A digital twin grain storage system and method integrating detection and adjustment based on hyperspectral shortwave infrared.
By using a hyperspectral shortwave infrared detection-adjustment integrated digital twin grain storage system, combined with hyperspectral grain storage lidar and LoRa IoT, rapid monitoring and location of insects, molds and volatiles in grain silos have been achieved, solving the problem of delayed grain condition detection and control, and improving the accuracy of grain condition assessment and the grain storage cycle.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF INFORMATION SCI & TECH
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-30
AI Technical Summary
Existing technologies face challenges in monitoring the location of insects and molds, interference from impurities such as rice bran in the grain storage environment, and delays in grain condition detection and control. Hyperspectral imaging technology cannot achieve real-time detection and positioning within the warehouse.
The integrated digital twin grain storage system employs hyperspectral shortwave infrared detection and adjustment, combining a hyperspectral grain storage lidar front-end device, a three-dimensional grain condition twin, LoRa IoT, and back-end execution devices. Through digital twin modeling, it enables rapid monitoring and location of insect and mold volatile gases, and allows for real-time control.
It achieves high-precision monitoring of insect and mold volatile gases, has strong anti-interference ability, accurate grain condition assessment, timely regulation, flexible multi-dimensional grain storage methods, significantly shortens the regulation response time, extends the grain storage period, and maintains stable grain quality.
Smart Images
Figure CN122306735A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of digital twins, specifically relating to a detection-adjustment integrated digital twin grain storage system based on hyperspectral shortwave infrared. Background Technology
[0002] Currently, the digital transformation of grain storage faces challenges such as difficulties in monitoring the location of insects and mold, interference from impurities like rice bran in the grain storage environment, and delays in grain condition detection and control. This is because insect and mold monitoring often relies on sensors and image recognition technology, but the internal environment of grain silos is complex, and factors such as temperature and humidity changes and insufficient lighting can affect the accuracy and stability of monitoring equipment. Impurities such as rice bran mainly originate from mechanical damage and shedding during grain harvesting, processing, and storage. These impurities easily accumulate in grain silos, forming sources of interference. Furthermore, the current equipment's untimely data processing or limited analytical capabilities lead to delays in grain condition detection and control.
[0003] Currently, hyperspectral imaging technology has been widely used in the detection of insect and mold volatiles and the quality rating of rice, but it cannot achieve real-time detection and positioning within the warehouse, and cannot effectively solve the above problems. Summary of the Invention
[0004] This invention addresses the problems of existing technologies, such as difficulty in monitoring the location of insects and molds, interference from impurities such as chaff in the grain storage environment, and lag in grain condition detection and control. It provides an integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and control. Compared with solutions using traditional hyperspectral imaging technology, it can more quickly monitor and locate volatile gases from insects and molds, and achieve real-time control of grain conditions through digital twin modeling.
[0005] To solve the above technical problems, the present invention provides the following technical solution: a digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment, comprising: a hyperspectral grain storage lidar front-end device, a three-dimensional grain condition twin, a LoRa Internet of Things, and a back-end execution device, wherein the back-end execution device comprises: a temperature-controlled grain storage device and a controlled atmosphere grain storage device.
[0006] A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is constructed using a grain condition database. Digital twin modeling is achieved using Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. Interactive control with the front-end detection device and the back-end execution device is achieved through LoRa IoT, realizing the synchronous fusion and remote feature extraction of data such as inversion map, temperature and humidity, and gas composition.
[0007] The three-dimensional grain condition twin is used to train a fully connected network on the spectral data signal of the same insect and mold volatiles. After multi-layer learning, it outputs the feature vectors of volatile substances, fumigation gases, and storage gases. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using the fully connected neural network and the trained grain condition dataset to output the grain condition database.
[0008] After receiving feedback, the temperature-controlled grain storage device automatically adjusts the temperature inside the storage chamber; the atmosphere-controlled grain storage device automatically adjusts the proportion of gas components inside the storage chamber based on the current grain condition assessment results, in order to inhibit insects and mold, extend the storage period, and maintain stable grain quality, thus constructing a detection-modeling-feedback closed loop.
