Intelligent detection system for viability of germplasm resources

Abstract

The invention discloses an intelligent germplasm resource viability detection system, and relates to the technical field of germplasm resource viability detection, the system comprises a data acquisition subsystem and a data processing subsystem; the data acquisition subsystem is used for acquiring near infrared spectrum data and volatilization data of a plurality of seeds to be detected and sending the near infrared spectrum data and the volatilization data to the data processing subsystem; the data processing subsystem is used for processing the near infrared spectrum data and the volatilization data of the multiple to-be-detected seeds by using a target fusion method selected by a user to obtain a final seed vigor prediction result, the method adapts to the operation habits of agricultural employees, the non-destructive testing technology is combined with an embedded intelligent model, and the accuracy of the seed vigor prediction is improved. And rapid and accurate determination of seed vitality is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of germplasm resource vitality detection technology, and in particular to an intelligent detection system for germplasm resource vitality. Background Art

[0002] Germplasm resources are important strategic resources for maintaining biodiversity, supporting sustainable agricultural development, and ensuring food security. Their vitality level directly determines the success of species continuation and genetic improvement. In the field of agricultural breeding, seeds with high vitality are the basis for achieving uniform and strong seedlings and stable and high yields, while seeds with low vitality may lead to low germination rates, weak seedling growth, and even cause large-scale yield reductions. Therefore, the establishment of a rapid, accurate, and large-scale testing technology system for assessing the vitality of germplasm resources is a major demand in the global agricultural science and technology field. Traditional seed vitality testing technology has problems of inefficiency and lag, and there is an urgent need to introduce new intelligent, high-throughput non-destructive testing technologies.

[0003] In recent years, electronic noses and near-infrared spectroscopy have demonstrated unique advantages in the field of biological detection. However, relying solely on a single sensor (such as an electronic nose or spectroscopy) is difficult to fully reflect the physiological state of seeds. Furthermore, currently, near-infrared and electronic nose devices operate independently, and data collection and processing typically rely on external devices, resulting in complex operations and poor real-time performance. These devices are difficult to integrate to complete real-time on-site detection, requiring complex manual operations and high technical barriers. Therefore, solutions for intelligent system integration and multimodal data fusion are urgently needed to break through the efficiency and accuracy bottlenecks of traditional methods through collaborative innovation in hardware and algorithms. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and specifically provide an intelligent detection system for the vitality of germplasm resources, which is as follows:

[0005] Including data acquisition subsystem and data processing subsystem;

[0006] The data acquisition subsystem is used to: collect near-infrared spectrum data and volatilization data of a plurality of seeds to be tested, and send them to the data processing subsystem;

[0007] The data processing subsystem is used to process the near-infrared spectrum data and volatility data of multiple seeds to be tested using a target fusion method selected by the user to obtain a final seed vitality prediction result.

[0008] Based on the above solution, the intelligent detection system for germplasm resource vitality of the present invention can also be improved as follows.

[0009] Furthermore, the data acquisition subsystem includes a near-infrared subsystem and an electronic nose system. The near-infrared subsystem is used to collect near-infrared spectral data of multiple seeds to be tested, and the electronic nose system is used to collect volatilization data of multiple seeds to be tested.

[0010] Furthermore, the near-infrared subsystem and the electronic nose system are integrated into one chassis.

[0011] Furthermore, the electronic nose system includes: a sample resting bottle, a gas sensor array, a gas signal acquisition device, and a gas signal processing device;

[0012] The sample resting bottle is used to place at least one seed to be tested. The gas in each sample resting bottle is transported to the gas sensor array. The gas sensor array detects the composition of the received gas. The signal output by the gas sensor array is collected and converted by the gas signal acquisition device, and then amplified and filtered by the gas signal processing device to obtain volatilization data.

[0013] Furthermore, the near-infrared subsystem includes: a spectrometer, a transmission optical fiber and an optical probe, and the spectrometer is connected to the optical probe; the optical probe obtains the original reflection signal of the seed to be tested and transmits it to the spectrometer via the transmission optical fiber, and the spectrometer processes the received original reflection signal to obtain near-infrared spectrum data.

[0014] Furthermore, the data acquisition subsystem is further used to: obtain volatile data, near infrared spectrum data and vitality label data of each sample seed in the sample germplasm resources, and obtain a sample seed information set of the sample germplasm resources;

[0015] The data processing subsystem is further used to: after processing the sample seed information set of the sample germplasm resources using fusion algorithms at different levels, train the seed vigor prediction model corresponding to the fusion algorithm at each level.

[0016] Furthermore, the data processing subsystem is also specifically used to: use the target fusion method selected by the user to process the near-infrared spectral data and volatile data of the multiple seeds to be tested, and combine the trained seed vitality prediction model corresponding to the target fusion method to obtain the final seed vitality prediction results of the seeds to be tested.

[0017] Furthermore, when the target fusion method selected by the user is the first fusion method, the seed viability prediction model trained corresponding to the first fusion method is the first seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0018] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are spliced ​​to obtain the spliced ​​data of each seed to be tested, and a first seed vitality prediction result is obtained by using a first seed vitality prediction model, and the first seed vitality prediction result is used as the final seed vitality prediction result.

[0019] Furthermore, when the target fusion method selected by the user is the second fusion method, the seed viability prediction model trained corresponding to the second fusion method is the second seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0020] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction and feature splicing are performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested respectively to obtain feature splicing data of each seed to be tested, and a second seed vitality prediction result is obtained by using a second seed vitality prediction model, and the second seed vitality prediction result is used as the final seed vitality prediction result.

[0021] Furthermore, when the target fusion method selected by the user is the third fusion method, the trained seed viability prediction models corresponding to the third fusion method include a third seed vigor prediction model and a fourth seed vigor prediction model. The process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0022] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction is performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested, respectively, to obtain the near-infrared spectral features of each seed to be tested and the volatile features of each seed to be tested. The near-infrared spectral features of each seed to be tested and the third seed vitality prediction model are used to obtain a third seed vitality prediction result. The volatile features of each seed to be tested and the fourth seed vitality prediction model are used to obtain a fourth seed vitality prediction result. The third seed vitality prediction model and the fourth seed vitality prediction result are fused to obtain a fifth seed vitality prediction result, and the fifth seed vitality prediction result is used as the final seed vitality prediction result.

[0023] The beneficial effects of the intelligent detection system for germplasm resource vitality of the present invention are as follows:

[0024] The near-infrared subsystem and electronic nose system simultaneously collect the seeds' near-infrared spectral reflectance and electronic nose response data (i.e., volatile data). The near-infrared spectral sensor covers the near-infrared spectral range and is used to capture changes in the seeds' internal biochemical properties, while the electronic nose system uses its gas sensor array to analyze the various volatile organic compounds released by seeds during capping and storage. The control subsystem is equipped with a data processing subsystem, which, through a deep coupling design of hardware circuits and algorithms, implements an integrated processing flow of data acquisition, preprocessing, and multimodal fusion analysis. The data processing subsystem is equipped with an interactive module, namely a touch screen, which can display test results in real time, including germination rate, germination potential, germination index, and vitality index. It also supports data storage and export functions, adapting to the operating habits of agricultural practitioners. By combining non-destructive testing technology with embedded intelligent models, it can achieve rapid and accurate judgment of seed vitality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments of the present invention:

[0026] Figure 1 This is a structural diagram of an intelligent seed vitality detection system according to an embodiment of the present invention;

[0027] Figure 2 A front view of an intelligent seed vitality detection system according to an embodiment of the present invention;

[0028] Figure 3 This is a rear view of an intelligent seed vitality detection system according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of the structure of the electronic nose system;

[0030] Figure 5 is a structural diagram of a light source;

[0031] Figure 6 Schematic diagram of the seed viability detection software interface. DETAILED DESCRIPTION

[0032] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0033] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0034] like Figure 1 As shown, an intelligent detection system for germplasm resource vitality according to an embodiment of the present invention includes a data acquisition subsystem and a data processing subsystem;

[0035] The data acquisition subsystem is used to: collect near-infrared spectrum data and volatilization data of a plurality of seeds to be tested, and send them to the data processing subsystem;

[0036] The data processing subsystem is used to process the near-infrared spectrum data and volatility data of multiple seeds to be tested using a target fusion method selected by the user to obtain a final seed vitality prediction result.

