CN114414551B - Fish identification method based on LIBS-RAMAN spectrum technology - Google Patents
Fish identification method based on LIBS-RAMAN spectrum technology Download PDFInfo
- Publication number
- CN114414551B CN114414551B CN202210070023.4A CN202210070023A CN114414551B CN 114414551 B CN114414551 B CN 114414551B CN 202210070023 A CN202210070023 A CN 202210070023A CN 114414551 B CN114414551 B CN 114414551B
- Authority
- CN
- China
- Prior art keywords
- libs
- raman
- fish
- spectrum
- data
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/71—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light thermally excited
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a fish meat identification method based on LIBS-RAMAN spectrum technology, which comprises the steps of firstly grinding fish meat to be identified into minced meat by a meat grinder to obtain a fish meat sample with a flat surface; then respectively acquiring LIBS spectrum data and RAMAN spectrum data of a fish sample to be identified, carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1; and finally, merging the normalized RAMAN and LIBS spectrum data to obtain RAMAN-LIBS merged picture data, and inputting the RAMAN-LIBS merged picture data into a trained CNN model to obtain a fish identification result. The method combines the RAMAN spectrum technology and the LIBS spectrum technology, and utilizes the complementarity of the spectrum information of the RAMAN spectrum technology and the LIBS spectrum technology to obviously improve the accuracy of fish identification, so as to realize effective identification of different fish, close fish and even different parts of fish, and provide a practical and feasible scheme for fish identification.
Description
Technical field:
the invention belongs to the technical field of aquatic product identification, and particularly relates to a fish meat identification method based on LIBS-RAMAN spectrum technology.
The background technology is as follows:
with the development of human society and the improvement of economic level, the demand of human beings on aquatic products is continuously increased, and food doping and adulteration become one of main problems of food quality and safety. In recent years, as the processing scale of meat value-added products is gradually increased, the phenomenon that meat products are "sub-full and pseudo-spurious" is frequently used, such as replacing high-price meat with low-price meat and replacing chilled meat with frozen-thawed meat. Among them, the aquatic products are of various kinds and large consumption, and the species quality and price difference between the near relationship are very different, so that the phenomena of aquatic product adulteration and mislabeling are endangered and the benefits and even health of consumers are damaged. The market supervision department is urgently required to be a quick and accurate technical means for distinguishing the authenticity of meat and striking counterfeit goods.
RAMAN spectroscopy (RAMAN) technology has been widely used as a novel spectroscopic detection technique in mass physicochemical structural analysis. The technology can realize rapid and nondestructive detection of meat products, is one of the technology of meat component analysis, and has extremely insensitive spectrum to polar substances such as water, thus having good application prospect in meat research. The Laser Induced Breakdown Spectroscopy (LIBS) technology is a spectroscopic technology for qualitatively and quantitatively analyzing elements, and plasma spectrums of a sample to be measured are obtained by performing laser ablation on the surface of the sample, so that qualitative or quantitative analysis of the sample to be measured is realized. LIBS technology has been widely used in the fields of mineral analysis, environmental pollution monitoring, metallurgical analysis, biopharmaceutical, space exploration, etc. With the development of LIBS technology, the LIBS technology is gradually an emerging material detection and identification analysis technology in the food industry in recent years. In terms of species identification of meat, bilge et al first identified pork, beef and chicken using LIBS technology, creating a method for effectively identifying meat species. The LIBS technology is combined with the random forest method to identify and classify fish products in the earlier stage of the subject group.
LIBS and RAMAN spectrum technology have good complementarity on the detection target: the LIBS spectrum technology can acquire element information of a sample to be detected, and the RAMAN spectrum technology can acquire molecular information of the sample to be detected. By combining the two spectrum technologies, more comprehensive substance components and structural information of the sample to be detected can be obtained, so that the sample to be detected can be identified more favorably. At present, research on fish identification by singly adopting LIBS and RAMAN spectrum technologies is reported, but research on fish identification by combining the LIBS and the RAMAN spectrum technologies is not reported yet.
The invention comprises the following steps:
the invention aims to establish a fish meat identification method based on LIBS-RAMAN spectrum technology.