[0009] Furthermore, the aforementioned hyperspectral grain storage lidar front-end device includes: a supercontinuum laser, a speckle optical path, a grating beam splitter, and a staring short-wave infrared imaging module connected in sequence, used to achieve rapid monitoring and positioning of insect and mold volatile gases, thus solving the problem of interference from bran impurities in the grain storage environment.
[0010] Among them, the supercontinuum laser emits lasers of various wavelengths, intensities, and spectral characteristics required for detection; the speckle optical path provides the path required for laser detection; the grating beam splitter decomposes the laser information of the required wavelengths; and the staring short-wave infrared imaging module collects point cloud data and spectral information related to volatile gases from insects and molds.
[0011] Furthermore, the aforementioned hyperspectral grain storage lidar front-end device has a ranging accuracy of ±3cm, a horizontal field of view of 360° and an angular resolution between 0.1° and 0.4°, a vertical field of view of 30° and a vertical resolution of 2°, and is used to accurately perform all-round trace gas scanning and tracking in grain depots.
[0012] Furthermore, the aforementioned dynamic grain condition prediction model is constructed based on point cloud data and spectral information related to insect and mold volatile gases. A deep model is constructed by extracting feature values from the spectral information, wherein the spectral signal is measured through multiple channels, and the dynamic grain condition prediction results are output.
[0013] Furthermore, the aforementioned detection-modeling-feedback closed-loop interactive control process is as follows:
[0014] A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is constructed using a grain condition database. Digital twin modeling is achieved with the help of Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. The process of rice storage is then virtually simulated, dynamically predicted, and visualized.
[0015] By leveraging LoRa IoT to achieve interactive control with front-end detection and back-end execution devices, the system enables synchronous fusion and remote feature extraction of data such as inversion maps, temperature, humidity, and gas composition. When abnormalities in insects and mold or imbalances in the grain storage environment are detected, the system automatically generates control strategies and uses LoRa IoT technology to link temperature-controlled and atmosphere-controlled grain storage devices in real time to regulate the grain condition. This constructs a detection-modeling-feedback closed loop, enabling the selection of multi-dimensional grain storage methods and mitigating the aging and deterioration of japonica rice.
[0016] This invention also provides a method for monitoring and controlling grain storage using a hyperspectral shortwave infrared integrated digital twin based on detection and adjustment, which is applied to the monitoring and control of grain storage using a hyperspectral shortwave infrared integrated digital twin based on detection and adjustment, and includes the following steps:
[0017] S1. A hyperspectral grain storage lidar front-end device is adopted to achieve rapid monitoring and location of volatile gases from insects and molds;
[0018] S2. Using a three-dimensional grain condition twin, a fully connected network is used to train the spectral data signal of the same insect and mold volatiles. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using a fully connected neural network and the trained grain condition dataset to output the grain condition database.
[0019] S3. A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from a hyperspectral grain storage lidar front-end. A spectral-grain condition inversion map is built using a grain condition database, and digital twin modeling is achieved using Unity 3D. Combined with multi-source parameters such as storage temperature and humidity, and gas concentration, a dynamic grain condition prediction model is established to perform virtual simulation, dynamic prediction, and visualization of the rice storage process. Interactive control with the front-end detection and back-end execution devices is achieved through LoRa IoT, enabling synchronous fusion and remote feature extraction of data such as inversion maps, temperature and humidity, and gas composition. When abnormal insect and mold growth or imbalance in the grain storage environment is detected, an automatic control strategy is generated, and the temperature-controlled grain storage device and atmosphere-controlled grain storage device are linked in real time through LoRa IoT technology to control the grain condition in real time, constructing a detection-modeling-feedback closed loop.
[0020] S4. Based on the gas chromatography-mass spectrometry analysis results, the algorithm is improved and upgraded using a comparative training group.
[0021] Furthermore, the aforementioned step S2 includes the following sub-steps:
[0022] S201. Based on a short-wave infrared detection array, and in conjunction with a laser speckle signal as a light source, the laser speckle short-wave infrared signal of volatiles is collected.
[0023] S202. Volatile spectral data were generated using the relationship graph between insects, molds and volatiles, and a fully connected model was constructed using a fully connected network and a CNN neural network.
[0024] S203. The spectral data signal is trained using a fully connected network, and the co-occurrence matrix is trained using a CNN neural network. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The volatile inversion spectrum is output through the fully connected neural network. Xavier initialization is used to initialize the network weights and biases. The cross-entropy loss function is used to measure the difference between the model's predicted values and the true values. Finally, gradient descent is used to update the network weights and biases to minimize the loss function. Through multiple iterations of training, the network's predicted values are made closer to the true values.
[0025] Furthermore, the aforementioned cross-entropy formula is as follows:
[0026]
[0027] The formula for calculating the loss function using cross-entropy is as follows:
[0028]
[0029] Where m represents the number of samples, n represents the total number of categories in the classification task, P(x) represents the true distribution of the samples, and Q(x) represents the distribution distance predicted by the model.
[0030] Furthermore, the aforementioned step S3 includes the following sub-steps:
[0031] S301. The construction of the dynamic prediction model revolves around the point cloud data and spectral information related to volatile gases of insects and molds. A deep model is constructed by extracting feature values from the spectral information, where the spectral signal can be measured through multiple channels.
[0032] S302 The system is connected to nitrogen-filled storage, new cold storage and dry storage back-end execution devices. When the system detects insects and mold, it will immediately make on-site adjustments in the storage environment and realize the selection of multi-dimensional grain storage methods, effectively slowing down the aging and deterioration of japonica rice.
[0033] Compared with the prior art, the beneficial technical effects of the present invention using the above technical solution are as follows:
[0034] 1. High detection accuracy and strong anti-interference capability: The front-end device of hyperspectral grain storage lidar is adopted. By combining a supercontinuum laser with a speckle optical path and a staring short-wave infrared imaging module, trace gas detection of 0.2 ppm is achieved. It effectively overcomes the interference of impurities such as rice bran in the grain storage environment, and can quickly locate insect and mold volatiles, solving the problem that traditional monitoring methods are easily affected by environmental factors.
[0035] 2. Intelligent modeling and accurate prediction: By constructing a three-dimensional grain condition twin, integrating the spectral-grain condition inversion map with multi-source parameters such as storage temperature and humidity and gas concentration, and using fully connected neural networks and CNNs to establish a dynamic grain condition prediction model, the virtual simulation and dynamic prediction of the rice storage process are realized, improving the accuracy and interpretability of grain condition assessment.
[0036] 3. Timely regulation and closed-loop system: Based on LoRa IoT, the system enables real-time interaction between front-end detection and back-end execution devices. When abnormalities such as insects and mold or imbalances in the grain storage environment are detected, the system automatically generates regulation strategies and links temperature-controlled and atmosphere-controlled grain storage devices for in-situ regulation. This constructs an integrated closed-loop system of "detection-modeling-feedback", which significantly shortens the regulation response time.
[0037] 4. Diverse grain storage methods and significant loss reduction: The system can be connected to various back-end execution devices such as nitrogen-filled storage, new cold storage and dry storage, supporting flexible selection of multi-dimensional grain storage methods, effectively inhibiting the growth of insects and molds, delaying the aging and deterioration of japonica rice, extending the grain storage period, and providing reliable technical support for saving grain and reducing losses and stabilizing grain quality. Attached Figure Description
[0038] Figure 1 This is an overall structural diagram of the system of the present invention.
[0039] Figure 2 This is a schematic diagram of the structure of the hyperspectral grain storage lidar front-end device used in this invention.
[0040] Figure 3 It is a graph showing the relationship between insect molds and their volatiles, as well as their inversion graph.
[0041] Figure 4 It utilizes staring shortwave infrared imaging technology to locate insects and molds and monitor volatile gases.
[0042] Figure 5 This is a diagram showing the interface of the three-dimensional grain condition twin of this invention. Detailed Implementation
[0043] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.
[0044] In this invention, various aspects of the invention are described with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. Embodiments of the invention are not limited to those depicted in the drawings. It should be understood that the invention is implemented through any of the various concepts and embodiments described above, as well as the concepts and embodiments described in detail below, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.
[0045] The embodiment provides an integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment, such as... Figure 1 As shown, it includes: a hyperspectral grain storage lidar front-end device, a three-dimensional grain condition twin, a LoRa Internet of Things, and a back-end execution device. The back-end execution device includes: a temperature-controlled grain storage device and a controlled atmosphere grain storage device.
[0046] Figure 2 The structure of the hyperspectral grain storage lidar front-end device is shown. The hyperspectral grain storage lidar front-end device takes advantage of the good penetration and strong anti-interference of laser. By penetrating the rice pile, it collects the infrared reflected laser signal and collects the laser speckle infrared characteristics, realizing the detection of trace gases at the 0.2 ppm level in the grain depot.
[0047] A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is built using a grain condition database, and digital twin modeling is achieved with the help of Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. Interactive control with the front-end detection device and the back-end execution device is achieved through LoRa Internet of Things, realizing the synchronous fusion and remote feature extraction of data such as inversion map, temperature and humidity, and gas composition.
[0048] The hyperspectral grain storage lidar front-end device comprises: a supercontinuum laser (900–2500 nm), a speckle optical path, a grating beam splitter (900–2500 nm), and a staring short-wave infrared imaging module (900–2200 nm), connected sequentially. This front-end device enables rapid monitoring and location of volatile gases from insects and molds, solving the interference problem caused by impurities such as rice bran in the grain storage environment. The supercontinuum laser emits lasers of various wavelengths, intensities, and spectral characteristics required for detection; the speckle optical path provides the path required for laser detection; the grating beam splitter decomposes the laser information of the required wavelengths; and the staring short-wave infrared imaging module collects point cloud data and spectral information related to volatile gases from insects and molds. The hyperspectral grain storage lidar front-end device has a ranging accuracy of ±3 cm, a horizontal field of view of 360° with an angular resolution between 0.1° and 0.4°, and a vertical field of view of 30° with a vertical resolution of 2°, enabling precise omnidirectional scanning and tracking of trace gases in grain depots. Meanwhile, this lidar has strong environmental adaptability and can operate normally under different lighting conditions and temperature ranges.
[0049] A three-dimensional grain condition twin is used to train a fully connected network on the spectral data signal of the same insect and mold volatiles. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using a fully connected neural network and the trained grain condition dataset to output the grain condition database.
[0050] During system operation, the speckle optical path expands the laser beam emitted by the supercontinuum laser (900–2500 nm) and scans the grain silo. The laser echo signal is collected by adjusting the grating beam splitter and the staring short-wave infrared imaging module lens, enabling comprehensive trace gas scanning and tracking of the grain silo. After detection and identification by the hyperspectral grain storage lidar front-end, a three-dimensional grain condition twin is constructed using point cloud data and spectral information collected by the hyperspectral grain storage lidar front-end. A spectral-grain condition inversion map is built using a grain condition database, and digital twin modeling is achieved using Unity 3D. Combined with multi-source parameters such as storage temperature and humidity, and gas concentration, a dynamic grain condition prediction model is established. Interactive control with the front-end detection and back-end execution devices is achieved through LoRa IoT, enabling synchronous fusion and remote feature extraction of data such as the inversion map, temperature and humidity, and gas composition.
[0051] After receiving feedback, the temperature-controlled grain storage device automatically adjusts the temperature inside the storage chamber; the atmosphere-controlled grain storage device automatically adjusts the proportion of gas components inside the storage chamber based on the current grain condition assessment results, in order to inhibit insects and mold, extend the storage period, and maintain stable grain quality, thus constructing a detection-modeling-feedback closed loop.
[0052] This invention also provides a method for integrated detection and control of grain storage using a digital twin based on hyperspectral shortwave infrared, comprising the following steps:
[0053] S1. A hyperspectral grain storage lidar front-end device is adopted to achieve rapid monitoring and location of volatile gases from insects and molds;
[0054] S2. Using a 3D grain condition twin, a fully connected network is used to train the spectral data signals of the same insect and mold volatiles. After multi-layer learning, feature vectors of volatile substances, fumigation gases, and storage gases are output. The database corresponding to the insect and mold volatiles-spectral data signals is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using a fully connected neural network and the trained grain condition dataset to output the grain condition database. Specifically, the following sub-steps are included:
[0055] S201. Based on a short-wave infrared detection array, and in conjunction with a laser speckle signal as a light source, the laser speckle short-wave infrared signal of volatiles is collected.
[0056] S202. Volatile spectral data were generated using the relationship graph between insects, molds and volatiles, and a fully connected model was constructed using a fully connected network and a CNN neural network.
[0057] S203. The spectral data signal is trained using a fully connected network, and the co-occurrence matrix is trained using a CNN neural network. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The volatile inversion spectrum is output through the fully connected neural network. Xavier initialization is used to initialize the network weights and biases. The cross-entropy loss function is used to measure the difference between the model's predicted values and the true values. Finally, gradient descent is used to update the network weights and biases to minimize the loss function. Through multiple iterations of training, the network's predicted values are made closer to the true values.
[0058] S3. A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from a hyperspectral grain storage lidar front-end. A spectral-grain condition inversion map is built using a grain condition database, and digital twin modeling is achieved using Unity 3D. Combining multi-source parameters such as storage temperature and humidity, and gas concentration, a dynamic grain condition prediction model is established to perform virtual simulation, dynamic prediction, and visualization of the rice storage process. Interactive control with the front-end detection and back-end execution devices is achieved through LoRa IoT, enabling synchronous fusion and remote feature extraction of inversion map, temperature and humidity, and gas composition data. When abnormal insect and mold growth or imbalances in the grain storage environment are detected, an automatic control strategy is generated, and the temperature-controlled and atmosphere-controlled grain storage devices are linked in real-time via LoRa IoT technology to regulate the grain condition, constructing a detection-modeling-feedback closed loop. Specifically, this includes the following sub-steps:
[0059] S301. The construction of the dynamic prediction model revolves around the point cloud data and spectral information related to insect and mold volatile gases. A deep model is constructed by extracting feature values from the spectral information. The spectral signal can be measured through multiple channels. Since volatile substances, fumigation gases, and storage gases have peaks in their respective specific bands, and their concentrations also have a strong influence on them, the bands in which the peaks appear and the intensity of the spectral information represent the corresponding characteristics.
[0060] S302 The system is connected to nitrogen-filled storage, new cold storage and dry storage back-end execution devices. When the system detects insects and mold, it will immediately make on-site adjustments in the storage environment and realize the selection of multi-dimensional grain storage methods (more than 50 grain warehouse terminals are connected), which effectively slows down the aging and deterioration of japonica rice.
[0061] The grain condition dynamic prediction model is constructed based on point cloud data and spectral information related to insect and mold volatile gases. A deep model is built by extracting feature values from the spectral information. The spectral signal is measured through multiple channels, and the grain condition dynamic prediction result is output. Since volatile substances, fumigation gases, and storage gases have peaks in their respective specific wavelength bands, and their concentrations also have a strong influence on them, the wavelength bands where peaks appear and the intensity of spectral information represent the corresponding characteristics.
[0062] S4. Based on the gas chromatography-mass spectrometry analysis results, the algorithm is improved and upgraded using a comparative training group.
[0063] In this embodiment, the detection-modeling-feedback closed-loop interactive control process is as follows:
[0064] A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is constructed using a grain condition database. Digital twin modeling is achieved with the help of Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. The process of rice storage is then virtually simulated, dynamically predicted, and visualized.
[0065] By leveraging LoRa IoT to achieve interactive control with front-end detection and back-end execution devices, the system enables synchronous fusion and remote feature extraction of data such as inversion maps, temperature, humidity, and gas composition. When abnormalities in insects and mold or imbalances in the grain storage environment are detected, the system automatically generates control strategies and uses LoRa IoT technology to link temperature-controlled and atmosphere-controlled grain storage devices in real time to regulate the grain condition. This constructs a detection-modeling-feedback closed loop, enabling the selection of multi-dimensional grain storage methods and mitigating the aging and deterioration of japonica rice.
[0066] like Figure 3 As shown, the three-dimensional grain condition twin uses a fully connected network to train the spectral data signal of the same insect and mold volatiles. After multi-layer learning, it outputs feature vectors of volatile substances, fumigation gases, and storage gases. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using the fully connected neural network and the trained grain condition dataset to output the grain condition database.
[0067] like Figure 4 As shown, staring shortwave infrared imaging technology is used to locate insects and molds and monitor volatile gases.
[0068] like Figure 5 As shown, the three-dimensional grain condition twin constructs a spectral-grain condition inversion map using a grain condition database, and uses Unity 3D to achieve digital twin modeling. Combining multi-source parameters such as storage temperature and humidity, gas concentration, etc., a dynamic grain condition prediction model is established to perform virtual simulation, dynamic prediction and visualization of the rice storage process.
[0069] While the present invention has been described above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.
Claims
1. A digital twin grain storage system integrating detection and adjustment based on hyperspectral shortwave infrared, characterized in that, include: The system includes a hyperspectral grain storage lidar front-end device, a three-dimensional grain condition twin, a LoRa Internet of Things, and a back-end execution device. The back-end execution device includes a temperature-controlled grain storage device and a controlled atmosphere grain storage device. A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is constructed using a grain condition database. Digital twin modeling is achieved using Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. Interactive control with the front-end detection device and the back-end execution device is achieved through LoRa IoT, realizing the synchronous fusion and remote feature extraction of data such as inversion map, temperature and humidity, and gas composition. The three-dimensional grain condition twin is used to train a fully connected network on the spectral data signal of the same insect and mold volatiles. After multi-layer learning, it outputs the feature vectors of volatile substances, fumigation gases, and storage gases. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using the fully connected neural network and the trained grain condition dataset to output the grain condition database. After receiving feedback, the temperature-controlled grain storage device automatically adjusts the temperature inside the storage chamber; the atmosphere-controlled grain storage device automatically adjusts the proportion of gas components inside the storage chamber based on the current grain condition assessment results, in order to inhibit insects and mold, extend the storage period, and maintain stable grain quality, thus constructing a detection-modeling-feedback closed loop.
2. The integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment as described in claim 1, characterized in that, The hyperspectral grain storage lidar front-end device includes: a supercontinuum laser, a speckle optical path, a grating beam splitter, and a staring short-wave infrared imaging module connected in sequence, which is used to realize the rapid monitoring and positioning of insect and mold volatile gases, and solves the problem of interference from bran impurities in the grain storage environment. Among them, the supercontinuum laser emits lasers of various wavelengths, intensities, and spectral characteristics required for detection; the speckle optical path provides the path required for laser detection; the grating beam splitter decomposes the laser information of the required wavelengths; and the staring short-wave infrared imaging module collects point cloud data and spectral information related to volatile gases from insects and molds.
3. The integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment as described in claim 1, characterized in that, The hyperspectral grain storage lidar front-end device has a ranging accuracy of ±3cm, a horizontal field of view of 360° and an angular resolution between 0.1° and 0.4°, a vertical field of view of 30° and a vertical resolution of 2°, and is used to accurately perform all-round trace gas scanning and tracking in grain depots.
4. The integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment as described in claim 1, characterized in that, The grain condition dynamic prediction model is constructed based on point cloud data and spectral information related to insect and mold volatile gases. A deep model is constructed by extracting feature values from the spectral information. The spectral signal is measured through multiple channels, and the grain condition dynamic prediction result is output.
5. The integrated digital twin grain storage system based on hyperspectral shortwave infrared detection and adjustment as described in claim 1, characterized in that, The detection-modeling-feedback closed-loop interactive control process is as follows: A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from the front end of a hyperspectral grain storage lidar. A spectral-grain condition inversion map is constructed using a grain condition database. Digital twin modeling is achieved with the help of Unity 3D. A dynamic grain condition prediction model is established by combining multi-source parameters such as storage temperature and humidity and gas concentration. The process of rice storage is then virtually simulated, dynamically predicted, and visualized. By leveraging LoRa IoT to achieve interactive control with front-end detection and back-end execution devices, the system enables synchronous fusion and remote feature extraction of data such as inversion maps, temperature, humidity, and gas composition. When abnormalities in insects and mold or imbalances in the grain storage environment are detected, the system automatically generates control strategies and uses LoRa IoT technology to link temperature-controlled and atmosphere-controlled grain storage devices in real time to regulate the grain condition. This constructs a detection-modeling-feedback closed loop, enabling the selection of multi-dimensional grain storage methods and mitigating the aging and deterioration of japonica rice.
6. A method for monitoring and controlling grain storage using a hyperspectral shortwave infrared integrated digital twin system, applied to the hyperspectral shortwave infrared integrated digital twin system for grain storage as described in any one of claims 1-3, characterized in that... Includes the following steps: S1. A hyperspectral grain storage lidar front-end device is adopted to achieve rapid monitoring and location of volatile gases from insects and molds; S2. Using a three-dimensional grain condition twin, a fully connected network is used to train the spectral data signal of the same insect and mold volatiles. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The database corresponding to the insect and mold volatiles-spectral data signal is used as the trained grain condition dataset. The spectral-grain condition inversion map is constructed by using a fully connected neural network and the trained grain condition dataset to output the grain condition database. S3. A three-dimensional grain condition twin is constructed by collecting point cloud data and spectral information from a hyperspectral grain storage lidar front-end. A spectral-grain condition inversion map is built using a grain condition database, and digital twin modeling is achieved using Unity 3D. Combined with multi-source parameters such as storage temperature and humidity, and gas concentration, a dynamic grain condition prediction model is established to perform virtual simulation, dynamic prediction, and visualization of the rice storage process. Interactive control with the front-end detection and back-end execution devices is achieved through LoRa IoT, enabling synchronous fusion and remote feature extraction of data such as inversion maps, temperature and humidity, and gas composition. When abnormal insect and mold growth or imbalance in the grain storage environment is detected, an automatic control strategy is generated, and the temperature-controlled grain storage device and atmosphere-controlled grain storage device are linked in real time through LoRa IoT technology to control the grain condition in real time, constructing a detection-modeling-feedback closed loop. S4. Based on the gas chromatography-mass spectrometry analysis results, the algorithm is improved and upgraded using a comparative training group.
7. The integrated digital twin grain storage monitoring and control method based on hyperspectral shortwave infrared as described in claim 6, characterized in that, Step S2 includes the following sub-steps: S201. Based on a short-wave infrared detection array, and in conjunction with a laser speckle signal as a light source, the laser speckle short-wave infrared signal of volatiles is collected. S202. Volatile spectral data were generated using the relationship graph between insects, molds and volatiles, and a fully connected model was constructed using a fully connected network and a CNN neural network. S203. The spectral data signal is trained using a fully connected network, and the co-occurrence matrix is trained using a CNN neural network. After multi-layer learning, the feature vectors of volatile substances, fumigation gases, and storage gases are output. The volatile inversion spectrum is output through the fully connected neural network. Xavier initialization is used to initialize the network weights and biases. The cross-entropy loss function is used to measure the difference between the model's predicted values and the true values. Finally, gradient descent is used to update the network weights and biases to minimize the loss function. Through multiple iterations of training, the network's predicted values are made closer to the true values.
8. The integrated digital twin grain storage monitoring and control method based on hyperspectral shortwave infrared detection and adjustment as described in claim 6, characterized in that, The cross-entropy formula is as follows: ， The formula for calculating the loss function using cross-entropy is as follows: ， Where m represents the number of samples, n represents the total number of categories in the classification task, P(x) represents the true distribution of the samples, and Q(x) represents the distribution distance predicted by the model.
9. The integrated digital twin grain storage monitoring and control method based on hyperspectral shortwave infrared as described in claim 6, characterized in that, Step S3 includes the following sub-steps: S301. The construction of the dynamic prediction model revolves around the point cloud data and spectral information related to volatile gases of insects and molds. A deep model is constructed by extracting feature values from the spectral information, where the spectral signal can be measured through multiple channels. S302 The system is connected to nitrogen-filled storage, new cold storage and dry storage back-end execution devices. When the system detects insects and mold, it will immediately make on-site adjustments in the storage environment and realize the selection of multi-dimensional grain storage methods, effectively slowing down the aging and deterioration of japonica rice.