[0037] The seeds to be tested may be rice seeds, corn seeds or wheat seeds, etc., and may be set according to actual conditions.

[0038] Optionally, in the above technical solution, the data acquisition subsystem includes: a near-infrared subsystem and an electronic nose system, the near-infrared subsystem is used to: collect near-infrared spectral data of multiple seeds to be tested, and the electronic nose system is used to: collect volatilization data of multiple seeds to be tested.

[0039] Optionally, in the above technical solution, the near infrared subsystem and the electronic nose system are integrated in one chassis.

[0040] Optionally, in the above technical solution, the electronic nose system includes: a sample standing bottle, a gas sensor array, a gas signal acquisition device and a gas signal processing device;

[0041] The sample resting bottle is used to place at least one seed to be tested. The gas in each sample resting bottle is transported to the gas sensor array. The gas sensor array detects the composition of the received gas. The signal output by the gas sensor array is collected and converted by the gas signal acquisition device, and then amplified and filtered by the gas signal processing device to obtain volatilization data.

[0042] Optionally, in the above technical solution, the near-infrared subsystem includes: a spectrometer, a transmission optical fiber and an optical probe, and the spectrometer is connected to the optical probe; the optical probe obtains the original reflection signal of the seed to be tested and transmits it to the spectrometer via the transmission optical fiber, and the spectrometer processes the received original reflection signal to obtain near-infrared spectral data.

[0043] Currently, near-infrared devices and electronic noses operate independently and cannot be integrated to jointly complete detection tasks. This requires complex manual operations for data collection, data transmission, and data fusion modeling of near-infrared spectral and volatile data. Therefore, the present invention proposes an intelligent seed viability detection system that collaboratively designs an integrated hardware architecture and a multimodal fusion algorithm to reduce manual operations and further improve practicality. This invention integrates the electronic nose system and near-infrared subsystem into a single device, providing the hardware structural foundation for multimodal, multi-level fusion seed viability detection. The following examples illustrate the invention.

[0044] like Figures 2 to 4 As shown, the seed vitality intelligent detection system in this embodiment includes a chassis, a near-infrared subsystem, an electronic nose system, a control subsystem, a data processing subsystem, and a power supply system. Each component is described as follows:

[0045] Among them, in the chassis, an upper horizontal plate 1 and a middle horizontal plate 26 are arranged in sequence from top to bottom along the vertical direction, and a vertical plate 27 is arranged in the chassis along the vertical direction. The upper horizontal plate 1, the middle horizontal plate 26 and the vertical plate 27 divide the space in the chassis into four parts: upper, middle, lower and back. The near-infrared subsystem is located in the middle and upper part of the chassis, and the electronic nose system is located in the lower part of the chassis. It should be noted that the positions of the near-infrared subsystem and the electronic nose system can be set according to actual conditions.

[0046] A display screen 31 is installed on the front of the upper part of the chassis, and a flip cover 9 is set on the top and right side of the upper space of the chassis to facilitate assembly and daily maintenance. Among them, a middle door and a lower door are set on the front of the middle and lower spaces respectively, which are connected by axle pin hinges for easy opening and closing.

[0047] 1) The electronic nose system includes: a sample resting bottle 22, a sample resting rack 23, an air intake unit 19, a cleaning unit 20, an exhaust unit 21, an air intake pump 43, a flow sensor 42, a gas sensor array 18, a gas signal acquisition device 25, a gas signal processing device 41, and an electronic nose power supply unit 24. Specifically:

[0048] ① The sample resting bottle 22 is used to place at least one seed, and the sample resting rack 23 is used to place the sample resting bottle 22. The usage process is: after the sample to be tested for volatile data (such as the seed to be tested, etc.) is loaded into the sample resting bottle 22, the sample resting bottle 22 needs to be sealed and capped to ensure the enrichment of the volatile gas of the seed, and then placed on the sample resting rack 23.

[0049] The sample standing bottles 22 of different capacities can be selected according to the size and quantity of the seeds to be tested, and the sample standing racks 23 can be provided according to the shapes of the sample standing bottles 22 .

[0050] The sample resting rack 23 can be placed inside or outside the chassis, and can be set according to actual conditions.

[0051] ② The air intake unit 19 includes an air intake probe 192, an air intake filter 193 and an air intake hose 191. The air intake pump 43 allows the external air to pass through the air intake unit 19 and reach the gas sensor array 18. Specifically, the air intake pump 43 is started to allow the air intake probe 192 to inhale the external air and transport it to the gas sensor array 18 through the air intake hose 191. During the transportation process, the inhaled external air is filtered by the air intake filter 193.

[0052] ③ The flow sensor 42 controls the flow of external gas delivered to the gas sensor array 18 by controlling the pump opening level. The flow delivered to the gas sensor array 18 can be detected by the flow sensor 42. After the external gas is delivered to the gas sensor array 18, the gas sensor array 18 detects the composition of the external gas. Each sensor element in the gas sensor array 18 responds to different gas molecules, causing its physical properties to change. At this time, each sensor element outputs a signal. The signal output by each sensor element is collected and converted by the gas signal acquisition device 25. The signal is then amplified and filtered by the gas signal processing device 41 to obtain volatility data, which is then transmitted to the data processing subsystem.

[0053] Among them, the gas sensor array 18 includes multiple sensor elements that are sensitive to different gas components, and the multiple sensor elements are arranged in sequence inside the gas sensor array 18. Among them, the number of sensor elements can be 10 (in this case, the gas sensor array 18 is a ten-channel gas sensor), and can also be set according to actual conditions. The arrangement of multiple sensor elements can also be set according to actual conditions.

[0054] ④ The cleaning unit 20 can ensure that the gas sensor array 18 remains clean and smooth during use. The cleaning unit 20 includes a cleaning filter element 202 and a cleaning hose 201. The two ends of the cleaning hose 201 are respectively connected to the air intake pump 43 and the cleaning filter element 202. External gas can be inhaled through the cleaning filter element 202 and then transported to the gas sensor array 18.

[0055] ⑤ The exhaust unit 21 is used to discharge exhaust gas. The exhaust unit 21 includes an exhaust probe 211 and an exhaust hose 212. The two ends of the exhaust hose 212 are respectively connected to the exhaust port of the gas sensor array 18 and the exhaust probe 211. The gas (exhaust gas) discharged from the exhaust port of the gas sensor array 18 is discharged in turn through the exhaust hose 212 and the exhaust probe 211.

[0056] The air intake hose 191 and the cleaning hose 201 may share a section of hose, through which external air is transported to the air intake pump 43 .

[0057] The data processing subsystem includes an operation interface displayed on display screen 31, which is used to set and adjust parameters such as the flow rate delivered to the gas sensor array 18, sampling duration, and cleaning duration. The control subsystem is responsible for coordinating operations such as intake and exhaust to ensure the normal sampling operation of the electronic nose system.

[0058] The process of collecting volatile data by the electronic nose system is as follows:

[0059] S101. The user first presses the power switch 38 to ensure power supply to the system of the present invention. The user then enters parameters such as the flow rate setting value, sampling time setting value, and cleaning time setting value to be delivered to the gas sensor array 18 in the operation interface of the display screen 31. The parameter information is sent to the control board 39 through the communication interface of the microcomputer 2 (data processing subsystem). The control board 39 generates a control signal based on the received parameters.

[0060] S102. Press the electronic nose start / stop switch 32. The control subsystem will receive a start signal to collect volatile data according to the generated control signal. A gas change signal during the cleaning process will appear on the display interface. After the cleaning is completed, a sampling prompt will be displayed. After the air intake probe 192 is inserted into the sample rest bottle 22 containing the sample to a certain depth (the depth can be set according to actual conditions), the user clicks to confirm the sampling signal and transmits it to the micro host 2. The micro host 2 inputs it into the control subsystem, and the control subsystem controls the air pump to start. At this time, the seed volatile compounds in the sample rest bottle 22 enter through the air intake unit 19 and are transmitted to the flow sensor 42 through the air intake pump 43. The control subsystem generates a control signal based on the received flow setting value, adjusts the air pump speed to control the gas flow through the gas sensor array 18. At the same time, the flow sensor 42 monitors the actual flow and transmits it to the control board 39. The control board 39 calculates the set flow value and the actual flow value fed back by the flow sensor 42, and the PID control algorithm of the control board 39 calculates the control signal amount that needs to be adjusted to achieve precise flow control. When gas is transmitted to the gas sensor array 18, it responds. The generated signal is collected and converted by the gas signal acquisition device 25. The gas signal processing device 41 performs signal amplification and filtering to obtain volatility data, which is then transmitted to the data processing subsystem. The exhaust gas is then discharged through the exhaust unit 21.

[0061] 2) The near-infrared subsystem includes: a cooling fan 3, a spectrometer 5, a spectrometer baffle 4, a transmission fiber 13, an optical probe 44, a light source 16, a sample placement plate 17, and a three-axis slide module. Specifically:

[0062] ① The sample placement plate 17 is used to place samples to be measured for near-infrared spectrum (such as seeds to be measured). For example, the sample placement plate 17 is provided with a plurality of grooves to facilitate placement of seeds to be measured in the grooves.

[0063] It should be noted that the sample placement plate 17 with the adapted grooves can be replaced according to the size of the seeds to be tested, so as to ensure that each seed to be tested can be placed exactly in a single groove of the sample placement plate 17 .

[0064] ② One end of the transmission optical fiber 13 is fixedly installed in the transmission optical fiber 13 interface of the optical probe 44, and the other end is connected to the spectrometer 5. The optical signal collected by the optical probe 44 is transmitted to the spectrometer 5 through the transmission optical fiber 13. The spectrometer 5 performs signal processing such as splitting, photoelectric conversion, analog-to-digital conversion, amplification and filtering on the received optical signal to obtain near-infrared spectral data.

[0065] The light source 16 may be an annular light source 16, and the optical probe 44 is fixed at the center of the annular light source 16, such as Figure 5 shown.

[0066] ③ The three-axis slide module consists of an X-axis slide 30, a Y-axis slide 29, a Z-axis slide 12, a limit switch 11 at each end of each axis, and a stepper motor 28. The three-axis slide module is installed and fixed on the inner plane of the chassis, specifically fixedly connected to the upper horizontal plate 1 and the side plate of the chassis, and the optical probe 44 is fixed to the bottom end of the Z-axis slide track. The motor controller 35 drives the stepper motor 28 in three directions to complete the control of movement speed and position; the three-axis slide module drives the light source 16 and the optical probe 44 to the top of each groove center of the sample placement plate 17 in turn, continuously obtains the near-infrared spectrum of the seed to be tested at each groove, and then returns to the initial position. The optical probe 44 obtains the original reflection signal of the seed to be tested (the original reflection signal is a light signal) and transmits it to the spectrometer 5 through the transmission optical fiber 13. The spectrometer 5 converts the original reflection signal into a digital signal. The spectrometer 5 records the light intensity value of each wavelength as a data point to form a set of near-infrared spectrum data, thereby obtaining the near-infrared spectrum data of each seed to be tested.

[0067] ④ After receiving the input of the external switch signal, the motor controller 35 controls the motor driver 40 and the stepper motor 28 to complete the control task of the three-axis slide module.

[0068] ⑤ The control board 39 receives the trigger signal, i.e., the external switch signal, from the spectral acquisition switch 10, and sends a control signal to the motor controller 35 according to the pre-written spectral acquisition code. The motor controller 35 receives the control signal and converts it into an instruction format suitable for the motor driver 40 and sends it to the motor driver 40. Finally, the motor driver 40 drives the stepper motor 28 according to the received instruction to complete the driving task of the three-axis slide module.

[0069] ⑥ The light source 16 and the optical probe 44 are located at the bottom end of the signal-collecting optical fiber 13, and are connected to the Y-axis slide 29 via a fixing block 14. The light source 16 is arranged in a ring shape, i.e., the light source 16 is a ring-shaped light source 16 with the optical probe 44 in the middle. The individual lamp beads of the ring-shaped light source 16 are connected in series, and a voltage stabilizing module 15 is provided between the light source 16 and the power supply to improve the stability of the light source 16.

[0070] 7. Spectrometer 5 is connected to a near-infrared detector. The near-infrared reflection signal from the seeds is collected by optical fiber 13 and then transmitted to spectrometer 5 for spectrum analysis, photoelectric conversion, analog-to-digital conversion, amplification, and filtering, generating near-infrared spectral data. The generated near-infrared spectral data is then transmitted via a signal transmission line to the data processing subsystem for data processing and modeling.

[0071] The data processing subsystem is equipped with an operation interface for displaying on the display screen 31, which is used to set parameters such as the near-infrared band sampling range, sampling speed, exposure time, etc., to ensure the normal sampling operation of the near-infrared subsystem.

[0072] ⑧The cooling fan 3 is located at the upper part of the chassis and is installed on the vertical plate 27. The spectrometer 5 and the spectrometer baffle 4 are located at the upper part of the electronic nose system and are installed on the upper horizontal plate 1. The spectrometer 5 is located inside the spectrometer baffle 4 and the interface of the spectrometer 5 is located facing the spectrometer baffle 4 to prevent dust and debris from entering the interior of the spectrometer 5 from the interface of the spectrometer 5.

[0073] The process of collecting near-infrared spectral data by the near-infrared subsystem is as follows:

[0074] S201. The user first presses the power switch 38 to ensure power supply to the entire system, places the seeds (sample seeds or seeds to be tested) in the grooves of the sample placement plate 17 in sequence, presses the light source switch 6 and the spectrometer start-stop switch 7, and starts the light source 16 and the spectrometer 5.

[0075] S202: The user inputs parameters such as the near-infrared band sampling range, sampling speed, and exposure time set in the operation interface of the display screen 31, and the parameter information is sent to the spectrometer 5 through the communication interface of the microcomputer 2 (data processing subsystem); the spectrometer 5 receives the parameter information;

[0076] S203: The user then presses the spectrum acquisition switch 10. The control subsystem receives the trigger signal from the spectrum acquisition switch 10 and, based on the pre-programmed spectrum acquisition code, sends a control signal to the motor controller 35. The motor controller 35 receives the control signal, converts it into a command format suitable for the motor driver 40, and sends it to the motor driver 40. Finally, the motor driver 40 drives the stepper motor 28 according to the received command to complete the task of driving the three-axis slide module. The three-axis slide module drives the optical probe 44 to the top of the first seed, aligning the center of the optical probe 44 with the center of the seed. The distance between the optical probe 44 and the seed can be adjusted according to actual conditions. After moving to above the first seed, the control subsystem will trigger the spectrometer 5 to collect near-infrared spectra. The spectrometer 5 controls the optical probe 44 to collect the near-infrared reflection signal of the first seed according to the set parameters. The near-infrared reflection signal is transmitted to the spectrometer 5 by the transmission optical fiber 13, and the near-infrared spectrum data of the first seed is obtained after signal conversion and processing; then the spectrometer 5 triggers the control subsystem to control the stepper motor 28 to move above the second seed, repeat the collection operation, and obtain the near-infrared spectrum data of the second seed; until the near-infrared spectrum data of the seeds in each groove position of the sample placement plate 17 are collected, the control subsystem controls the three-axis slide module to return to the initial position.

[0077] The workflow of the spectrum acquisition code is as follows:

[0078] The user presses the spectrum acquisition switch 10, and the control subsystem receives the trigger signal sent by the switch, and then the control subsystem starts the near-infrared spectrum acquisition code process;

[0079] First, define the state variables of the spectrum acquisition switch 10, and store the position of the center of each groove of the sample placement plate 17 on the XY plane in the form of a samplePositions array, and set a fixed height of the Z axis, where the XY plane is a plane perpendicular to the Z-axis slide 12, and the Z axis is an axis parallel to the Z-axis slide 12.

[0080] Define the Spectral_Collection() function, access the center coordinates of each groove of the sample placement plate 17 in sequence through a nested loop, and call the sample_Collect() function to control the XY axis to move to the position of each sample for data collection;

[0081] Among them, the sample_Collect() function is defined, and the X-axis slide 30, Y-axis slide 29, and Z-axis slide 12 are set to move to the specified coordinates through motor_Command(). Then the HAL_Delay delay function waits for the X-axis slide 30, Y-axis slide 29, and Z-axis slide 12 to reach the specified coordinates in the samplePositions array. Then the HAL_UART_Transmit() function sends a valid trigger instruction to the spectrometer 5 through the interface. Finally, the delay function is set to the time for the spectrometer 5 to collect data.

[0082] Among them, the motor_Command() stepper motor control function is defined. First, the stepper motor speed is set and the stepper motor 28 is detected to be in the initial position. By inputting the names and step numbers of the three-axis slide module corresponding to the X-axis slide 30, Y-axis slide 29, and Z-axis slide 12, the slide movement in the corresponding direction is controlled.

[0083] In the main function, the GPIO_Init() and UART_Init() functions are called to initialize the hardware configuration. The state variable of the spectrum acquisition switch 10 is then detected. When the spectrum acquisition switch 10 is pressed, the Spectral_Collection() function is called to start spectrum acquisition. After the spectrum acquisition is completed, a while loop is used to wait for the button to be released to avoid continuous triggering of acquisition.

[0084] Among them, the GPIO_Init() function module is defined to initialize the pins of the spectrum acquisition switch 10 and the motor controller 35, etc., set the pin of the spectrum acquisition switch 10 to input mode and enable the pull-up resistor so that the pin is high when the spectrum acquisition switch 10 is not pressed and low when pressed; the UART_Init() function is defined to initialize the communication serial port.

[0085] S204. Before collecting the near-infrared spectrum data of the seeds, the near-infrared subsystem is made to collect a background signal once to eliminate the dark current noise of the instrument. Specifically, the control subsystem controls the near-infrared subsystem to collect the background signal and transmits the background signal to the spectrometer 5 through the transmission optical fiber 13. The background signal is split and photoelectrically converted inside the spectrometer 5, and the corrected spectrum signal is obtained through background correction, and the spectrum signal is output to the data processing subsystem.

[0086] S205. According to the sampling sequence set by the control subsystem, the three-axis slide module is controlled to drive the optical probe 44 and the light source 16 to the top of each seed for signal collection, until the signal of each seed is collected, and the collected near-infrared spectrum data of each seed is uploaded to the data processing subsystem. When the detection of all seeds is completed, the three-axis slide module is controlled to drive the optical probe 44 and the light source 16 to automatically reset.

[0087] The data processing subsystem analyzes and processes all received near-infrared spectral data, extracts the characteristic wavelengths of seed vitality, and then establishes a seed vitality detection model based on PLSR through multi-information fusion, and then trains to obtain a third seed vitality prediction model.

[0088] 3) The data processing subsystem is connected to the near-infrared subsystem and the electronic nose system, and is used to read, preprocess, extract features, divide data sets, establish training models, and perform other operations on the volatile data (volatile data can also be called electronic nose response data) collected by the electronic nose system and the near-infrared spectral data collected by the near-infrared subsystem. The data processing subsystem includes a microprocessor (the microprocessor can be an industrial computer or a computer, etc.) and a display screen 31.

[0089] Optionally, in the above technical solution, the data acquisition subsystem is further used to: obtain volatile data, near-infrared spectrum data, and vitality label data of each sample seed in the sample germplasm resources to obtain a sample seed information set of the sample germplasm resources;

[0090] Among them, the vitality label data includes germination (vitality) and no germination (no vitality), as well as vitality indicators such as germination rate, germination potential, germination index and vitality index.

[0091] The data processing subsystem is further used to: after processing the sample seed information set of the sample germplasm resources using fusion algorithms at different levels, train the seed vigor prediction model corresponding to the fusion algorithm at each level.

[0092] Optionally, in the above technical solution, the data processing subsystem is also specifically used to: use the target fusion method selected by the user to process the near-infrared spectral data and volatile data of the multiple seeds to be tested, and combine the trained seed vitality prediction model corresponding to the target fusion method to obtain the final seed vitality prediction result of the seeds to be tested.

[0093] Among them, the target fusion methods are data-level fusion method, feature-level fusion algorithm and decision-level fusion algorithm. The data-level fusion method is the first fusion method, the feature-level fusion algorithm is the second fusion method, and the decision-level fusion algorithm is the third fusion method.

[0094] Optionally, in the above technical solution, when the target fusion method selected by the user is the first fusion method, the trained seed viability prediction model corresponding to the first fusion method is the first seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0095] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are spliced ​​respectively to obtain the spliced ​​data of each seed to be tested, and a first seed vitality prediction result is obtained by using a first seed vitality prediction model, and the first seed vitality prediction result is used as the final seed vitality prediction result.

[0096] Optionally, in the above technical solution, when the target fusion method selected by the user is the second fusion method, the seed viability prediction model trained corresponding to the second fusion method is the second seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0097] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction and feature splicing are performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested respectively to obtain feature splicing data of each seed to be tested, and a second seed vitality prediction result is obtained by using a second seed vitality prediction model, and the second seed vitality prediction result is used as the final seed vitality prediction result.

[0098] Optionally, in the above technical solution, when the target fusion method selected by the user is the third fusion method, the trained seed viability prediction model corresponding to the third fusion method includes a third seed vigor prediction model and a fourth seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes:

[0099] After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction is performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested to obtain the near-infrared spectral features of each seed to be tested and the volatile features of each seed to be tested. The near-infrared spectral features of each seed to be tested and the third seed vitality prediction model are used to obtain a third seed vitality prediction result. The volatile features of each seed to be tested and the fourth seed vitality prediction model are used to obtain a fourth seed vitality prediction result. The third seed vitality prediction model and the fourth seed vitality prediction result are fused to obtain a fifth seed vitality prediction result, and the fifth seed vitality prediction result is used as the final seed vitality prediction result.

[0100] The training model is embedded in the microprocessor. The training model is established based on the volatility data of multiple groups of sample seeds collected by the electronic nose system and the near-infrared spectrum data of multiple groups of sample seeds collected by the near-infrared subsystem. Specifically:

[0101] S301. The data processing subsystem obtains the volatile data of multiple groups of sample seeds collected by the electronic nose system and the near-infrared spectral data of multiple groups of sample seeds collected by the near-infrared subsystem, and obtains the value of the vitality index of each group of sample seeds. The multiple seed vitality indicators include germination rate, germination potential, germination index and vitality index.

[0102] S302: Preprocess the volatile data and near-infrared spectrum data of each sample seed in each group, specifically by smoothing and denoising, and then perform processing and modeling at the data level, feature level, and decision level based on the preprocessed volatile data and preprocessed near-infrared spectrum data of each sample seed in each group. Specifically:

[0103] a. At the data level, the preprocessed near-infrared spectral data and the preprocessed volatile data of each sample seed in each group are spliced ​​to obtain the spliced ​​data of each sample seed in each group, and combined with the value of each seed vitality index of each group of sample seeds, multiple samples are constructed and recorded as first samples. All first samples are divided into a training set (also called a calibration set) and a prediction set, which are recorded as the first training set and the first prediction set respectively. After the first training model is established, the first training set and the first prediction set are used to train the first training model to obtain a first seed vitality prediction model. The accuracy of the seed vitality prediction results obtained by the first seed vitality prediction model can also be evaluated. This process is completed through the data-level fusion module.

[0104] Among them, during preprocessing, the size of the preprocessed near-infrared spectral data was adjusted to 233×1, the preprocessed volatile data was adjusted to 200×1, and the size of the spliced ​​data was 433×1. The coefficient of determination R 2 The root mean square error (RMSE) and the root mean square error (RMSE) characterize the accuracy of the seed viability prediction results obtained by the first seed viability prediction model.

[0105] b. At the feature level, feature extraction and feature splicing are performed on the preprocessed near-infrared spectral data and preprocessed volatile data of each sample seed in each group to obtain the feature splicing data of each sample seed, and combined with the value of each seed vitality index of each group of sample seeds, multiple samples are constructed and recorded as second samples. All second samples are divided into a training set and a prediction set, which are recorded as the second training set and the second prediction set respectively. After the second training model is established, the second training set and the second prediction set are used to train the second training model to obtain a second seed vitality prediction model. The accuracy of the seed vitality prediction results obtained by the second seed vitality prediction model can also be evaluated. This process is completed through the feature-level fusion module.

[0106] Among them, an electronic nose feature extraction (E-NFE) unit and a near-infrared feature extraction (NIR-FE) unit are constructed. The electronic nose feature extraction unit is used to perform feature extraction on the preprocessed volatile data of each sample seed in each group to obtain the volatile characteristics of each sample seed. The near-infrared feature extraction (NIR-FE) unit is used to perform feature extraction on the preprocessed near-infrared spectral data of each sample seed in each group to obtain the near-infrared spectral characteristics of each sample seed.

[0107] c. At the decision-making level, feature extraction is performed on the preprocessed near-infrared spectral data and preprocessed volatile data of each seed sample in each group to obtain near-infrared spectral features and volatile features for each seed sample in each group. These features are combined with the near-infrared spectral features of each seed sample in each group and the values ​​of each seed vitality index in each seed sample group to construct multiple samples, denoted as a third sample set. All third samples are divided into a training set and a prediction set, denoted as a third training set and a third prediction set, respectively. After establishing a third training model, the third training set and the third prediction set are used to train the third training model to obtain a third seed vitality prediction model. Combining the volatile features of each seed sample in each group and the values ​​of each seed vitality index in each seed group, multiple samples are constructed, denoted as a fourth sample set. All fourth samples are divided into a training set and a prediction set, denoted as a fourth training set and a fourth prediction set, respectively. After establishing a fourth training model, the fourth training set and the fourth prediction set are used to train the fourth training model to obtain a fourth seed vitality prediction model. This process is completed by the decision-making level fusion module.

[0108] Among them, the first training model, the second training model, the third training model and the fourth training model can all be partial least squares regression models PLSR. When dividing the first sample set, the second sample set, the third sample set and the fourth sample set, the parameter required by the data set division function is the random seed number. When the value is fixed, the division result will be fixed, and the function can be controlled to select a random data set division mode; the important parameter of the partial least squares regression model is the number of principal components, and the number of principal components determines the dimension of the data after dimensionality reduction; in order to obtain appropriate random seed parameter values ​​of the data division function and principal component number parameter values ​​of the partial least squares regression, the grid search method is used to traverse the possible random seed value range and the principal component number value range, and calculate the prediction determination coefficient of the partial least squares model under each combination of the two. The random seed value and the principal component number value when the determination coefficient value is larger are the optimal parameter combination.

[0109] The decision-level fusion module is also used to assign weights to the seed viability prediction results obtained by the third seed viability prediction model and the seed viability prediction results obtained by the fourth seed viability prediction model, and to obtain the final seed viability prediction results by weighting.

[0110] The data reading process, data preprocessing process, first seed viability prediction model, second seed viability prediction model, third seed viability prediction model, and fourth seed viability prediction model are embedded in the program software. The seeds to be tested are placed in a sample resting bottle 22 and placed on a sample resting rack 23. The electronic nose system in the data acquisition subsystem collects volatile data from the seeds to be tested. The seeds to be tested are then removed and placed on a sample placement plate 17. The near-infrared subsystem in the data acquisition subsystem collects near-infrared spectral data from the seeds to be tested. The software embedded in the data processing subsystem then processes the near-infrared spectral data and volatile data (hereinafter referred to as volatile signals) of the multiple seeds to be tested to obtain a final seed viability prediction result. The housing of the present invention can be sealed during operation to prevent external light from affecting near-infrared detection. The device is compact, easy to operate, and has high detection accuracy. The software has the function of analyzing and processing data, integrating testing and analysis for ease of operation.

[0111] In another embodiment, the system includes an electronic nose signal acquisition unit, a near-infrared spectrum acquisition unit, a target tag reading and information integration unit, a signal processing unit, a model building unit, and a fusion training unit. The electronic nose signal acquisition unit is configured to acquire the volatile signal of the target sample after it has been allowed to rest for a certain period of time, including the signal collected by each channel of the gas sensor array 18. The near-infrared spectrum acquisition unit is configured to acquire the spectral signal of the target sample, including signals within a wavelength range. The target tag reading and information integration unit is configured to integrate the data into a single system, either wired or wirelessly. The signal processing and model building unit is configured to perform the following steps on the aforementioned signals: processing and formatting the volatile signal and near-infrared spectrum signal, performing feature acquisition on the volatile signal, and performing data preprocessing on the spectrum. Model training is performed on the multimodal data and target information labels to generate network weight information and network structure information. A multimodal fusion detection network is constructed based on the training results, the aforementioned network weight information, and the target network structure information. The fusion training unit is configured to fuse the multimodal fusion detection network into a multimodal detection model, input the target volatility and spectral information into the above multimodal seed vitality detection model, and obtain the target vitality detection result.

[0112] The upper and middle dark boxes of the near-infrared integrated device include a cooling fan 3, a spectrometer 5, a spectrometer baffle 4, a transmission fiber 13, an optical probe 44, a light source 16, a sample placement plate 17, and a three-axis slide module. One end of the transmission fiber 13 is fixedly installed in the interface of the transmission fiber 13 to collect light signals, and the other end is connected to the spectrometer 5 to transmit the spectrum collected by the transmission fiber 13 to the spectrometer 5 for spectrometry and photoelectric conversion. The optical probe 44 is fixed at the center of the annular light source 16.

[0113] The electronic nose is located at the lower layer of the integrated device, including external detection interfaces and internal structures. The gas flows through the sensor array after the flow rate is adjusted by the flow control device. The collected signals pass through the data acquisition system (analog-to-electric conversion) and the processing unit (amplification and filtering, etc.), and the gas is discharged through the exhaust port.

[0114] The process of obtaining sample seeds is as follows: seeds with no surface damage, uniform size and shape are selected and placed in an aging box for aging treatment, and are taken out at different times to obtain sample seeds with different aging days and different vitality levels. The sample seeds taken out of the aging box are placed at room temperature for a certain period of time to balance.

[0115] The detection operation process is as follows:

[0116] S30. After the seed sample is placed in the sample standing bottle 22 and placed on the sample standing rack 23 to stand at room temperature, the electronic nose system is used to collect the volatile odor information of the seed. The specific steps are as follows: press the electronic nose start-stop switch 32 to turn on the device, the electronic nose system takes in air to clean the residual gas in the pipeline, click to start sampling on the display interface, and insert the sampling probe of the electronic nose system into a certain position of the sample standing bottle 22 to collect the enriched gas in the bottle. After the collection is completed, the volatile data will be saved in the corresponding folder in sequence.

[0117] S31. After collecting the volatilization data of the seeds to be tested, the near-infrared subsystem is used to collect the near-infrared spectrum of the seeds. The specific steps are as follows: take out the seeds in the sample static bottle 22, place them in the seed grooves of the shelf in turn, press the light source switch 6 and the spectrometer start-stop switch 7, start the light source 16 and the spectrometer 5, then press the spectrum acquisition switch 10, and the three-axis slide module drives the optical probe 44 to collect the sample near-infrared spectrum in turn. After the acquisition is completed, the near-infrared spectrum data will be saved in the corresponding folders in turn.

[0118] S32. Manually use a standard germination test to obtain the germination rate, germination potential, germination index and vitality index of the tested seeds, and manually record the data and save them in the corresponding file.

[0119] S33. Use the data processing subsystem to read and preprocess the volatilization data and near-infrared spectrum data of the seeds respectively.

[0120] S34. The data processing subsystem is used to perform feature extraction on the seed volatile data. The E-NFE (electronic nose feature extraction) unit extracts the volatile features of the seeds as input for the electronic nose feature-level modeling of the model.

[0121] S35. The data processing subsystem of the multi-level multimodal detection system for seed vitality detection is used to extract features from the near-infrared spectral data of the seeds. The NIR-FE (near-infrared feature extraction) unit extracts the near-infrared spectral features of the seeds as input for the near-infrared spectral feature-level modeling of the model.

[0122] Establish a seed viability prediction model in the data processing subsystem,

[0123] The volatile characteristics and near-infrared spectral characteristics of the seeds are input, and a partial least squares regression (PLSR) model is trained in Python. Then, the trained model is called in the data processing subsystem to predict the vitality of the unknown seeds. Among them, the vitality prediction indicators are germination rate, germination potential, germination index and vitality index.

[0124] The display screen 31 in the data processing subsystem is connected to the microprocessor via a data line, and the display interface of the program software, i.e., the seed vitality detection software interface, is displayed on the display screen 31. Figure 6 As shown, through human-computer interaction, functions such as sample data reading, data preprocessing, feature extraction, and model prediction are realized. It also includes a near-infrared spectrum display area and an electronic nose response signal display area. Users can click on the interface function modules to implement corresponding functions, such as selecting the target fusion method.

[0125] 4) The control subsystem is described as follows:

[0126] The control subsystem includes a motor driver 40, a motor controller 35 and a control board 39. A control program is written into the control board 39, and the motor controller 35 and the motor driver 40 are controlled by the control program to drive the stepper motor 28 on the three-axis slide module to move. In addition, the control subsystem also completes the control of the light source switch 6, the spectrometer start and stop switch 7, the spectrum collection switch 10 and the electronic nose collection switch, as well as the control of data transmission. A plurality of wiring terminals 34 are provided above the motor controller 35.

[0127] The control subsystem is used to control the opening and closing of the spectrometer 5, the opening and closing of the light source 16, the collection of near-infrared spectrum data and the collection of volatile data, as well as the data transmission. Specifically:

[0128] The light source switch 6, spectrometer start / stop switch 7, and spectrum acquisition switch 10 are located on the upper right side of the chassis, facing outward. The electronic nose start / stop switch 32 and the network cable interface 33 are located on the upper left side of the chassis, facing outward. The USB adapter 8 is located on the flip cover 9, on the upper right side of the chassis and facing outward.

[0129] a. When it is detected that the spectrometer start-stop switch 7 is pressed, the spectrometer 5 is started. When the spectrometer start-stop switch 7 is pressed again, the spectrometer 5 is turned off.

[0130] b. When it is detected that the light source switch 6 is pressed, the light source 16 is turned on. When the light source switch 6 is pressed again, the light source 16 is turned off.

[0131] c. When it is detected that the spectrum start / stop switch is pressed, the stepper motor 28 drives the three-axis slide module to move, thereby driving the optical probe 44 to sequentially obtain the near-infrared spectrum data of each seed to be tested and transmit it to the data processing subsystem.

[0132] d. When it is detected that the electronic nose start-stop switch 32 is pressed, the electronic nose system is started. When the electronic nose start-stop switch 32 is pressed again, the electronic nose system is turned off.

[0133] e. Data transmission is mainly completed through the USB conversion interface 8 and the network cable interface 33.

[0134] 5) The power supply system is used to supply power to the data acquisition subsystem and the data processing subsystem. The power supply subsystem includes an electronic nose power supply unit 24 for supplying power to the electronic nose system, and a power supply unit 36 ​​for supplying power to the near-infrared subsystem and the data processing subsystem. The electronic nose power supply unit 24 is located in the lower space of the chassis, and the power supply unit 36 ​​for supplying power to the near-infrared subsystem and the data processing subsystem is located at the back of the chassis. A junction box 37 is provided below the power supply unit 36.

[0135] In an intelligent seed vitality detection system according to an embodiment of the present invention, a data acquisition subsystem collects information about seed vitality; a control subsystem detects whether a switch is pressed and controls the device's startup and shutdown; a data processing subsystem builds and trains a training model, as well as predicts seed vitality; and a power supply subsystem supplies power to all system components. The system integrates the electronic nose system and near-infrared subsystem into one, achieving spatial integration. Furthermore, data transmission, processing, and control are integrated with both the electronic nose system and the near-infrared subsystem, achieving operational integration. This integrated system enables the simultaneous acquisition and transmission of volatile and near-infrared data, as well as their processing in the data processing subsystem.

[0136] The data processing subsystem is equipped with an interactive design interface, a seed vitality detection software interface based on multimodal information fusion, with four options on the left side for importing data, preprocessing, fusion model method, and prediction results. Clicking on import data will display near-infrared spectral data, volatile data, and seed vitality data on the right interface. After selecting the data, an image of the data will be automatically drawn below. Clicking on preprocessing will display options for multiple preprocessing methods on the right interface. After selecting one of the preprocessing methods, the preprocessed image will be automatically drawn below. Clicking on the fusion model method will enable the right interface to implement options for three fusion methods: data-level fusion, feature-level fusion, and decision-level fusion. Select one of the fusion methods and click on the prediction result option on the left interface to display the modeling results under the fusion method. The results are expressed as the determination coefficient and root mean square error of the correction set and prediction set.

[0137] The operating process of the software program used in the system of the present invention is as follows:

[0138] S401, signal triggering: the data acquired by the electronic nose and near infrared are automatically transmitted to the operating system of the micro host 2 through the interface. The interface type can be selected from serial port interface, USB interface or Ethernet interface.

[0139] S402, protocol analysis: After receiving the data, the micro host 2 needs to analyze the data according to the communication protocol of the electronic nose or near infrared system.

[0140] S403. Format conversion: The data output by the electronic nose and near-infrared system is usually in text or binary formats such as txt, csv, and bin. Therefore, after receiving the data parsed by the protocol, the reading program in the software program of the micro host 2 will further convert the data format of the volatile data or near-infrared spectrum into a data structure such as an array or matrix that is convenient for analysis. The format-converted data can be stored in the local storage of the micro host 2. At the same time, the reading program will display the parsed data on the display screen 31, presenting the electronic nose response curve or spectral curve through a graphical interface.

[0141] S404, pre-processing: Then, in the pre-processing program of the software program of the micro host 2, the user can click to select a pre-processing method such as noise removal, baseline correction, etc., and pre-process the electronic nose response curve or spectrum curve.

[0142] S405, modeling: The model building program in the software program in the host performs modeling and analysis, and the user clicks to select the data level, feature level, and decision level fusion modeling method to implement the fusion model training.

[0143] S406, result display: The result prediction program in the software program in the host can calculate the results of the above fusion model and display them on the interface.

[0144] The FusionAnalysisApp class is defined to build a graphical interface application for multi-level modeling and analysis of volatile and near-infrared spectral data fusion. Specifically:

[0145] First, the basic interface and controls of the application are initialized. The main window of the application is passed in through self.root. The title of the main window is set to "Multimodal and Multi-level Fusion Detection of Seed Viability" by self.root.title, and the size of the main window is set by self.root.geometry. Then, the left menu is created using self.left_frame. The menu frame is placed on the left side of the window and fills the entire window vertically with self.left_frame.pack. The button with the text "Import Data" is created by self.import_button and placed in the left menu bar for importing data. The button with the text "Preprocessing" is created by self.preprocessing_button and placed in the left menu bar for selecting the preprocessing method. self.fusion_button creates a button with the text "Fusion Model Method" and places it in the left menu bar to select the fusion model method; self.prediction_button creates a button with the text "Prediction Result" and places it in the left menu bar to display the prediction result; self.right_frame creates the right display area, sets its placement and fills the entire window horizontally and vertically; self.preprocessing_methods creates a list containing the text of available preprocessing methods, such as "Standardization", "SG Smoothing", "Normalization", etc.; self.fusion_methods creates a list containing the available fusion methods, in this case there are "data-level fusion", "feature-level fusion" and "decision-level fusion".

[0146] Optionally, in S403, the data reading code is completed in Python, and the import_enose_data(self) function is defined to read the electronic nose file, use a judgment statement to determine the file type, and then import the data into an electronic nose array according to the file type and return the array; the import_nir_data(self) function is defined to read the near-infrared spectrum file, use a judgment statement to determine the file type, and then import the data into a near-infrared spectrum array according to the file type and return it; the display_enose_plot(self) function is defined to clear the existing image on the right, and then draw the imported electronic nose array response radar image. For each sample electronic nose array, the first column is time, and the 2nd to 11th columns are the electronic nose sensor response values ​​corresponding to the 1st to 10th channels; the display_nir_plot(self) function is defined to clear the existing image on the right, and then draw the imported near-infrared spectrum curve graph. For each sample spectrum array, the first column is wavelength, and the second column is reflectance value.

[0147] Optionally, in S404, the data preprocessing code is completed in Python, the show_preprocessing_options(self) function is defined, the existing content on the right is cleared first, and the show_enose_preprocessing(self) function is called to create an electronic nose preprocessing method button component. When it is detected that the user has selected a preprocessing button, the preprocessing method is executed and the preprocessed volatile response image is drawn below. The function returns the electronic nose preprocessing array; the show_nir_preprocessing(self) function is called to create a near-infrared spectrum preprocessing method button component. When it is detected that the user has selected a preprocessing button, the preprocessing method is executed and the preprocessed spectral image is drawn below. The function returns the spectral preprocessing array.

[0148] Optionally, in the above technical solution, the show_enose_preprocessing(self) function uses Python's Tkinter library to create a label on the left right side of the interface to prompt the user to select the electronic nose preprocessing method, and then creates a button component for each electronic nose preprocessing method under the label. The button text displays the corresponding preprocessing method. When the button is clicked, the apply_preprocessing(self, method) function is called, and the corresponding preprocessing method is passed as the method parameter to the apply_preprocessing(self, method) function to perform the preprocessing to obtain a preprocessed array. After the data preprocessing is completed, the preprocessed array is used to draw the preprocessed electronic nose response radar image; finally, the function returns the electronic nose preprocessing array.

[0149] Optionally, in the above technical solution, the show_nir_preprocessing(self) function uses Python's Tkinter library to create another label on the right side of the interface to prompt the user to select a spectral preprocessing method, and creates a button component for each spectral preprocessing method under the label; the button text displays the corresponding preprocessing method, and when the button is clicked, the apply_preprocessing(self, method) function is called, and the corresponding preprocessing method is passed as the method parameter to the apply_preprocessing(self, method) function to perform the preprocessing to obtain a preprocessed array. After the data preprocessing is completed, the preprocessed array is used to draw the preprocessed spectral image; finally, the function returns the spectral preprocessing array;

[0150] Optionally, in the above technical solution, the apply_preprocessing(self,method) function accepts a parameter method, which represents the preprocessing method to be applied. The function then determines what kind of preprocessing method is, and if so, performs the corresponding preprocessing; using the messagebox.showinfo tool function in Tkinter, a pop-up information prompt box method, the value of method is inserted into the prompt box content string, so that the pop-up box displays that the preprocessing method has been completed.

[0151] Optionally, in S405, define the show_fusion_options(self) function, first clear the current right panel content, use Python's Tkinter library to create a label on the upper right side of the interface to display "Select fusion method" to prompt the user to select a multimodal fusion modeling method, and create button components for three fusion modeling methods at the data level, feature level, and decision level under the label. The button text displays the corresponding fusion modeling method. When the button is clicked, call the fusion_modes(self, mode) function, detect the button pressed by the user, and display the text as mode. Use the lambda expression to pass the current mode as a parameter to the fusion_modes(self, mode) function, then the function executes the selected fusion modeling method and obtains the predicted seed vitality array. The function returns the seed vitality prediction value of the fusion modeling.

[0152] Optionally, in the above technical solution, the fusion_modes(self,mode) function accepts a parameter mode, which represents the multimodal fusion method to be applied, that is, the data level, feature level, or decision level fusion method. Then the function determines which fusion method mode is, and if so, executes the corresponding fusion modeling method; uses the messagebox.showinfo tool function in Tkinter, a pop-up information prompt box method, to insert the value of mode into the prompt box content string, so that the pop-up box displays the selected fusion method; and finally returns the seed vitality prediction value of the fusion modeling.

[0153] Optionally, in S406, show_prediction_results(self) is defined, the function reads the true value of seed viability as y_true, reads the predicted value of the fusion model as y_pred; then uses r2_score(y_true,y_pred) to calculate the R between the true value and the predicted value. 2(coefficient of determination), which is used to measure the goodness of fit of the model prediction results; use np.sqrt(mean_squared_error(y_true,y_pred)): to calculate RMSE (root mean square error), which is used to measure the error between the predicted value and the true value; finally, the results of the coefficient of determination and root mean square error are displayed on the right side of the interface.

[0154] The present invention addresses issues of device independence, as well as multimodal information acquisition, multimodal information fusion, and low model detection efficiency. To address this issue, the present invention integrates the near-infrared subsystem and electronic nose system of the intelligent seed vitality detection system within a single housing, placing them in close proximity. Simultaneously, both subsystems communicate with the data processing subsystem, transmitting data to it. The data processing subsystem then performs operations such as reading, preprocessing, feature extraction, and model building on the data.

[0155] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art may adjust the execution order of S1, S2, etc. according to actual conditions, which is also within the scope of protection of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0156] The above description is merely a preferred embodiment of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the present invention.

[0157] It should be noted that the terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and to define a specific order or precedence. Where appropriate, the order used for similar objects may be interchanged, such that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0158] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An intelligent detection system for germplasm vitality, characterized in that: Including data acquisition subsystem and data processing subsystem; The data acquisition subsystem is used to: collect near-infrared spectrum data and volatilization data of a plurality of seeds to be tested, and send them to the data processing subsystem; The data processing subsystem is used to process the near-infrared spectrum data and volatility data of multiple seeds to be tested using a target fusion method selected by the user to obtain a final seed vitality prediction result.

2. The intelligent detection system for germplasm vitality according to claim 1, characterized in that: The data acquisition subsystem includes a near-infrared subsystem and an electronic nose system. The near-infrared subsystem is used to collect near-infrared spectrum data of multiple seeds to be tested, and the electronic nose system is used to collect volatilization data of multiple seeds to be tested.

3. The intelligent detection system for germplasm vitality according to claim 2, characterized in that: The near-infrared subsystem and the electronic nose system are integrated into one chassis.

4. The intelligent detection system for germplasm vitality according to claim 2, characterized in that: The electronic nose system includes: a sample static bottle, a gas sensor array, a gas signal acquisition device and a gas signal processing device; The sample resting bottle is used to place at least one seed to be tested. The gas in each sample resting bottle is transported to the gas sensor array. The gas sensor array detects the composition of the received gas. The signal output by the gas sensor array is collected and converted by the gas signal acquisition device, and then amplified and filtered by the gas signal processing device to obtain volatilization data.

5. The intelligent detection system for germplasm vitality according to claim 2, characterized in that: The near-infrared subsystem includes: a spectrometer, a transmission optical fiber and an optical probe, and the spectrometer is connected to the optical probe; the optical probe obtains the original reflection signal of the seed to be tested and transmits it to the spectrometer via the transmission optical fiber. The spectrometer processes the received original reflection signal to obtain near-infrared spectrum data.

6. The intelligent detection system for germplasm vitality according to claim 1, characterized in that: The data acquisition subsystem is further used to: obtain volatile data, near infrared spectrum data and vitality label data of each sample seed in the sample germplasm resources, and obtain a sample seed information set of the sample germplasm resources; The data processing subsystem is further used to: after processing the sample seed information set of the sample germplasm resources using fusion algorithms at different levels, train the seed vitality prediction model corresponding to the fusion algorithm at each level.

7. The intelligent detection system for germplasm vitality according to claim 5, characterized in that: The data processing subsystem is also specifically used to: use the target fusion method selected by the user to process the near-infrared spectral data and volatile data of the multiple seeds to be tested, and combine the trained seed vitality prediction model corresponding to the target fusion method to obtain the final seed vitality prediction result of the seeds to be tested.

8. The intelligent detection system for germplasm vitality according to claim 7, characterized in that: When the target fusion method selected by the user is the first fusion method, the seed viability prediction model trained corresponding to the first fusion method is the first seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes: After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are spliced ​​to obtain the spliced ​​data of each seed to be tested, and a first seed vitality prediction result is obtained by using a first seed vitality prediction model, and the first seed vitality prediction result is used as the final seed vitality prediction result.

9. The intelligent detection system for germplasm vitality according to claim 8, characterized in that: When the target fusion method selected by the user is the second fusion method, the seed viability prediction model trained corresponding to the second fusion method is the second seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes: After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction and feature splicing are performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested respectively to obtain feature splicing data of each seed to be tested, and the second seed vitality prediction model is used to obtain a second seed vitality prediction result, and the second seed vitality prediction result is used as the final seed vitality prediction result.

10. The intelligent detection system for germplasm vitality according to claim 9, characterized in that: When the target fusion method selected by the user is the third fusion method, the trained seed viability prediction models corresponding to the third fusion method include a third seed vigor prediction model and a fourth seed vigor prediction model, and the process of the data processing subsystem obtaining the final seed vigor prediction result includes: After preprocessing the near-infrared spectral data and volatile data of multiple seeds to be tested, the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested are obtained, and feature extraction is performed on the preprocessed near-infrared spectral data of each seed to be tested and the preprocessed volatile data of each seed to be tested, respectively, to obtain the near-infrared spectral features of each seed to be tested and the volatile features of each seed to be tested. The near-infrared spectral features of each seed to be tested and the third seed vitality prediction model are used to obtain a third seed vitality prediction result. The volatile features of each seed to be tested and the fourth seed vitality prediction model are used to obtain a fourth seed vitality prediction result. The third seed vitality prediction model and the fourth seed vitality prediction result are fused to obtain a fifth seed vitality prediction result, and the fifth seed vitality prediction result is used as the final seed vitality prediction result.

Images