In order to achieve the above purpose, the fish meat identification method based on LIBS-RAMAN spectrum technology comprises the following steps:
(1) Grinding the fish meat to be identified into minced meat by a meat grinder, filling the minced meat into a small-aperture culture dish with the diameter of 35mm, and finally compacting and refrigerating by using glass sheets to obtain a fish meat sample with a flat surface;
(2) Respectively acquiring LIBS spectrum data and RAMAN spectrum data of a fish sample to be identified, carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1;
(3) Simultaneously converting the normalized RAMAN spectrum data and LIBS spectrum data into RAMAN-LIBS fusion picture data; inputting the RAMAN-LIBS fusion picture data into a trained CNN model to obtain a fish identification result;
specifically, the training process of the CNN model is as follows:
(301) The method comprises the steps of adopting a meat emulsion method to process a sample, grinding fish meat into meat emulsion by a meat grinder, filling the meat emulsion into a small-aperture culture dish, and finally compacting and refrigerating by using glass sheets, wherein the sample comprises fish meat of different types and different parts of the fish meat;
(302) Respectively acquiring LIBS spectrum data and RAMAN spectrum data of fish meat to be identified, carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1;
(303) Simultaneously converting the normalized RAMAN spectrum data and LIBS spectrum data into RAMAN-LIBS fusion picture data;
(304) Dividing the RAMAN-LIBS fusion picture data into 3 parts according to a proper proportion, wherein the 3 parts are respectively a training set, a verification set and a test set;
(305) The training set data is used for network training, the training set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminary verification of the quality of the model, if the model is qualified through verification, the model is exported to stop training, and then the test set is used for making a final evaluation standard.
The CNN model comprises a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a third convolution layer, a third pooling layer, a first full-connection layer and a second full-connection layer which are sequentially connected.
Compared with the prior art, the invention has the following beneficial effects: (1) Combining the RAMAN spectrum technology and the LIBS spectrum technology, and obviously improving the accuracy of fish identification by utilizing the complementarity of the spectrum information of the RAMAN spectrum technology and the LIBS spectrum technology; (2) The CNN model is adopted to realize effective identification of different fish meat, close-range fish meat and even different parts of fish meat, and a practical and feasible scheme is provided for fish meat identification.
Description of the drawings:
FIG. 1 is a spectrum of salmon, danish.
FIG. 2 is a LIBS spectrum of salmon in Danish
FIG. 3 is a chart showing the fusion spectrum of the salmon RAMAN-LIBS in Danish.
Fig. 4 is a flow chart for classification and identification of CNN models.
Fig. 5 is a specific structural block diagram of the CNN model.
The specific embodiment is as follows:
the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments, and all other embodiments obtained by those skilled in the art without making creative efforts based on the embodiments of the present invention are included in the protection scope of the present invention.
Example 1:
the embodiment comprises the following three parts: sample preparation, spectrum experiment and classification and identification of the fully connected neural network.
1. Sample preparation:
in experiments, 13 different kinds of fish meat, namely white tuna, balk fish, plaice fish, hairtail, red tuna, longli fish, silver cod fish, rainbow trout, salmon, snapper, chilean salmon, norway salmon and Danish salmon, and 3 kinds of fish are selected, wherein the three kinds of fish are respectively different parts, namely the Chilean salmon (middle section and tail), the Norway salmon (middle section and tail) and the Danish salmon (middle section and tail), and total 16 samples are taken. Rainbow trout and salmon are common salmon counterfeiters, and balsa fish often camouflaged to a more expensive longli fish in the market. The possibility of meat identification is studied through the research of common adulterated fishes such as rainbow trout, salmon, balsa fish and the like. The possibility of origin and site identification was investigated by sampling salmon (middle, tail), norway salmon, danish (middle, tail).
The sample is processed by adopting a meat emulsion method, fish meat is ground into meat emulsion by a meat grinder, the meat emulsion is filled into a small-aperture culture dish, and finally, a glass sheet is used for compaction and refrigeration. For 16 samples, 6 samples were prepared for each class, totaling 96 experimental samples.
The preparation method can ensure the uniformity of the sample, and can effectively avoid the influence caused by uneven material distribution on the salmon sample containing fatty streaks. Meanwhile, the sample prepared by the meat emulsion method has good surface flatness and hardness after being compressed and frozen. LIBS spectrum has higher requirements on the flatness and hardness of the sample surface to ensure the quality of the acquired signals, and confocal micro-Raman spectrometer adopted in RAMAN spectrum experiment utilizes the frequency domain filtering effect of the focal plane to reduce the fluorescence intensity of the signals, so that LIBS and RAMAN spectrum technologies have higher requirements on the flatness of the sample surface.
2. Spectral experiments
2.1 Raman spectroscopy experiments
The experiment adopts a WITec confocal micro Raman spectrometer system to carry out RAMAN spectrum detection on fish samples. A 532nm diode laser built into the system was used to excite the raman spectrum. The experimental parameters were set as follows: the integration time was 3s, averaged 2 times, and the grating was set at 600g/mm with a power of 8.0mW. The laser is incident to the beam splitter through the single-mode fiber, reflected by the beam splitter and enters the micro-objective lens, is focused by the Zeiss 10/×0.25 micro-objective lens and is incident to the surface of the sample so as to excite a Raman signal, the signal and the incident laser are reflected by the sample and then are received by the multimode fiber after passing through the 532nm notch filter, and the signal received by the multimode fiber is detected and analyzed through the spectrometer so as to finally obtain the Raman spectrum signal. The Raman frequency shift range of the experimental instrument reaches 40-3800cm -1 . The experiment performed raman spectrum acquisition on 96 samples, resulting in 9600 raman spectrum data. FIG. 1 is a typical salmon RAMAN spectrum of Danish.
2.2 LIBS Spectrum experiment
The LIBS spectrum system built by a laboratory is adopted, the position of a sample platform is monitored by a cross laser and a CMOS camera, the movement of the sample is controlled by adjusting a three-position electric displacement platform, the level of the platform is adjusted by adopting each half-adjustment method, the light path is optimized by adjusting the positions of a focusing light spot and a light probe, a plasma image is shot by a CCD camera after plasma is generated, and spectrum signals are acquired by a spectrum acquisition system, so that LIBS spectrum data are obtained.
In the experiment, the laser adopts a laser Dava-200 Nd:YAG Q-switched pulse laser, the laser wavelength is 1064nm, the pulse width is 10ns, the laser energy is 75mJ, and the frequency is 10Hz. The spectral detection delay is 1.28 mu s and the gate width is 1.05ms. The spectrometer adopts an AVANTES four-channel spectrometer, the model is AvaSpec-2048-USB2, and the wavelength range of the channel 1 is: 195-370nm, channel 2 wavelength range: 360-513nm, channel 3 wavelength range: 502-632nm, channel 4 wavelength range: 620-837nm. Square areas with sides of 7.2mm are selected on the surface of the sample, one sampling point is punched 10 times, and 100 spectrums are obtained for each sample. LIBS spectrum acquisition is carried out on 96 samples, and 9600 LIBS spectrum data are obtained. FIG. 2 is a chart of LIBS spectra of typical salmon danish.
3. Full-connection neural network classification identification
3.1 classifying and identifying the RAMAN spectrum data by adopting the fully connected neural network
(1) And normalizing the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1. Then according to 0.7:0.15: a ratio of 0.15, the ram an spectral data was randomly divided into 3 parts, training set, validation set and test set, respectively. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(2) And performing dimension reduction treatment on the RAMAN spectrum data by adopting a PCA (Principal Component Analysis) method. PCA, principal component analysis, is a commonly used data dimension reduction algorithm. The main idea of PCA is to map n-dimensional features onto k-dimensions, which are completely new orthogonal features, also called principal components, and are k-dimensional features reconstructed on the basis of the original n-dimensional features. The RAMAN spectrum data after PCA dimension reduction is input into a fully connected neural network, and the method specifically comprises the following steps: the training set data is used for network training, the verification set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily checking the model, finally the model is exported, and the data of the test set is used as the final evaluation standard of the model, so that the classification accuracy of the model is 51%.
3.2 classifying and identifying LIBS spectral data by adopting fully-connected neural network
(1) And normalizing the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1. Then according to 0.7:0.15: the LIBS spectral data was randomly divided into 3 parts, training set, validation set and test set, at a ratio of 0.15. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(2) And performing dimension reduction treatment on LIBS spectrum data by adopting a PCA method. The LIBS spectrum data after the dimension reduction is input to a fully-connected neural network, and the method specifically comprises the following steps: the training set data is used for network training, the verification set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily checking the model, finally the model is exported, and the data of the test set is used as the final evaluation standard of the model, so that the classification accuracy of the model is 47%.
3.3. Classifying and identifying fusion spectrum data of RAMAN and LIBS by adopting fully-connected neural network
In view of the molecular information of substances reflected by the RAMAN spectrum, the element information of the substances is reflected by the LIBS spectrum, and the RAMAN spectrum data and the LIBS spectrum data of the same fish are spliced and fused end to end, specifically:
(1) And carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1.
(2) And splicing the normalized RAMAN spectrum data and LIBS spectrum data of the same fish end to obtain RAMAN-LIBS fusion spectrum data. 96 samples, 9600 total, of RAMAN-LIBS fusion data.
(3) According to 0.7:0.15: the ratio of 0.15, random divided the RAM-LIBS fusion data into 3 parts, training set, validation set and test set, respectively. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(4) The training set data is used for network training, the verification set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily checking the model, finally the model is exported, and the data of the test set is used as the final evaluation standard of the model, so that the classification accuracy of the model is 53%. The result shows that based on the fully connected neural network method, the classification accuracy of LIBS or RAMAN data and RAMAN-LIBS fusion data adopted independently is about 50%, and the effective distinction of different kinds of fish is difficult.
Example 2
In view of the fact that the full-connection neural network method cannot classify the RAMAN and LIBS spectrum data to achieve the purpose of fish identification, the Convolutional Neural Network (CNN) method is adopted to classify the RAMAN and LIBS spectrum data, and the specific process is as follows:
1. classifying and identifying RAMAN spectrum data by CNN
(1) And normalizing the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1. Then according to 0.7:0.15: a ratio of 0.15, the ram an spectral data was randomly divided into 3 parts, training set, validation set and test set, respectively. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(2) And converting the normalized RAMAN spectrum data into a RAMAN spectrum picture through a data processing function.
(3) The RAMAN spectrum picture is input into a CNN model, specifically: the training set is used for network training, the model parameters are corrected by the verification set, the training set is used for training repeatedly for a plurality of times to obtain a trained model, then the verification set is used for preliminary verification of the quality of the model, if the model is qualified, the model is exported to stop training, and then the test set is used as the final evaluation standard. And if the verification set is unqualified, repeating training by using the training set. And finally, the test set is used for model test, and the classification accuracy of the CNN model is 91%.
2. Classifying and identifying LIBS spectrum data by CNN
(1) And normalizing the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1. Then according to 0.7:0.15: the LIBS spectral data was randomly divided into 3 parts, training set, validation set and test set, at a ratio of 0.15. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(2) And converting the normalized LIBS spectrum data into LIBS spectrum pictures through a data processing function.
(3) LIBS spectrum picture data are input into a CNN model, and specifically: the training set data is used for network training, the training set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily verifying whether the model is good or bad, if the model is qualified, the model is exported to stop training, and then the test set is used for making a final assessment standard. And if the verification set is unqualified, repeating training by using the training set. Finally, the test set is used for model test, and the classification accuracy of the obtained model is 83%.
3. Classifying and identifying RAMAN spectrum and LIBS spectrum fusion data by CNN
(1) And carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1.
(2) And simultaneously converting the normalized RAMAN data and LIBS data into RAMAN-LIBS fusion picture data, as shown in figure 3. 96 samples, 9600 groups in total, resulted in 9600 groups of frame-LIBS fusion picture data.
(3) According to 0.7:0.15: the ratio of 0.15, random divided the man-LIBS fusion picture into 3, training set, validation set and test set, respectively. The training set is used for model training, the verification set is used for correcting model parameters, and the test set is used for detecting generalization performance of the model.
(4) The RAMAN-LIBS fusion picture data is input into a CNN model, and specifically comprises the following steps: the training set data is used for network training, the training set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily verifying whether the model is good or bad, if the model is qualified, the model is exported to stop training, and then the test set is used for making a final assessment standard. And if the verification set is unqualified, repeating training by using the training set. Finally, the test set is used for model test, and the classification accuracy of the obtained CNN model is 99%. It can be seen that the classification accuracy of the CNN model is remarkably improved after the RAMAN and LIBS spectrum data are fused.
Specifically, the classification and identification flowchart of the CNN model and the specific structural block diagram of the CNN model adopted in embodiment 2 are shown in fig. 4 and fig. 5, respectively. The CNN model comprises a 3-layer convolution layer (Conv 2 d), wherein the convolution layer is used for carrying out convolution operation to obtain local characteristics of the picture; a 3-layer pooling layer (MaxPool 2 d) which is used for data compression and reduces the calculated amount; two full link layers (Linear) are used to obtain the overall characteristics of the sample. Wherein ReLU is an activation function used to determine the gradient of the change in the parameter, dropout is used to prevent model overfitting. The input data of the model is a 3-channel (RGB color) picture with 640 multiplied by 480 pixels, and the output data is a label for predicting the fish meat type represented by the picture. That is, the CNN model includes a first convolution layer 1, a first pooling layer 2, a second convolution layer 3, a second pooling layer 4, a third convolution layer 5, a third pooling layer 6, a first fully connected layer 7, and a second fully connected layer 8, which are sequentially connected.
The model has fewer layers, is suitable for lightweight application, occupies less memory, and is suitable for being installed on some small equipment instruments.
Claims (1)
1. The fish meat identification method based on LIBS-RAMAN spectrum technology is characterized by comprising the following steps of:
(1) Grinding the fish meat to be identified into minced meat by a meat grinder, filling the minced meat into a small-aperture culture dish, and finally compacting and refrigerating by using a glass sheet to obtain a fish meat sample with a flat surface;
(2) Respectively acquiring LIBS spectrum data and RAMAN spectrum data of a fish sample to be identified, carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1;
(3) Merging the normalized RAMAN and LIBS spectrum data to obtain RAMAN-LIBS merged picture data, and then inputting the RAMAN-LIBS merged picture data into a trained CNN model to obtain a fish meat identification result; the training process of the CNN model comprises the following steps:
(301) Treating a sample by adopting a meat emulsion method, grinding fish meat into meat emulsion by a meat grinder, filling the meat emulsion into a small-aperture culture dish with the diameter of 35mm, and finally compacting and refrigerating by using glass sheets, wherein the sample comprises different types of fish meat and different parts of the fish meat;
(302) Respectively acquiring LIBS spectrum data and RAMAN spectrum data of fish meat to be identified, carrying out normalization processing on the acquired RAMAN spectrum data to limit the spectrum peak value to be between 0 and 1, and carrying out normalization processing on the acquired LIBS spectrum data to limit the spectrum peak value to be between 0 and 1;
(303) Simultaneously converting the normalized RAMAN spectrum data and LIBS spectrum data into RAMAN-LIBS fusion picture data;
(304) Dividing the RAMAN-LIBS fusion picture data into 3 parts according to a proper proportion, wherein the 3 parts are respectively a training set, a verification set and a test set;
(305) The training set data is used for network training, the training set is used for correcting model parameters, the training set is used for repeatedly training to obtain a trained model, then the verification set is used for preliminarily verifying whether the model is good or bad, if the model is qualified, the model is exported to stop training, and then the test set is used for making a final evaluation standard; the CNN model comprises a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a third convolution layer, a third pooling layer, a first full-connection layer and a second full-connection layer which are sequentially connected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210070023.4A CN114414551B (en) | 2022-01-21 | 2022-01-21 | Fish identification method based on LIBS-RAMAN spectrum technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210070023.4A CN114414551B (en) | 2022-01-21 | 2022-01-21 | Fish identification method based on LIBS-RAMAN spectrum technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114414551A CN114414551A (en) | 2022-04-29 |
| CN114414551B true CN114414551B (en) | 2023-09-01 |
Family
ID=81275351
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210070023.4A Active CN114414551B (en) | 2022-01-21 | 2022-01-21 | Fish identification method based on LIBS-RAMAN spectrum technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114414551B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103278464A (en) * | 2013-04-18 | 2013-09-04 | 北京工商大学 | Method and device for fish flesh detection |
| CN103760110A (en) * | 2013-05-23 | 2014-04-30 | 山东商业职业技术学院 | Method for rapidly identifying meat with different animal sources |
| JP6712691B1 (en) * | 2019-11-20 | 2020-06-24 | 中国計量大学 | Method and system for distinguishing mixed salmon production areas |
| CN111735806A (en) * | 2020-06-18 | 2020-10-02 | 中国海洋大学 | Rapid fish product identification method based on laser-induced breakdown spectroscopy technology |
-
2022
- 2022-01-21 CN CN202210070023.4A patent/CN114414551B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103278464A (en) * | 2013-04-18 | 2013-09-04 | 北京工商大学 | Method and device for fish flesh detection |
| CN103760110A (en) * | 2013-05-23 | 2014-04-30 | 山东商业职业技术学院 | Method for rapidly identifying meat with different animal sources |
| JP6712691B1 (en) * | 2019-11-20 | 2020-06-24 | 中国計量大学 | Method and system for distinguishing mixed salmon production areas |
| CN111735806A (en) * | 2020-06-18 | 2020-10-02 | 中国海洋大学 | Rapid fish product identification method based on laser-induced breakdown spectroscopy technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114414551A (en) | 2022-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11280777B2 (en) | Apparatus and method for multimode analytical sensing of items such as food | |
| Huang et al. | Development of a multispectral imaging system for online detection of bruises on apples | |
| Li et al. | Detection of common defects on oranges using hyperspectral reflectance imaging | |
| CN107101943B (en) | Device and method for detecting microalgae activity by optical flow control lens-free holographic imaging | |
| Qin et al. | Hyperspectral and multispectral imaging for evaluating food safety and quality | |
| US8902423B2 (en) | Classification using multivariate optical computing | |
| Sivertsen et al. | Automatic nematode detection in cod fillets (Gadus morhua) by transillumination hyperspectral imaging | |
| US20200392555A1 (en) | Substance or contamination detection | |
| EP3322969B1 (en) | System and method for adjusting cytometer measurements | |
| CN103900972B (en) | Multi-feature fusion-based meat freshness hyperspectral image visual detection | |
| Cozzolino | Near infrared spectroscopy and food authenticity | |
| Dong et al. | Automated analysis of microplastics based on vibrational spectroscopy: are we measuring the same metrics? | |
| CN107064084B (en) | Microminiature laser fluorescence spectrometer and spectrum detection method | |
| CN105548113A (en) | Plant physiology monitoring method based on chlorophyll fluorescence and multispectral image | |
| CN107389638A (en) | A kind of microscopic fluorescent spectral imaging marine phytoplankton original position classifying identification method and device | |
| Nakauchi et al. | Selection of optimal combinations of band-pass filters for ice detection by hyperspectral imaging | |
| CN109187443B (en) | Water body bacteria microorganism accurate identification method based on multi-wavelength transmission spectrum | |
| Song et al. | Detection of fish bones in fillets by Raman hyperspectral imaging technology | |
| Haraguchi et al. | Monitoring natural phytoplankton communities: A comparison between traditional methods and pulse-shape recording flow cytometry | |
| CN114414551B (en) | Fish identification method based on LIBS-RAMAN spectrum technology | |
| CN109142251A (en) | The LIBS quantitative analysis method of random forest indirect labor's neural network | |
| RU2498275C2 (en) | Remote classification method of oil impurities on water surface | |
| Bagheri et al. | Estimation of optical properties of nearshore water | |
| Vasilescu et al. | Analysis of seawater pollution using neural networks and channels relationship algorithms | |
| CN113906284A (en) | Hyperspectral quantitative imaging cytometry system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |