Lonicera caerulea active ingredient spectrum quantitative analysis method and system based on deep learning

Abstract

The invention relates to the technical field of spectral quantitative analysis, in particular to a spectral quantitative analysis method and system for active ingredients of lonicera caerulea based on deep learning, and the method comprises the following steps: material selection and cleaning, sampling environment setting, sample sampling, active ingredient sample spectrum determination and template area determination. The system comprises a base, a detection shell is installed at the upper end of the base, and a lifting mechanism is jointly installed in the base and the detection shell. The lonicera caerulea can be fully processed, the stability of the environment during sampling is ensured, the situation that the sampling quality is affected due to changes of the external environment condition is avoided, meanwhile, the spectrum information of active ingredients in the lonicera caerulea can be accurately obtained, and then through spectrum sampling on the surface of the lonicera caerulea, the quality of the active ingredients in the lonicera caerulea is improved. The comparison of the sampling spectrum information and the spectrum information of different active components is fully realized through an algorithm so as to analyze the active components; meanwhile, the structure scheme of the sampling analysis system is also shown.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral quantitative analysis, and particularly to a spectral quantitative analysis method and system for the active components of blue honeysuckle based on deep learning. Background Art

[0002] Blue honeysuckle is bitter in taste and cool in nature. It can clear heat and purge fire, disperse swelling and carbuncles, and treat heat-toxic carbuncle sores such as breast abscess, intestinal abscess, and erysipelas. Its fruit contains chemical components such as anthocyanins, carotenoids, and anthocyanidins; it has the effects of lowering blood pressure, improving myocardial ischemia, enhancing children's eyesight, and preventing skin aging.

[0003] The main active components of blue honeysuckle include anthocyanins, vitamin P, vitamin C, vitamin B1, B2, potassium, iron, zinc, calcium, and polyphenolic substances such as phenolic acids and flavonoids. These components endow blue honeysuckle with various health benefits.

[0004] Currently, the main methods for determining the content of active components in plants, traditional Chinese medicines, etc. are gas chromatography (GC), ultraviolet spectrophotometry (UV), high performance liquid chromatography (HPLC), gas chromatography-mass spectrometry (GC-MS), nuclear magnetic resonance (NMR), etc. Although these methods are technically mature and have high accuracy, their pretreatment is complex, the operation process is cumbersome, and it takes a long time, and the purpose of rapid detection of the active components of plants and traditional Chinese medicines cannot be achieved, so improvement is needed. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies in the prior art, and propose a spectral quantitative analysis method and system for the active components of blue honeysuckle based on deep learning.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A spectral quantitative analysis method for the active components of blue honeysuckle based on deep learning, comprising the following steps:

[0008] S1. Material selection and cleaning:

[0009] Obtain mature blue honeysuckle, clean the blue honeysuckle at the same time, dry it in the air, and then screen the blue honeysuckle to select those with plump grains and no damage on the surface for use;

[0010] S2. Sampling environment setting:

[0011] Set a stable sampling environment to avoid sampling deviation caused by environmental changes;

[0012] S3. Sample sampling:

[0013] Set the spectral sampling conditions, classify and sample the samples through the camera and the hyperspectral image acquisition device. At the same time, group the samples and the data obtained through the camera and the hyperspectral image acquisition device, and ensure that the samples correspond to the two sampling data;

[0014] S4. Spectral determination of active ingredient samples:

[0015] Classify the active ingredients in the blue honeysuckle and obtain the spectral data information of the corresponding ingredients;

[0016] S5. Define the sample area:

[0017] Identify the blue honeysuckle area in the picture through the samples obtained by the camera, and define the comparison area 1 in the blue honeysuckle area. The samples obtained by the hyperspectral image acquisition device are consistent with the corresponding camera sample specifications. The area of the comparison area 1 in the camera sample can be used to identify the comparison area 2 in the samples obtained by the hyperspectral image acquisition device;

[0018] S6. Active ingredient analysis:

[0019] Compare the spectral data information of the active ingredients obtained in S4 with the spectral information in the comparison area 2. Process the comparison area 2 in S5 through the YOLO algorithm, and compare the processed data with the spectral data information of the corresponding ingredients in S4 to ensure the accuracy of data acquisition.

[0020] Compared with the prior art, the present application can fully process the blue honeysuckle, ensure the stability of the sampling environment, avoid affecting the sampling quality due to changes in the external environment, accurately obtain the spectral information of the active ingredients in the blue honeysuckle, then sample the spectrum on the surface of the blue honeysuckle, and fully realize the comparison between the sampling spectrum and the spectrum information of different active ingredients through the algorithm for active ingredient analysis.

[0021] Preferably, the stable sampling environment in S2 includes constant temperature, brightness, and air flow velocity.

[0022] Furthermore, ensure the stability of the sampling environment and avoid the influence of the external environment on sampling.

[0023] Preferably, the hyperspectral image acquisition device in S3 includes a near-infrared camera GaiaField-N17, a light source, a dark box, and a loading plate; the near-infrared camera GaiaField-N17 has 256 spectral channels and can record the reflection spectrum in the range of 900-1700 nm.

[0024] Furthermore, ensure the sampling quality.

[0025] Preferably, the YOLO algorithm in S6 includes the following steps:

[0026] Step 1: Use the Input module of the YOLOv5 model to perform adaptive scaling and data augmentation preprocessing on the input hyperspectral image, and then unify the image size through the resize() function of the Python programming language as the input of the YOLO active ingredient hyperspectral recognition model;

[0027] Step 2: Input the preprocessed hyperspectral image with annotation information into the Backbone module of the YOLOv5 model for feature extraction to obtain multi-layer feature maps;

[0028] Step 3: Input the feature maps output by the Backbone module of the YOLOv5 model into the prediction module to obtain prediction results; the Backbone module uses the FOCUS layer to convert the information on the hyperspectral image plane to the channel dimension through slicing operations, and then extracts different features through convolutional layers; the extracted features of the Backbone module are input into the Neck module, and the Neck module further processes and integrates the features from the backbone network through feature enhancement, feature fusion, and feature processing operations; the prediction module consists of convolutional layers and fully connected layers, and is responsible for generating the recognition results of the YOLO hyperspectral image recognition model, including the bounding box position, category, and confidence of the target;

[0029] Step 4: Construct a loss function based on the prediction results, and continuously train and optimize the model based on the loss function.

[0030] Furthermore, implementing the training of the YOLO algorithm can facilitate the rapid processing of the acquired hyperspectral images, and can improve the comparison between the processed hyperspectral images and the spectral data of the active ingredients, thereby completing the analysis of the active ingredients.

[0031] The present invention also proposes a blueberry active ingredient spectral quantitative analysis system based on deep learning, which is applicable to the above-mentioned blueberry active ingredient spectral quantitative analysis method based on deep learning, including a base. An inspection housing is installed at the upper end of the base. A lifting mechanism is jointly installed in the base and the inspection housing. A support plate member is installed on the lifting mechanism. An installation box is provided through one side of the inspection housing. A linkage mechanism is installed in the installation box. An extension rod and a mounting bracket are provided on the linkage mechanism. The extension rod is connected to the support plate member;

[0032] A camera assembly and a hyperspectral image assembly are installed on one side of the mounting bracket. A flipping mechanism is connected to one side of the mounting bracket. A sealing cover is provided on the flipping mechanism. The sealing cover is hinged to one side of the upper end of the inspection housing;

[0033] A circulation mechanism is provided on the pallet member, and a plurality of carrying pallets are provided on the circulation mechanism.

[0034] Compared with the prior art, the present system can realize shooting and hyperspectral image acquisition operations in a closed environment, can fully avoid the influence of the external environment on image acquisition, and can realize cyclic acquisition operations at the same time, improving the efficiency and quality of acquisition.

[0035] Preferably, the circulation mechanism includes a linkage belt assembly installed on the pallet member, a circulation motor assembly is installed on one side of the lower end of the pallet member, and the output shaft of the circulation motor assembly is connected to the linkage belt assembly;

[0036] A plurality of bushings are equidistantly installed on the linkage belt assembly, a vertical shaft is rotatably sleeved in the bushing, a carrying pallet is installed at the upper end of the vertical shaft, a sliding member is installed at the lower end of the vertical shaft, and the sliding member is slidably installed on the pallet member.

[0037] Further, the action of the circulation motor assembly can drive the linkage belt assembly to perform a cyclic movement. In actual operation, the linkage belt assembly is composed of a belt component and a plurality of pulley components. Teeth can be provided on the belt and the pulley to engage and ensure the linkage effect. At the same time, a plurality of pulley components can be evenly distributed on the edge of the pallet member so that the linkage belt assembly can move cyclically stably.

[0038] Preferably, the lifting mechanism includes two lead screw members penetrating through the detection housing and the base. The lower ends of the two lead screw members are both located inside the base. Synchronous wheels are fixed at the lower ends of the two lead screw members. A synchronous belt is sleeved on the two synchronous wheels. A drive motor assembly is installed inside the base, and the drive motor assembly is connected to the lower end of one of the lead screw members;

[0039] One end of the lead screw member located inside the detection housing is threadedly sleeved with a lead screw nut member, and both lead screw nut members penetrate through the pallet member.

[0040] Further, the action of the drive motor assembly can make one of the lead screw members rotate. At the same time, the synchronous wheel and the synchronous belt cooperate to complete the synchronous rotation of the two lead screw members, and the rotation of the lead screw member can make the lead screw nut member drive the pallet member to lift stably.

[0041] Preferably, the linkage mechanism includes an elastic telescopic assembly installed at the bottom inside the installation box. The upper end of the elastic telescopic assembly is fixed with an extension rod. One end of the extension rod extends to the lower end of the pallet member. A swing rod is rotatably connected between the extension rod and the installation frame.

[0042] Furthermore, by lifting the pallet member, the extended rod member can be driven to be extruded, so that the pendulum rod member can pull the mounting frame to move into the detection housing, and the camera assembly and the hyperspectral image assembly can be moved into the detection housing, facilitating the camera assembly and the hyperspectral image assembly to perform image sampling on the bilberries placed on the carrying tray.

[0043] Preferably, the flipping mechanism includes a pushing member fixed to one side of the mounting frame. One end of the pushing member penetrates through the detection housing and extends to one side of the detection housing. A slanting rod member is rotatably connected to the end of the pushing member located outside the detection housing. The upper end of the slanting rod member is rotatably connected to a connecting member, and the connecting member is fixed to one side of the sealing cover.

[0044] Furthermore, the movement of the mounting frame can drive the pushing member to move, and the movement of the pushing member can cause the slanting rod member to drive the sealing cover to flip. That is, after the pallet member descends in place, the sealing cover and the detection housing are closed to avoid external influence.

[0045] Preferably, a control component is installed on one side of the detection housing.

[0046] Furthermore, the control component can control the operation of corresponding automated components and process and analyze the collected data information.

[0047] The beneficial effects of the present invention are as follows:

[0048] 1. Corresponding programs can be set in the control component to control the operation of automated components. At the same time, it can receive and transmit the collected data information, and process and analyze the data information, enabling the analysis of the active ingredients of bilberries;

[0049] 2. It can fully process bilberries and ensure the stability of the environment during sampling, avoiding the influence of changes in the external environment on the sampling quality. At the same time, it can accurately obtain the spectral information of the active ingredients in bilberries. Then, by sampling the spectrum on the surface of bilberries and using algorithms to fully compare the sampling spectrum with the spectral information of different active ingredients for the analysis of active ingredients;

[0050] 3. It can realize the shooting and hyperspectral image acquisition operations in a closed environment, fully avoiding the influence of the external environment on image acquisition. At the same time, it can realize cyclic acquisition operations, improving the efficiency and quality of acquisition;

[0051] 4. Through the action of the circulating motor assembly, the linkage belt assembly can be driven to perform a circulating motion. In actual operation, the linkage belt assembly is composed of a belt component and multiple belt pulley components. Teeth can be provided on the belt and the belt pulleys for meshing to ensure the linkage effect. At the same time, multiple belt pulley components can be evenly distributed on the edge of the pallet member to facilitate the stable circulating movement of the linkage belt assembly;

[0052] 5. The rotation of one of the lead screw members can be achieved through the action of the drive motor assembly. Meanwhile, with the cooperation of the synchronous pulley and the synchronous belt, the synchronous rotation of the two lead screw members can be completed, and the rotation of the lead screw members can enable the lead screw nut members to drive the pallet member to lift stably.

[0053] 6. The movement of the mounting bracket can drive the pushing member to move, and the movement of the pushing member can enable the inclined rod member to drive the sealing cover to flip, that is, after the pallet member descends in place, the sealing cover and the detection housing are closed to avoid external influence. Description of the Drawings

[0054] Figure 1 It is a step diagram of the method for spectral quantitative analysis of the active ingredients of blue honeysuckle based on deep learning proposed by the present invention;

[0055] Figure 2 It is a structural diagram of the system for spectral quantitative analysis of the active ingredients of blue honeysuckle based on deep learning proposed by the present invention;

[0056] Figure 3 It is a structural diagram inside the detection housing in the system for spectral quantitative analysis of the active ingredients of blue honeysuckle based on deep learning proposed by the present invention;

[0057] Figure 4 It is an enlarged view of part A of the appendix of the present invention Figure 3 of the present invention;

[0058] In the figure: 1 base, 2 pushing member, 3 detection housing, 4 control component, 5 connecting member, 6 sealing cover, 7 inclined rod member, 8 mounting box, 9 lead screw nut member, 10 lead screw member, 11 synchronous belt, 12 drive motor assembly, 13 elastic telescopic component, 14 synchronous pulley, 15 extension rod member, 16 swing rod member, 17 mounting bracket, 18 camera component, 19 hyperspectral image component, 20 bearing tray, 21 vertical shaft, 22 bushing, 23 sliding member, 24 linkage belt component, 25 pallet member, 26 circulating motor assembly. Detailed Embodiments

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0060] Referring to Figure 1 , the method for spectral quantitative analysis of the active ingredients of blue honeysuckle based on deep learning includes the following steps:

[0061] S1. Material selection and cleaning:

[0062] Obtain mature blue honeysuckle, clean the blue honeysuckle at the same time, dry it in the air, and then screen the blue honeysuckle to select those with plump grains and no damage on the surface for use.

[0063] S2. Sampling environment setting:

[0064] Set a stable sampling environment to avoid sampling deviation caused by environmental changes; the stable sampling environment includes constant temperature, brightness, and air flow velocity.

[0065] S3. Sample sampling:

[0066] Set spectral sampling conditions, classify and sample the samples through the camera and the hyperspectral image acquisition device, group the samples and the data obtained through the camera and the hyperspectral image acquisition device at the same time, and ensure that the samples correspond to the two sampling data; the hyperspectral image acquisition device consists of a near-infrared camera GaiaField-N17, a light source, a dark box, and a loading plate; the near-infrared camera GaiaField-N17 has 256 spectral channels and can record the reflection spectra in the range of 900 - 1700 nm.

[0067] S4. Spectral determination of active ingredient samples:

[0068] Classify the active ingredients in the blue honeysuckle and obtain the spectral data information of the corresponding ingredients.

[0069] S5. Define the sample area:

[0070] Identify the blue honeysuckle area in the picture through the samples obtained by the camera, and define area 1 for comparison in the blue honeysuckle area. The samples obtained by the hyperspectral image acquisition device are consistent with the corresponding camera samples in specifications. By comparing the area of area 1 in the camera samples, area 2 for comparison can be identified in the samples obtained by the hyperspectral image acquisition device.

[0071] S6. Active ingredient analysis:

[0072] Compare the spectral data information of the active ingredients obtained in S4 with the spectral information in comparison area 2. Process comparison area 2 in S5 through the YOLO algorithm, and compare the processed data with the spectral data information of the corresponding ingredients in S4 to ensure the accuracy of data acquisition; the YOLO algorithm includes the following steps: each grid is responsible for predicting the objects whose center points fall in this area.

[0073] Each grid predicts B bounding boxes (usually 2 or 5), and each box contains:

[0074] Coordinate parameters (center point position x, y, width and height w, h)

[0075] Confidence (the probability of containing the object and the accuracy of the prediction box)

[0076] Class probability (using one-hot encoding to represent the object class)

[0077] Loss function design

[0078] Train by integrating three types of errors:

[0079] Position error: Calculate the coordinate difference between the predicted box and the ground truth box through mean squared error

[0080] Confidence error: Distinguish whether an object is included and combine cross-entropy loss

[0081] Class error: Cross-entropy loss between the predicted class and the ground truth class

[0082] Post-processing and optimization

[0083] Use non-maximum suppression (NMS) to filter redundant boxes: Keep the predicted box with the highest confidence and remove the low-confidence boxes that highly overlap with it.

[0084] Step 1: Use the Input module of the YOLOv5 model to perform adaptive scaling and data augmentation preprocessing on the input hyperspectral image, and then unify the image size through the resize() function of the Python programming language as the input of the YOLO active ingredient hyperspectral recognition model;

[0085] Step 2: Input the preprocessed hyperspectral image with annotation information into the Backbone module of the YOLOv5 model for feature extraction to obtain multi-layer feature maps;

[0086] Step 3: Input the feature maps output by the Backbone module of the YOLOv5 model into the prediction module to obtain the prediction results; The Backbone module uses the FOCUS layer to convert the information on the hyperspectral image plane to the channel dimension through slicing operations, and then extracts different features through convolutional layers; The extracted features of the Backbone module are input into the Neck module, and the Neck module further processes and integrates the features from the backbone network through feature enhancement, feature fusion, and feature processing operations; The prediction module consists of convolutional layers and fully connected layers, and is responsible for generating the recognition results of the YOLO hyperspectral image recognition model, including the bounding box position, class, and confidence of the target;

[0087] Step 4: Build a loss function based on the prediction results and continuously train and optimize the model based on the loss function.

[0088] Refer to Figures 2 - 4, the present invention also proposes a spectral quantitative analysis system for the active components of blue honeysuckle based on deep learning, which is applicable to the above-mentioned spectral quantitative analysis method for the active components of blue honeysuckle based on deep learning. It includes a base 1, and a detection housing 3 is installed at the upper end of the base 1. A control component 4 is installed on one side of the detection housing 3. Through the control component 4, the corresponding automated components can be controlled to operate, and the collected data information can be processed and analyzed.

[0089] Refer to Figures 2 - 4 , a lifting mechanism is jointly installed inside the base 1 and the detection housing 3. A tray member 25 is installed on the lifting mechanism. An installation box 8 is provided through one side of the detection housing 3. A linkage mechanism is installed inside the installation box 8. An extension rod 15 and a mounting bracket 17 are provided on the linkage mechanism. The extension rod 15 is connected to the tray member 25; by controlling the operation of the lifting mechanism and the linkage mechanism, the positions of the extension rod 15 and the mounting bracket 17 can be controlled, and the blue honeysuckle can be photographed by the camera assembly 18 and the hyperspectral image assembly 19.

[0090] Refer to Figures 2 - 4 , a camera assembly 18 and a hyperspectral image assembly 19 are installed on one side of the mounting bracket 17. A flipping mechanism is connected to one side of the mounting bracket 17. A sealing cover 6 is provided on the flipping mechanism. The sealing cover 6 is hinged to one side of the upper end of the detection housing 3; through the flipping mechanism, the sealing cover 6 can be made to contact and seal with the detection housing 3, which can ensure the stability of the sampling environment.

[0091] Refer to Figures 2 - 4 , a circulating mechanism is provided on the tray member 25, and a plurality of loading trays 20 are provided on the circulating mechanism; the circulating mechanism includes a linkage belt assembly 24 installed on the tray member 25, and a circulating motor assembly 26 is installed on one side of the lower end of the tray member 25. The output shaft of the circulating motor assembly 26 is connected to the linkage belt assembly 24;

[0092] A plurality of bushings 22 are equidistantly installed on the linkage belt assembly 24. A vertical shaft 21 is rotatably sleeved inside the bushing 22. A loading tray 20 is installed at the upper end of the vertical shaft 21. A sliding member 23 is installed at the lower end of the vertical shaft 21. The sliding member 23 is slidably installed on the tray member 25; through the action of the circulating motor assembly 26, the linkage belt assembly 24 can be driven to perform a circulating movement. In actual operation, the linkage belt assembly 24 is composed of a belt component and a plurality of pulley components. Teeth can be provided on the belt and the pulleys for meshing to ensure the linkage effect. At the same time, a plurality of pulley components can be evenly distributed on the edge of the tray member 25 so that the linkage belt assembly 24 can circulate stably.

[0093] Refer to Figures 2 - 4, the lifting mechanism includes two lead screw members 10 penetrating through the detection housing 3 and the base 1. The lower ends of the two lead screw members 10 are both located inside the base 1. Synchronous wheels 14 are fixed to the lower ends of the two lead screw members 10. A synchronous belt 11 is sleeved on the two synchronous wheels 14 together. A drive motor assembly 12 is installed inside the base 1, and the drive motor assembly 12 is connected to the lower end of one of the lead screw members 10. One end of the lead screw member 10 located inside the detection housing 3 is threadedly sleeved with a lead screw nut member 9, and the two lead screw nut members 9 both penetrate through the support plate member 25. Through the action of the drive motor assembly 12, one of the lead screw members 10 can be rotated. At the same time, with the cooperation of the synchronous wheels 14 and the synchronous belt 11, the synchronous rotation of the two lead screw members 10 can be completed, and through the rotation of the lead screw member 10, the lead screw nut member 9 can drive the support plate member 25 to lift stably.

[0094] Refer to Figures 2 - 4 , the linkage mechanism includes an elastic telescopic component 13 installed at the bottom inside the installation box 8. The upper end of the elastic telescopic component 13 is fixed with an extension rod 15. One end of the extension rod 15 extends to the lower end of the support plate member 25. A swing rod 16 is rotatably connected between the extension rod 15 and the installation frame 17. Through the lifting of the support plate member 25, the extension rod 15 can be driven to be extruded, so that the swing rod 16 can pull the installation frame 17 to move into the detection housing 3, and the camera component 18 and the hyperspectral image component 19 can be moved into the detection housing 3, facilitating the camera component 18 and the hyperspectral image component 19 to perform image sampling on the blue honeysuckle placed on the bearing tray 20.

[0095] Refer to Figures 2 - 4 , the flipping mechanism includes a pushing member 2 fixed to one side of the installation frame 17. One end of the pushing member 2 penetrates through the detection housing 3 and extends to one side of the detection housing 3. One end of the pushing member 2 located outside the detection housing 3 is rotatably connected with an inclined rod 7. The upper end of the inclined rod 7 is rotatably connected with a connecting member 5, and the connecting member 5 is fixed to one side of the sealing cover 6. Through the movement of the installation frame 17, the pushing member 2 can be driven to move. The movement of the pushing member 2 can enable the inclined rod 7 to drive the sealing cover 6 to flip, that is, after the support plate member 25 descends in place, the sealing cover 6 and the detection housing 3 are closed to avoid external influence.

[0096] In the present invention, the staff can place the blue honeysuckle in the carrying tray 20. When the pallet member 25 descends, it can squeeze the extension rod member 15 through the pallet member 25 so that the swing rod member 16 pulls the mounting bracket 17 to move the camera assembly 18 and the hyperspectral image assembly 19 out, facilitating the shooting and sampling of the blue honeysuckle moving sequentially from its lower end, controlling the shooting conditions of the camera assembly 18 and the hyperspectral image assembly 19 to make the specifications of the captured images the same, so as to ensure that the blue honeysuckle is in the same position in the pictures, determining the skin condition of the blue honeysuckle through the captured images of the camera assembly 18, and further clarifying the sampling range in the pictures, so as to clarify the pictures obtained in the hyperspectral image assembly 19 to frame the analysis range;

[0097] The collected data information is transmitted and can be effectively processed through the control component 4. The processing effect on small objects can be improved through the YOLOv5 model algorithm. After selecting the corresponding sampling range, it can be divided into a grid of S×S, and then each grid is responsible for predicting the situation where the center point falls in this area, and the analysis of the active ingredients is achieved through multi-data comparison.

[0098] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent replacements or changes should be covered within the protection scope of the present invention.

Claims

1. A spectral quantitative analysis method for the active ingredients of blue honeysuckle based on deep learning, characterized in that, It includes the following steps: S1. Material selection and cleaning: Obtain mature blue honeysuckle fruits, clean the blue honeysuckle fruits at the same time, dry them, and then screen the blue honeysuckle fruits to select those with plump grains and no damage on the surface for use; S2. Sampling environment setting: Set a stable sampling environment to avoid sampling deviation caused by environmental changes; S3. Sample sampling: Set spectral sampling conditions, classify and sample the samples through a camera and a hyperspectral image acquisition device, group the samples and the data obtained through the camera and the hyperspectral image acquisition device at the same time, and ensure that the samples correspond to the two sampling data; S4. Spectral determination of active ingredient samples: Classify the active ingredients in the blue honeysuckle fruits and obtain the spectral data information of the corresponding ingredients; S5. Define the sample area: Identify the blue honeysuckle fruit area in the picture through the samples obtained by the camera, and define the comparison area 1 in the blue honeysuckle fruit area. The samples obtained by the hyperspectral image acquisition device are consistent with the corresponding camera sample specifications. The comparison area 1 in the camera sample can be used to identify the comparison area 2 in the samples obtained by the hyperspectral image acquisition device; S6. Active ingredient analysis: Compare the spectral data information of the active ingredients obtained in S4 with the spectral information in the comparison area 2, process the comparison area 2 in S5 through the YOLO algorithm, and compare the processed data with the spectral data information of the corresponding ingredients in S4 to ensure the accuracy of data acquisition.

2. The spectral quantitative analysis method of the active ingredients of blue honeysuckle based on deep learning according to claim 1, characterized in that: The stable sampling environment in S2 includes the constancy of temperature, brightness, and air flow velocity.

3. The spectral quantitative analysis method of the active ingredients of blue honeysuckle based on deep learning according to claim 1, characterized in that: The hyperspectral image acquisition device in S3 consists of a near-infrared camera GaiaField-N17, a light source, a dark box, and a loading plate; the near-infrared camera GaiaField-N17 has 256 spectral channels and can record the reflection spectra in the range of 900 - 1700 nm.

4. The spectral quantitative analysis method for active components of blue honeysuckle based on deep learning according to claim 1, characterized in that: The YOLO algorithm in S6 includes the following steps: Step 1: Use the Input module of the YOLOv5 model to perform adaptive scaling and data augmentation preprocessing on the input hyperspectral image, and then unify the image size through the resize() function of the python programming language as the input of the YOLO active ingredient hyperspectral recognition model; Step 2: Input the preprocessed hyperspectral image with annotation information into the Backbone module of the YOLOv5 model for feature extraction to obtain multi-layer feature maps; Step 3: The feature map output by the Backbone module of the YOLOv5 model is input into the prediction module to obtain the prediction results. The Backbone module uses a FOCUS layer to convert the information on the hyperspectral image plane to the channel dimension through slicing operations, and then extracts different features through convolutional layers. The features extracted by the Backbone module are input into the Neck module, which further processes and integrates the features from the backbone network through feature enhancement, feature fusion, and feature processing operations. The prediction module consists of convolutional layers and fully connected layers and is responsible for generating the recognition results of the YOLO hyperspectral image recognition model, including the bounding box position, category, and confidence of the target. Step 4: Construct a loss function based on the prediction results, and continuously train and optimize the model based on the loss function.

5. A spectroscopic quantitative analysis system for active components of Lonicera caerulea L. based on deep learning, applicable to the spectroscopic quantitative analysis method for active components of Lonicera caerulea L. based on deep learning according to any one of the above claims 1-4, comprising a base (1), characterized in that: A detection housing (3) is installed at the upper end of the base (1). A lifting mechanism is installed in the base (1) and the detection housing (3) together. A tray member (25) is installed on the lifting mechanism. An installation box (8) penetrates through one side of the detection housing (3). A linkage mechanism is installed in the installation box (8). An extension rod member (15) and a mounting frame (17) are provided on the linkage mechanism. The extension rod member (15) is connected to the tray member (25). A camera assembly (18) and a hyperspectral image assembly (19) are installed on one side of the mounting frame (17). A flipping mechanism is connected to one side of the mounting frame (17). A sealing cover (6) is provided on the flipping mechanism. The sealing cover (6) is hinged to the upper end side of the detection housing (3). A circulation mechanism is provided on the tray member (25). A plurality of carrier trays (20) are provided on the circulation mechanism.

6. The spectral quantitative analysis system for the active ingredients of blue honeysuckle based on deep learning according to claim 5, characterized in that: The circulation mechanism includes a linkage belt assembly (24) installed on the tray member (25). A circulation motor assembly (26) is installed on the lower end side of the tray member (25). The output shaft of the circulation motor assembly (26) is connected to the linkage belt assembly (24). A plurality of bushings (22) are equidistantly installed on the linkage belt assembly (24). A vertical shaft (21) is rotatably sleeved in the bushing (22). A carrier tray (20) is installed at the upper end of the vertical shaft (21). A sliding member (23) is installed at the lower end of the vertical shaft (21). The sliding member (23) is slidably installed on the tray member (25).

7. The spectral quantitative analysis system for the active ingredients of blue honeysuckle based on deep learning according to claim 5, wherein: The lifting mechanism includes two lead screw members (10) penetrating through the detection housing (3) and the base (1). The lower ends of the two lead screw members (10) are both located in the base (1). Synchronous wheels (14) are fixed to the lower ends of the two lead screw members (10). A synchronous belt (11) is sleeved on the two synchronous wheels (14) together. A drive motor assembly (12) is installed in the base (1). The drive motor assembly (12) is connected to the lower end of one of the lead screw members (10). One end of the lead screw member (10) located in the detection housing (3) is threadedly sleeved with a lead screw nut member (9). The two lead screw nut members (9) both penetrate through the tray member (25).

8. The spectral quantitative analysis system for the active ingredients of Lonicera caerulea L. based on deep learning according to claim 5, wherein: The linkage mechanism includes an elastic telescopic component (13) installed at the inner bottom of the installation box (8). The upper end of the elastic telescopic component (13) is fixed with an extension rod (15). One end of the extension rod (15) extends to the lower end of the support plate member (25). A swing rod (16) is rotatably connected between the extension rod (15) and the installation frame (17).

9. The spectral quantitative analysis system for the active ingredients of Lonicera caerulea based on deep learning according to claim 5, characterized in that: The flipping mechanism includes a pushing member (2) fixed to one side of the installation frame (17). One end of the pushing member (2) penetrates through the detection housing (3) and extends to one side of the detection housing (3). An inclined rod (7) is rotatably connected to the end of the pushing member (2) located outside the detection housing (3). The upper end of the inclined rod (7) is rotatably connected to a connecting member (5). The connecting member (5) is fixed to one side of the sealing cover (6).

10. The spectral quantitative analysis system for active components of Lonicera caerulea L. based on deep learning according to claim 5, characterized in that: A control component (4) is installed on one side of the detection housing (3).

Images