A method for identifying biomass based on machine vision and raman laser coupling

Abstract

The application discloses a method for recognizing biomass based on machine vision and Raman laser coupling, comprising the following steps: obtaining images, Raman spectra and physicochemical property data of various biomasses; extracting image characteristic parameters of the image data to obtain a first characteristic data set; and extracting Raman characteristic parameters of the Raman spectra to obtain a second characteristic data set; respectively establishing a Hardgrove grindability prediction model, a moisture content prediction model and a carbon-hydrogen element ratio prediction model based on the first characteristic data set and the second characteristic data set; obtaining images, Raman spectra and physicochemical property data of a to-be-detected biomass, and recognizing the to-be-detected biomass by using the three prediction models. The method has the rapidity of machine vision recognition and the accuracy of Raman laser recognition, can obtain the grindability and moisture content of the to-be-recognized biomass, and can also obtain the chemical components of the to-be-recognized biomass; compared with single method recognition, the method expands the results of biomass detection, and can also overcome the interference of surface impurities of the to-be-detected object on Raman laser recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomass detection, and particularly relates to a method for recognizing biomass based on machine vision and Raman laser coupling. BACKGROUND

[0002] As a renewable energy, the advantage of biomass power generation technology is to reduce the consumption of fossil energy and improve the structure of energy utilization. When a power plant uses biomass fuel for power generation, the power generation efficiency and the smooth operation of the power plant are directly related to the quality of the biomass fuel. Especially when multiple types of biomass are mixed for combustion, it is necessary to make the combustion quality of the mixed biomass closer to the combustion characteristics of the combustion boiler. Therefore, it is necessary to detect the biomass before it is sent into the boiler for combustion to determine its type and chemical composition.

[0003] The traditional biomass detection method is to sample and analyze the sample in the laboratory to determine its composition and chemical composition. Although the results obtained by the traditional biomass detection method are accurate, it cannot realize online real-time non-destructive detection, cannot be coupled with the intelligent control system of the power plant, and is not conducive to improving the power generation efficiency of the power plant.

[0004] Therefore, there is an urgent need for a non-destructive, online biomass rapid detection method to not only improve the detection efficiency, but also reduce the loss of raw materials under the condition of ensuring the detection accuracy. SUMMARY

[0005] In view of the above technical problems, the present application provides a method for recognizing biomass based on machine vision and Raman laser coupling to realize real-time, non-destructive online detection of biomass.

[0006] The technical scheme provided by the present application is as follows:

[0007] In a first aspect, a method for recognizing biomass based on machine vision and Raman laser coupling is provided, comprising the following steps:

[0008] Obtaining images, Raman spectra and physicochemical property data of multiple types of biomass;

[0009] Extracting image feature parameters of the image data to obtain a first feature data set, and extracting Raman feature parameters of the Raman spectrum to obtain a second feature data set;

[0010] Based on the first feature data set and the second feature data set, respectively establishing a Hardgrove grindability prediction model, a moisture content prediction model and a carbon-hydrogen element ratio prediction model;

[0011] Obtaining the images, Raman spectra and physicochemical property data of the to-be-detected biomass, and identifying the to-be-detected biomass by using the three prediction models.

[0012] In a possible implementation, the image is an image of the biomass after being ground into particles.

[0013] In a possible implementation, the image feature parameters include average particle size and grayscale.

[0014] In a possible implementation, the physicochemical property data include a Hardgrove grindability index, a moisture content, and a carbon-hydrogen element ratio.

[0015] In a possible implementation, the method for obtaining the Raman feature parameters includes:

[0016] obtaining a total area S1 of first-order Raman peaks and a total area S2 of second-order Raman peaks in a Raman spectrum, and a peak area of each feature peak in a second-order Raman spectrum range;

[0017] selecting an area ratio that can represent characteristics of the biomass as the Raman feature parameter.

[0018] In a possible implementation, the method for establishing the Hardgrove grindability prediction model includes:

[0019] taking the first feature data set and the second feature data set as input, and taking the Hardgrove grindability index corresponding to the biomass as output;

[0020] training by using a neural network to obtain the Hardgrove grindability prediction model.

[0021] In a possible implementation, the method for establishing the moisture content prediction model includes:

[0022] coupling the particle size feature and the grayscale feature in the first feature data set to obtain a picture coupling feature;

[0023] taking the picture coupling feature and the second feature data set as input, and taking the mass moisture content as output;

[0024] training by using a neural network to obtain the moisture content prediction model.

[0025] In a possible implementation, the method for establishing the carbon-hydrogen element ratio prediction model includes:

[0026] coupling the particle size feature and the grayscale feature in the first feature data set to obtain a picture coupling feature;

[0027] taking the particle size feature and the grayscale feature as input, and taking the picture coupling feature as output, to fit a first equation;

[0028] taking the picture coupling feature and the second feature data set as input, and taking the carbon-hydrogen element ratio prediction as output, to fit a second equation;

[0029] Substitute the first equation into the second equation to obtain a carbon-hydrogen element ratio prediction model.

[0030] Further, the method for coupling the particle size feature and the gray scale feature in the first feature data set to obtain the picture coupling feature comprises:

[0031] Feature fusion: weighted sum and normalization of the particle size feature and the gray scale feature to obtain a fusion feature set;

[0032] Depth information coupling: calculating the depth information of each pixel point of each image in the image data, and forming a binary array with the corresponding fusion feature set to obtain the picture coupling feature.

[0033] In a second aspect, a device for identifying biomass based on machine vision and Raman laser coupling is provided, comprising:

[0034] The acquisition module is configured to obtain images, Raman spectra and physicochemical property data of a plurality of biomasses;

[0035] The extraction module is configured to extract image feature parameters of the image data to obtain a first feature data set, and extract Raman feature parameters of the Raman spectra to obtain a second feature data set;

[0036] The construction module is configured to establish a hardness grindability prediction model, a moisture content prediction model and a carbon-hydrogen element ratio prediction model based on the first feature data set and the second feature data set, respectively;

[0037] The identification module is configured to obtain images, Raman spectra and physicochemical property data of a to-be-tested biomass, and identify the to-be-tested biomass by using the three prediction models.

[0038] The present application has the following beneficial effects:

[0039] (1) The method described in the present application has the advantages of Raman laser identification and machine vision identification, and the identification process is accurate, fast and non-destructive.

[0040] (2) Compared with the use of Raman laser identification alone, when the surface of the biomass is contaminated, the detection of the Raman laser will fail, but the method described in the present application can still accurately reveal the type of the biomass through machine vision means.

[0041] (3) The method described in the present application can obtain more comprehensive and accurate physicochemical information of the biomass, such as the grindability index, the moisture content and the carbon-hydrogen element ratio, and has a wide range of applications. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart is a schematic diagram of the present application;

[0043] Figure 2 The schematic diagram is a part of the biomass samples in the embodiment of the present application;

[0044] Figure 3 For the embodiment of the application, the sample biomass picture collected is obtained by image processing to obtain the particle size picture; wherein the upper left is peanut straw powder, the lower left is sugarcane residue, the upper right is corn cob powder, and the lower right is peanut shell;

[0045] Figure 4 For the original picture and gray picture and gray value of peanut shell powder with different moisture contents;

[0046] Figure 5 For the spectrum obtained by laser Raman testing of a part of the collected biomass sample;

[0047] Figure 6 For the Raman spectrum of the biomass to be tested and the baseline correction processing schematic diagram;

[0048] Figure 7 For the peak separation schematic diagram of the second-order Raman spectrum in the embodiment of the application;

[0049] Figure 8 (a) is a schematic diagram of the BP neural network model structure in the Hardgrove grindability prediction model; Figure 8 (b) is a schematic diagram of the construction of the Hardgrove grindability prediction model;

[0050] Figure 9 For the fitting effect of the Hardgrove grindability index in the database;

[0051] Figure 10 For the schematic diagram of the construction of the moisture content prediction model;

[0052] Figure 11 For the fitting effect of the mass moisture content index in the database;

[0053] Figure 12 For the schematic diagram of the construction of the carbon-hydrogen element ratio C / H prediction model;

[0054] Figure 13 For the fitting effect of the carbon-hydrogen element ratio in the database;

[0055] Figure 14 For the comparison of the predicted value and the true value of the Hardgrove grindability index, the mass moisture content, and the carbon-hydrogen element ratio of the biomass to be detected;

[0056] Figure 15 The structure schematic diagram of the device for recognizing biomass based on machine vision and Raman laser coupling provided by the application. DETAILED DESCRIPTION

[0057] The content of the application will be further described below in conjunction with specific embodiments, and the content of the application is not limited thereto.

[0058] Reference Figure 1The application discloses a method for identifying biomass based on machine vision and Raman laser coupling, and comprises the following steps:

[0059] S100, obtaining images, Raman spectra and physicochemical property data of a plurality of biomasses.

[0060] In a possible implementation manner, the images are images of the biomasses after being ground into particles.

[0061] It can be understood that, in order to ensure the accuracy of subsequent model training, the particle size distribution of the ground biomass is ensured to be consistent to a certain extent. For example, the consistency of grinding can be ensured by unifying related parameters of grinding.

[0062] In a possible implementation manner, the physicochemical property data comprises a Hardgrove grindability index, a moisture content and a carbon-hydrogen element ratio.

[0063] S200, extracting image feature parameters of the image data to obtain a first feature data set, and extracting Raman feature parameters of the Raman spectrum to obtain a second feature data set.

[0064] In a possible implementation manner, the image feature parameters comprise an average particle size and a gray scale.

[0065] Further, the method for obtaining the average particle size comprises the following steps:

[0066] Image preprocessing: the collected image of the biomass is preprocessed, including denoising, contrast enhancement and the like, so as to improve the image quality;

[0067] Edge detection: an edge detection algorithm (such as a Canny algorithm) is used to identify the edges of the biomass particles in the image;

[0068] Particle segmentation: the biomass particles are segmented from the background by the edge detection result, so as to form independent particle images;

[0069] Particle size calculation: the size of the segmented particle image is calculated, and the diameter of each particle is measured;

[0070] Average particle size calculation: the average value of all particle diameters is calculated to obtain the average particle size.

[0071] Further, the method for obtaining the gray scale comprises the following steps:

[0072] Image graying: the collected image is converted into a gray scale image, so as to reduce the interference of color information;

[0073] Gray scale normalization: the maximum pixel value and the minimum pixel value in the gray scale image are calculated, and then each pixel value is mapped to the range of 0 to 255, so as to enhance the contrast of the image;

[0074] Gray scale feature extraction: gray scale features of the image are extracted by using a gray level co-occurrence matrix (GLCM) and the like; the gray scale features include contrast, uniformity, energy and the like of the texture;

[0075] Gray scale projection: gray scale value projections of the image in horizontal and vertical directions are calculated to obtain gray scale distribution features of the image.

[0076] In a possible implementation manner, the method for obtaining the Raman characteristic parameter comprises:

[0077] obtaining a total area S1 of a first-order Raman peak and a total area S2 of a second-order Raman peak by using Raman spectrum, and peak areas of each characteristic peak in the second-order Raman spectrum range;

[0078] selecting an area ratio capable of representing characteristics of the biomass as the Raman characteristic parameter.

[0079] S300, based on the first characteristic data set and the second characteristic data set, respectively establishing a Hardgrove grindability prediction model, a moisture content prediction model and a carbon-hydrogen element ratio prediction model.

[0080] In a possible implementation manner, the method for establishing the Hardgrove grindability prediction model comprises:

[0081] taking the first characteristic data set and the second characteristic data set as inputs and taking the Hardgrove grindability index corresponding to the biomass as output;

[0082] training by using a neural network to obtain the Hardgrove grindability prediction model.

[0083] In a possible implementation manner, the method for establishing the moisture content prediction model comprises:

[0084] coupling the particle size feature and the gray scale feature in the first characteristic data set to obtain picture coupling features;

[0085] taking the picture coupling features and the second characteristic data set as inputs and taking the moisture content as output;

[0086] training by using a neural network to obtain the moisture content prediction model.

[0087] In a possible implementation manner, the method for establishing the carbon-hydrogen element ratio prediction model comprises:

[0088] coupling the particle size feature and the gray scale feature in the first characteristic data set to obtain picture coupling features;

[0089] taking the particle size feature and the gray scale feature as inputs and taking the picture coupling features as output, and fitting to obtain a first equation;

[0090] The picture coupling feature and the second feature dataset are taken as inputs, and the carbon-hydrogen element ratio prediction is taken as output, and a second equation is fitted;

[0091] The first equation is substituted into the second equation to obtain a carbon-hydrogen element ratio prediction model.

[0092] Further, the method for coupling the particle size feature and the gray scale feature in the first feature dataset to obtain the picture coupling feature comprises:

[0093] Feature fusion: weighted sum and normalization of the particle size feature and the gray scale feature to obtain a fusion feature set;

[0094] Depth information coupling: calculate the depth information of each pixel point in the image data of each image, and form a binary array with the corresponding fusion feature set to obtain the picture coupling feature.

[0095] Specifically, the steps of feature fusion are as follows:

[0096] Weighted sum: weighted sum of the particle size feature and the gray scale feature, formula: , wherein is a weight coefficient.

[0097] Normalization processing: normalization processing is performed on the weighted value to ensure that the feature value is within a suitable range, and a fusion feature set is obtained.

[0098] It should be noted that the information depth of the pixel point refers to additional information obtained by calculating the depth value of each pixel point in the image processing process. Depth information can be used to enhance the expression ability of features, which can be:

[0099] Depth value: in a three-dimensional image or depth image, the depth value of each pixel point represents the distance of the point from the observation plane.

[0100] Contrast information: the contrast information of the image is obtained by calculating the depth difference of adjacent pixel points in the image.

[0101] For example, the particle size feature is PZ=3.11 (average particle size), the gray scale feature is G=105 (gray scale value), and the weight is assumed to be The weighted value F is obtained by weighted sum 0.6*3.11+0.4*105=43.666.

[0102] After calculating the weighted value F of all image data, normalization processing is performed, and the fusion feature set obtained after normalization , The value of the fusion feature set is between 0 and 1, and the normalized data is not affected by the unit.

[0103] Then, the distance L of each pixel in the image data relative to the camera lens reference plane is calculated, and the distance L is combined with the corresponding image to form a binary array (L, L), which is the image coupling feature. And the corresponding L is combined to form a binary array (L, L), which is the image coupling feature. , L), which is the image coupling feature.

[0104] Further, the neural network includes a BP neural network algorithm and a random forest algorithm; and the neural network model is evaluated for performance by cross-validation.

[0105] S400, obtaining the image, Raman spectrum and physicochemical property data of the biomass to be tested, and identifying the biomass to be tested by using three prediction models.

[0106] The following will be described in more specific embodiments, as follows.

[0107] A method for identifying biomass based on machine vision and Raman laser coupling, comprising the following steps:

[0108] (1) Obtain the image, Raman spectrum and physicochemical property data of a plurality of biomass, and extract the characteristic parameters

[0109] Ten types of biomass particles were selected as samples to establish a database. The selected samples were all on an air-dry basis. The images of these biomass were obtained, and their average particle size and image gray value were obtained after image processing, and their Hardgrove grindability data were measured. Part of the biomass image is shown in Figure 2 The average particle size of the biomass particles obtained by image processing is shown in Figure 3 The image gray value of part of the biomass is shown in Figure 4 The results are shown in Table 1.

[0110] Table 1

[0111] Biomass Grindability index HGI Particle image size PZ Grey value G Peanut straw 20.01 3.11 105 Corn cob 22.03 3.19 90.35 Bagasse 10.92 10.09 77.52 Peanut shell 42.23 1.97 142.26 Corn straw 21.98 3.22 114.86 Rice husk 41.01 2 83.21 Traditional Chinese medicine residue 54.56 1.56 50.21 Wood chip 24.45 3.14 86.26 Coffee shell 38.21 2.03 135.25 Cotton stalk 27.98 3.16 140.45

[0112] Ten types of biomass particles were selected as samples to establish a database. The selected samples were all on an air-dry basis. The images of these biomass were obtained, and their average particle size and image gray value were obtained after image processing, and their Hardgrove grindability data were measured. Part of the biomass image is shown in

[0113] Table 2

[0114] Biomass Mass moisture content W Particle image size PZ Grey value G Peanut straw 8.87% 3.11 105 Corn cob 12.45% 3.19 90.35 Bagasse 15.43% 10.09 77.52 Peanut shell 7.21% 1.97 142.26 Corn straw 22% 3.22 114.86 Rice husk 10.09% 2 83.21 Traditional Chinese medicine residue 40% 1.56 50.21 Wood chip 15% 3.14 86.26 Coffee shell 8.62% 2.03 135.25 Cotton stalk 9.04% 3.16 140.45

[0115] The selected 10 types of biomass were tested by Raman spectroscopy. The Raman spectrum is shown in Figure 5 Table 3 is the condition of Raman test when the sample substance is used to make the biomass-Raman laser database sample. The baseline of part of the Raman spectrum of the biomass drifts, as shown in Figure 6 The baseline needs to be corrected.

[0116] Table 3

[0117] Laser wavelength Laser power Eyepiece magnification Scanning time Scanning range 532.16 nm 5 mw X50 10 s 500-3400 cm-1

[0118] After completing the above steps, the second-order Raman peak is split into three peaks, one of which is shown in the example below. Figure 7 As shown. The Raman spectrum parameters obtained after peak splitting include the total area S1 of the first-order Raman peak, the total area S2 of the second-order Raman peak, and the area within 2940 cm⁻¹ of the second-order Raman spectrum. -1 3200cm -1 Isopal area I 2940 ,I 3200 and their combinations I 2940 / S1,I 3200 / S1,I 3200 / S2,I 2940 / S2,I 2940 / I 3200 ,(I 2940 +I 3200 ) / S1,(I 2940 +I 3200 ) / S2, S2 / S1, etc. Among them (I 2940 +I 3200 ) indicates I 2940 and I 3200 sum.

[0119] After peak fractionation calculations, a mapping relationship was established between the selected sample biomass Raman spectral characteristic parameters and its chemical composition parameters, thus successfully constructing a biomass-Raman database. In this specific application example, (I2940+I3200) / S2 was used as the Raman characteristic parameter of the biomass. The carbon-hydrogen ratio and Raman characteristic parameters of the sample biomass are shown in Table 4.

[0120] Table 4

[0121]

[0122] (2) Establish the Hardy grindability prediction model, moisture content prediction model and carbon-hydrogen ratio prediction model.

[0123] (2.1) Establish a predictive model for Hardy grindability with respect to {PZ,G,(I2940+I3200) / S2}

[0124] The construction method is as follows: A BP neural network algorithm is adopted, using three sets of data, PZ, G, and (I2940+I3200) / S2, as inputs, and the Hardy Grindability Index of biomass as the output. A set of hidden layers and output layers are set up. The structure of the BP neural network is shown in Figure 8(a). The model construction is shown in Figure 8(b).

[0125] During the model training process, first, the particle size data PZ, the gray data G and the Raman characteristic data (I2940+I3200) / S2 of the biomass are standardized, the data set is divided into a training set and a test set according to 8:2, 10 hidden layer neurons are set, the iteration number is set to 1000 times, and the learning rate is set to 0.01, and the grindability prediction model is obtained after training. The fitting effect of the model on the test set is as shown in Figure 9 According to the fitting effect, it can be known that the grindability index calculated by fitting is basically consistent with the actual grindability index of the biomass.

[0126] (2.2) Establish a prediction model of the moisture content W of the biomass particles about {PZ, G, (I2940+I3200) / S2}

[0127] The construction method is as follows: first, the particle size characteristics and the picture gray value characteristics are coupled to obtain the picture characteristics ImageFeature, abbreviated as IF, and the IF and the RF two groups of data are taken as inputs and sent into the BP neural network model, and the mass moisture content W is taken as the output. The prediction model construction is as shown in Figure 10 The fitting effect of the model on the test set is as shown in Figure 11 According to the fitting result, it can be known that the mass moisture content of the biomass calculated by fitting is basically consistent with the actual mass moisture content.

[0128] (2.3) Establish a prediction model of the carbon-hydrogen element ratio C / H of the biomass particles about {PZ, G, (I2940+I3200) / S2}

[0129] The construction method is as follows: first, PZ and G are taken as inputs, and image characteristics IF is taken as output for linear fitting, and the linear fitting equation is ; then, the IF and the RF two groups of data of the biomass are taken as inputs, and C / H is taken as output for linear fitting, and the fitting equation is . The prediction model construction is as shown in Figure 12 The fitting effect of C / H is as shown in Figure 13 According to the fitting result, it can be known that the carbon-hydrogen ratio of the biomass calculated by fitting is basically consistent with the actual carbon-hydrogen ratio.

[0130] (3) Obtain the image, Raman spectrum and physicochemical property data of the to-be-tested biomass, and identify the to-be-tested biomass by using the three prediction models

[0131] (3.1) Select the biomass to be tested, process it according to the same method in step (2) to obtain the average particle size PZ, the picture gray G and the Raman characteristic parameters as shown in Table 5. It is noted that the peak area is as follows: 2940 ,I 3200S1, S2 and the like are signal strength, without a certain physical unit. The calculation is kept in the same scale. Meanwhile, the industrial element composition analysis of the biomass is carried out, and the results are shown in Table 6.

[0132] Table 5

[0133]

[0134] Table 6

[0135]

[0136] (3.2) the obtained characteristic parameters are sent into the coupling model constructed in step (2), and the hardness grindability index, mass moisture content and carbon-hydrogen element ratio of the biomass are respectively predicted, the obtained prediction results are shown in Table 6, the prediction results are consistent with the characteristic parameters of peanut shell in the database, so it can be determined that the biomass to be measured is peanut shell. Figure 14

[0137] The following describes a device for identifying biomass based on machine vision and Raman laser coupling provided by the present application, and the device for identifying biomass based on machine vision and Raman laser coupling described below can be correspondingly referred to the method for identifying biomass based on machine vision and Raman laser coupling described above.

[0138] Figure 15 is a structural schematic diagram of a device for identifying biomass based on machine vision and Raman laser coupling provided by an embodiment of the present application, as shown in Figure 15 , comprising: an acquisition module 1511, an extraction module 152, a construction module 153 and an identification module 154, wherein:

[0139] The acquisition module 151 is used for obtaining images, Raman spectra and physicochemical property data of a plurality of biomasses.

[0140] The extraction module 152 is used for extracting image characteristic parameters of the image data to obtain a first characteristic data set, and extracting Raman characteristic parameters of the Raman spectrum to obtain a second characteristic data set.

[0141] The construction module 153 is used for respectively establishing a hardness grindability prediction model, a moisture content prediction model and a carbon-hydrogen element ratio prediction model based on the first characteristic data set and the second characteristic data set.

[0142] The identification module 154 is used for obtaining images, Raman spectra and physicochemical property data of a biomass to be measured, and identifying the biomass to be measured by using the three prediction models.

[0143] ​It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying biomass based on machine vision and Raman laser coupling, characterized in that, The method comprises the following steps: Obtaining images, Raman spectra and physicochemical property data of a plurality of biomasses; the images are images of the biomasses after being ground into particles; the physicochemical property data comprises Hargen grindability index, moisture content and carbon-hydrogen element ratio; Extracting image feature parameters of the image data to obtain a first feature data set; And extracting Raman feature parameters of the Raman spectra to obtain a second feature data set; the image feature parameters comprise average particle size and gray scale; Based on the first feature data set and the second feature data set, Hargen grindability prediction model, moisture content prediction model and carbon-hydrogen element ratio prediction model are respectively established; Obtaining images, Raman spectra and physicochemical property data of a to-be-tested biomass, and identifying the to-be-tested biomass by using the three prediction models.

2. The method of identifying biomass based on machine vision and Raman laser coupling according to claim 1, characterized in that, The method for obtaining the Raman feature parameters comprises: Obtaining total area S1 of first-order Raman peaks and total area S2 of second-order Raman peaks by Raman spectrum; and peak area of each feature peak in the second-order Raman spectrum range; Selecting an area ratio capable of representing characteristics of the biomass as the Raman feature parameter.

3. The method of claim 1, wherein the biomass is identified based on machine vision and Raman laser coupling. The method for establishing the Hargen grindability prediction model comprises: Taking the first feature data set and the second feature data set as inputs and taking the Hargen grindability index corresponding to the biomass as output; Training by using a neural network to obtain the Hargen grindability prediction model.

4. The method of claim 1, wherein the biomass is identified based on machine vision and Raman laser coupling. The method for establishing the moisture content prediction model comprises: Coupling particle size features and gray scale features in the first feature data set to obtain picture coupling features; Taking the picture coupling features and the second feature data set as inputs and taking mass moisture content as output; Training by using a neural network to obtain the moisture content prediction model.

5. The method of identifying biomass based on machine vision and Raman laser coupling of claim 1, wherein, The method for establishing the carbon-hydrogen element ratio prediction model comprises: Coupling particle size features and gray scale features in the first feature data set to obtain picture coupling features; Taking the particle size features and the gray scale features as inputs and taking the picture coupling features as output, a first equation is fitted; Taking the picture coupling features and the second feature data set as inputs and taking carbon-hydrogen element ratio prediction as output, a second equation is fitted; Substituting the first equation into the second equation to obtain the carbon-hydrogen element ratio prediction model.

6. The method of identifying biomass based on machine vision and Raman laser coupling according to claim 4 or 5, characterized in that, The method for coupling the particle size features and the gray scale features in the first feature data set to obtain the picture coupling features comprises: Feature fusion: weighted summation and normalization of the particle size features and the gray scale features are performed to obtain a fusion feature set; Depth information coupling: depth information of each pixel point of each image in the image data is calculated, and the depth information is coupled with the corresponding fusion feature set to form a binary array, and the picture coupling features are obtained.

7. An apparatus for identifying biomass based on machine vision and Raman laser coupling, characterized in that, The method comprises: A collection module is configured to obtain images, Raman spectra and physicochemical property data of a plurality of biomasses; the images are images of the biomasses after being ground into particles; the physicochemical property data comprises Hargen grindability index, moisture content and carbon-hydrogen element ratio; An extraction module is configured to extract image feature parameters of the image data to obtain a first feature data set; And extract Raman feature parameters of the Raman spectra to obtain a second feature data set; the image feature parameters comprise average particle size and gray scale; A construction module is configured to, based on the first feature data set and the second feature data set, respectively establish Hargen grindability prediction model, moisture content prediction model and carbon-hydrogen element ratio prediction model; The recognition module is used for acquiring images, Raman spectrograms and physicochemical property data of the to-be-tested biomass, and recognizing the to-be-tested biomass by using three prediction models.

Images