bottle on the bottle conveying module, the hyperspectral image processing module performs data processing on the acquired hyperspectral image, a chemical/activity index processing unit of the hyperspectral image processing module analyzes and obtains a chemical index prediction result and an activity index prediction result of the traditional Chinese medicine injection, a chromaticity index processing unit analyzes and obtains a chromaticity index detection result, and a visible foreign matter index processing unit analyzes and obtains a visible foreign matter index detection result.
In this embodiment, the hyperspectral camera 7 is a visible-shortwave near-infrared hyperspectral camera, the collected wavelength range includes all wavelength range combinations within 400-1000nm, and the wavelength resolution is 1.28 nm.
The detection of the Shuxuening injection comprises the processes of sampling, correcting, data processing and the like when in specific use, and the specific process is as follows:
1. simultaneous quantitative determination of chemical/activity indicators
1) Image acquisition: hyperspectral images of a total of 60 samples from 7 companies were collected, including 11 colorless 2ml bottles, 37 colorless 5ml bottles and 12 brown 5ml ampoules. Because the technology adopts a mode of collecting transmission spectrum, and the cylindrical liquid has a light condensation effect, the hyperspectral image of the purified water sample filled in the ampoule bottle with the same specification and the same volume is collected as a background for white board correction of the subsequent sample. In order to eliminate errors caused by different colors of bottles, images of ampoules containing water with three different specifications are respectively used as correction images.
2) And (3) image correction: in this embodiment, the spectra of 60 sample sampling points are directly divided by the spectra of 3 corrected image sampling points, and the obtained result is used as the corrected spectrum of each sample.
3) Image processing: for the 63 hyperspectral images, firstly, selecting pixel points with the maximum RGB value in the hyperspectral images, and defining a rectangular area on the bottle body, wherein the RGB value of the pixel points in the rectangular area is more than or equal to 70% of the maximum value, and the rectangular area is different from other areas, represents the bottle body of the traditional Chinese medicine injection and serves as a hyperspectral sampling processing area; present hyperspectral detection technique is generally to set up whole sample image region for image processing region, is used for quantitative correction to model after calculating average spectrum, and because the characteristic of ampoule cylinder, the optical path of different positions is different in the sample region, and above method can cause great error, so the utility model discloses evenly gather 3 x 5 little pixel blocks in each selected sampling process region, be applied to 63 highlight blocks that chemistry/activity detectedThe spectral image, based on the width and height of the selected sample processing region, uniformly acquires 15 sample processing points (in a 3 x 5 matrix). The concrete form is as follows: as shown in FIG. 5, let the size of the hyperspectral data be LW×LH×LBWherein L isW×LHLength and width in spatial dimension, and LBIs the number of bands. Five vertical lines are set in the rectangular area of the bottle body:
setting three transverse lines simultaneously
And the intersection point of the horizontal and vertical lines is a sampling processing point. And calculating the average spectrum of the pixel block covered by the sampling processing point, the volume of the sample liquid covered by the sampling processing point and the optical path (the length of the sampling processing point region is multiplied by the width and multiplied by the optical path depth). And randomly dividing the average spectrum of all sampling processing points and corresponding chemical/activity indexes into a correction set and a verification set according to a ratio of 3:1, inputting the average spectrum of a pixel block of the correction set as a quantitative correction model, and outputting a value obtained by multiplying the corresponding volume by the chemical component content and the activity data of the sample measured by a conventional method as the quantitative correction model. The chemical/activity index processing unit establishes a prediction model by using a Convolutional Neural Network (CNN) method, as shown in FIG. 4, the CNN network comprises an input layer, three convolutional pooling layers, a full-link layer and four parallel output modules, wherein each module comprises two full-link layers and one output layer; the input layer inputs the original spectrum of the sample without pretreatment, and the output layer simultaneously outputs to obtain four quantitative results: two chemical component contents (total flavonol glycosides and bilobalide), and two activity indexes (antioxidant activity and anticoagulant activity).
And testing the verification set on a prediction model obtained after training is completed to obtain a result which is better than that of a traditional algorithm-partial least square regression method. The results of model evaluation and comparison with the conventional method are shown in the following table:
the utility model can effectively output the detection results of 4 quality indexes and passes through the decision coefficient (R)2The closer to 1, the better), the root mean square error (RMSE, the smaller the better) and the residual prediction deviation (RPD, the larger the better) to evaluate the prediction performance of the established model, and compared with the traditional partial least squares method, each quality index obtains the better model performance on a correction set and a verification set.
The conventional CNN network model models 4 indexes, and a manner of establishing 4 individual models is adopted. And the CNN model of the utility model can predict 4 indexes by establishing a model, and only 1/4 time is needed. In addition, the CNN model of the present invention uses a common convolution layer for 4 indices, i.e., common spectral features are extracted. According to the weight coefficients of the convolutional layer and the input layer, characteristic variables in the spectrum can be reversely deduced, the interpretability of CNN is increased, meanwhile, a hyperspectral system can be converted into a multispectral system, and the detection speed can be greatly improved.
The utility model provides a CNN network model is revised as parallelly connected a plurality of with the output module on a full tie layer, reaches the target of many outputs.
In order to simultaneously consider the prediction accuracy of the 4 indexes, the sum of Mean Square Error (MSE) of the 4 indexes is used as model loss (loss) for training, and therefore the prediction effect of the model on the 4 indexes is comprehensively evaluated. In order to eliminate the bias of gradient descending direction during model training caused by dimension difference, normalization processing needs to be carried out on output data before model training, and anti-normalization processing needs to be carried out when the output data is used for prediction.
For the problem of pre-treatment: the traditional spectral quantitative correction modeling method generally comprises three steps of spectral preprocessing, characteristic wave band extraction and quantitative modeling. The existing spectrum pretreatment methods include: a smoothing method; b derivative method; c, standard normal variable transformation; d correcting multiple scattering. The more common method is to extract the characteristic wave band by adopting a supervised or unsupervised mode after the spectrum is preprocessed. And after the methods of preprocessing and extracting the characteristic wave bands are fully combined, screening according to the modeling effect. This way of traversing the combination of algorithms results in an unnecessary computational burden. Moreover, the preprocessing method is different with different environmental factors (temperature, humidity and ambient light) and subjective factors of operators, and the popularization performance of the model is limited. Erroneous use of the preprocessing mode may also distort the spectral signal, causing the model accuracy to drop. Although the algorithm for extracting the characteristic bands selects the key bands in the spectrum, the algorithm may cause loss of effective information. Neural Network (NN) can theoretically achieve the purpose of completing modeling without preprocessing and extracting characteristic bands, but because parameter variables in a Neural Network are too many, the training speed is slow, and meanwhile, the risk of overfitting is caused.
The CNN technology does not need to preprocess and extract characteristic wave bands, and simultaneously, due to the adoption of the structure of the deep neural network and the convolution layer, the number of variables is greatly reduced, and the expandability of the model is improved on the premise of reducing the calculated amount and ensuring the precision of the model. The convolutional neural network model comprises a convolutional layer and a fully-connected layer, wherein the convolutional layer is essentially a characteristic extraction process and can extract characteristic information in a spectrum and filter part of irrelevant information, so that a relatively ideal modeling effect can be obtained under the condition of no spectrum preprocessing.
2. Detection of colorimetric indicators
1) Image acquisition: the hyperspectral images of the body areas of 63 ampoules which were subjected to "simultaneous quantitative determination of chemical/activity indices" were used.
2) Image processing: for the 63 hyperspectral images, firstly, selecting pixel points with the maximum RGB value in the hyperspectral images, and defining a rectangular area on the bottle body, wherein the RGB value of the pixel points in the rectangular area is more than or equal to 70% of the maximum value, and the rectangular area is different from other areas, represents the bottle body of the traditional Chinese medicine injection and serves as a hyperspectral sampling processing area; the chromaticity index processing unit calculates the average spectrum of the hyperspectral images in the hyperspectral sampling processing area, the light intensities of the spectrum at 700nm (red), 546.1nm (green) and 435.8nm (blue) are selected from the hyperspectral images, and the chromaticity distribution of 7 manufacturers is shown in figure 7.
The method specifically comprises the following steps: the acquired light intensities of the three bands are set as R ', G ' and B '. The formula for calculating the chromaticity is as follows:
Δ＝max(R′,G′,B′)-min(R′,G′,B′)
3. detection of visible foreign body indicators
1) Image acquisition: hyperspectral images of 100 bottles of samples (32 bottles were judged to be abnormal when shipped) were collected in total. In addition, in order to examine whether the method is applicable to detection of fine particles of several tens of micrometers, a hyperspectral image of a 60 μm standard particle was acquired. For the hyperspectral images of the 101 ampoule bottles, all pixel points of the hyperspectral images are traversed by small windows of 3 x 3 pixel points, except for one circle of pixel points at the outermost edge of the images. And solving the variance of the spectrum of the pixel points in each small window, and taking the variance value as the pixel value of a new image so as to obtain a single-layer gray-scale image. And (4) obtaining a bottom mask of the gray-scale image by adopting a watershed algorithm, so as to obtain a region of interest at the bottom.
2) Image processing: and extracting the outline of the bottom region, and extracting a half of the bottom edge close to the lens by using a chain code technology, wherein the bottom edge is called as a lower curve. Firstly, a large-scale detection method is adopted for a lower curve, possible corner points (points with abrupt curvature) are detected on the lower curve, then adjacent corner points are connected by line segments, and a possible foreign matter area is determined as shown in fig. 6 (1). And judging whether the foreign matter area is false alarm or not according to the standard deviation of the root mean square of the spectrum in the foreign matter area, wherein if the standard deviation of the root mean square is less than 0.2, the foreign matter area is false alarm. These false alarms may be caused by dust or spectral noise spots attached to the outer wall.
The root mean square standard deviation calculation formula is as follows:
wherein n isiAnd RSD (i) is the difference between the number of pixel points in the ith possible impurity region and the root mean square, PIj，kIs the light intensity of the kth wave band of the jth pixel point.
And (3) continuing small-scale inspection on the lower curve part where the foreign matters cannot be detected by the large-scale detection method, directly inspecting the curvature of the curve by the small-scale inspection, and determining that small-particle foreign matters exist in the sample if the curvature of pixel points on the curve is larger than 0.3 as shown in (2) of fig. 6. The method can detect visible foreign matters with particle size as low as 60 μm under static condition.
The number of samples used for visual foreign matter examination was 100, and 32 of them were abnormal samples. As shown in fig. 6(3), L represents large-scale particles, and S represents small-scale particles. The large-scale inspection and the small-scale inspection of the method are sequentially operated, wherein 33 bottles are detected in a large scale, 12 bottles are detected in a small scale, and 35 bottles are detected in total by integrating the results of the large scale and the small scale. And 32 bottles are detected by the light inspection machine, so that all abnormal samples can be detected by adopting the method, but 3 bottles of false alarms exist. For a part of samples with abnormal detection in large scale, the samples detect additional impurity particles, and the small scale detection can effectively avoid the condition of missing detection. In addition, the small-scale inspection method adopted by the method can detect 60 μm particles, which is close to the detection standard of pharmacopoeia that can not detect 50 μm particles.
Unlike "light inspection machine" detection: the current comparatively ripe lamp inspection machine is through the image of high-speed CCD camera collection liquid medicine upset in-process, judge whether there is visible particle through the change condition of analysis pixel point grey level in the image, its mechanical structure is complicated and to the shooting speed and the imaging accuracy requirement of CCD camera higher, solid high price, and this system need not overturn the operation to the sample, mechanical structure complexity has been reduced, and owing to combined spectral information, but solid reduction to a certain extent is to the requirement of pixel resolution ratio, and then reduce cost.
In summary, when the multi-index detection system of the traditional Chinese medicine injection is applied to the shuxuening injection, as shown in fig. 3, after the shuxuening injection is scanned to obtain a hyperspectral image, the body region is used for detecting the chromaticity index, the bottom region is used for detecting the visible foreign object index according to the discrimination of the spectral variance and the edge curvature, a quantitative correction model is made on the pixel block of the matrix sampling point in the body region based on the convolutional neural network, and the chemical and activity index results are predicted.
The utility model provides a many indexes of traditional chinese medicine injection detecting system based on hyperspectrum, its hyperspectral image collection module only need scan the traditional chinese medicine injection bottle on bottle conveying module and form the hyperspectral image, and hyperspectral image processing module carries out data processing to the hyperspectral image who gathers, draws spectral information just can predict many chemical index and many activity index and acquire the colourity testing result; the image information is extracted to judge whether the injection has visible foreign matters; after the modeling is accomplished, need not any preliminary treatment, just can realize that many indexes detect simultaneously, greatly simplified detection mode and check out test set, labour saving and time saving adopts the utility model provides a system has improved the detection efficiency of traditional chinese medicine injection.
The utility model discloses a many indexes detecting system can be applied to the on-line measuring on the production line, detects all injection products high-throughput, realizes that the product examines entirely, has overcome because of the leak testing problem that the selective examination leads to, and product quality more can obtain the guarantee, has effectively promoted the security when the medicine uses, and need not to damage the injection sample and just can measure the index result, and is clean environmental protection.
In addition, the system can also be applied to other liquid preparations, corresponding adjustment is carried out according to specific chemical components and activity indexes of different preparations, and chromaticity and visible foreign matters are general inspection items of the water injection, so that the system can be directly applied and is suitable for popularization.
The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be equivalent replacement modes, and all are included in the scope of the present invention.
Claims (1)
1. A hyperspectral-based multi-index detection system for a traditional Chinese medicine injection is characterized by comprising a traditional Chinese medicine injection ampoule bottle conveying module, a hyperspectral image acquisition module and a hyperspectral image processing module;
the ampoule bottle conveying module conveys the Chinese medicine injection to be detected to a hyperspectral image acquisition area, and comprises an inclined vibration conveying platform (4), an inclined flat conveying platform (3) and a fixed support (2) for enabling the bottles to be vertically and orderly conveyed, wherein the table surfaces of the inclined flat conveying platform (3) and the inclined vibration conveying platform (4) are in an inclined state, the inclination angle is 5-15 degrees, and the injection bottles are conveyed forwards from the inclined vibration conveying platform (4) and the inclined flat conveying platform (3) in sequence;
the hyperspectral image acquisition module acquires a hyperspectral image of an injection, the hyperspectral image acquisition module comprises a halogen lamp light source (1), a hyperspectral camera (7) and a light shielding cover, the halogen lamp light source (1) and the hyperspectral camera (7) are respectively arranged on two sides of the ampoule bottle conveying module, wherein the hyperspectral camera (7) acquires all combinations of wavelength ranges within 400 plus 1700nm, the wavelength resolution is 5nm or less, and the hyperspectral camera (7) scans a traditional Chinese medicine injection bottle body on the inclined flat conveying platform (3) to acquire the hyperspectral image;
and the hyperspectral image processing module is used for carrying out data processing on the acquired hyperspectral image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020138594.3U CN213779856U (en) | 2020-01-21 | 2020-01-21 | Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020138594.3U CN213779856U (en) | 2020-01-21 | 2020-01-21 | Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN213779856U true CN213779856U (en) | 2021-07-23 |
Family
ID=76865468
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202020138594.3U Active CN213779856U (en) | 2020-01-21 | 2020-01-21 | Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN213779856U (en) |
-
2020
- 2020-01-21 CN CN202020138594.3U patent/CN213779856U/en active Active
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111257228A (en) | Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection | |
| Chen et al. | Machine vision technology for agricultural applications | |
| Lu et al. | Hyperspectral imaging for safety inspection of food and agricultural products | |
| Delwiche et al. | Classification of wheat by visible and near-infrared reflectance from single kernels | |
| Teerachaichayut et al. | Non-destructive prediction of total soluble solids, titratable acidity and maturity index of limes by near infrared hyperspectral imaging | |
| Qin et al. | Raman chemical imaging system for food safety and quality inspection | |
| Gowen et al. | Recent applications of chemical imaging to pharmaceutical process monitoring and quality control | |
| ElMasry et al. | Chemical-free assessment and mapping of major constituents in beef using hyperspectral imaging | |
| Ariana et al. | Evaluation of internal defect and surface color of whole pickles using hyperspectral imaging | |
| Serranti et al. | The development of a hyperspectral imaging method for the detection of Fusarium-damaged, yellow berry and vitreous Italian durum wheat kernels | |
| CN108267414B (en) | Near infrared spectrum analysis method for textile fiber content | |
| Chao et al. | Use of hyper–and multi–spectral imaging for detection of chicken skin tumors | |
| EP1362233A2 (en) | System and method for grouping reflectance data | |
| CN107976428A (en) | The lossless real-time detection method and device of aquatic products chemical pollutant | |
| Park et al. | Multi-spectral image analysis using neural network algorithm for inspection of poultry carcasses | |
| Zhong et al. | An online, non-destructive method for simultaneously detecting chemical, biological, and physical properties of herbal injections using hyperspectral imaging with artificial intelligence | |
| Soares et al. | Classification of individual cotton seeds with respect to variety using near-infrared hyperspectral imaging | |
| EP0965393B1 (en) | Method for discrimination of product units and apparatus therefor | |
| CN113008817A (en) | Method for rapidly identifying authenticity and quality of bitter apricot kernels based on hyperspectral imaging technology | |
| KR20220038602A (en) | Methods and systems for identifying microorganisms | |
| Park et al. | Detection of cecal contaminants in visceral cavity of broiler carcasses using hyperspectral imaging | |
| Guo et al. | Detection of foreign materials on surface of ginned cotton by hyper-spectral imaging | |
| CN213779856U (en) | Hyperspectrum-based multi-index detection system for traditional Chinese medicine injection | |
| Zheng et al. | Detection of Huanglongbing (citrus greening) based on hyperspectral image analysis and PCR | |
| JP2009168746A (en) | Inspection method and inspection device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant |