Intelligent detection and screening system for combined harvesting quality of forage grass and grass seeds

Abstract

The invention relates to the technical field of physical measurement and data processing, in particular to an intelligent detection and screening system for combined harvesting quality of forage grass and grass seeds. An excitation and acquisition module; the signal processing module is used for separating single seed data from the mixed signal, extracting high-discrimination features and outputting an acoustic feature vector of each seed; a classification and sorting module; wherein the preprocessing module is connected with the excitation and acquisition module through seed matrix coding information, the excitation and acquisition module is connected with the signal processing module through multichannel time domain waveform data, and the signal processing module is connected with the classification and sorting module through acoustic feature vectors. By means of the combined harvesting quality intelligent detection and screening system, the efficiency and quality of forage grass seed harvesting are remarkably improved, the loss rate is reduced, various technical problems faced by the current industry are effectively solved, and therefore a solid foundation is laid for sustainable development of the forage grass industry.

Description

Technical Field

[0001] The invention relates to the technical field of physical measurement and data processing, and in particular to an intelligent detection and screening system for the quality of combined harvesting of forage grass seeds. Background Art

[0002] Seed vitality detection and efficient harvesting are key links in improving the quality and efficiency of modern agriculture, and directly affect the sustainable development of the forage industry. Due to the influence of high-altitude, mountainous and other geographical environments, mechanized harvesting of forage seeds faces severe challenges.

[0003] In addition, auxiliary technologies such as chemical detection and machine vision are difficult to cooperate with mechanized harvesting due to problems such as destructiveness and environmental sensitivity. For example, the Kjeldahl method requires crushing seeds, which cannot achieve non-destructive testing; image segmentation technology (such as patent CN112258456A) relies on germination pretreatment, and the misjudgment rate is as high as 15%. Although acoustic detection technology (such as patent CN118874278A) is mature in the industrial field, its frequency range (20-100 kHz) is mismatched with the acoustic impedance of seeds, making it difficult to characterize physiological indicators such as cell activity. Summary of the invention

[0004] Based on the above purpose, the present invention provides a combined harvesting quality intelligent detection and screening system for forage grass seeds, the system comprising: The pre-processing module is used to realize batch loading, position calibration and environmental stabilization of grass seeds, and output matrix samples of single seeds occupying independent grids and seed matrices with position coding; The excitation and acquisition module is used to perform frequency-band acoustic excitation on the matrixed samples and collect mixed signals, and output multi-channel time-domain waveform data; The signal processing module is used to separate the single seed data from the mixed signal and extract the high-discrimination features, and output the acoustic feature vector of each seed; The classification and sorting module is used to determine the vitality level of multiple seeds based on feature vectors and perform accurate sorting, and output the sorted seed collection and full-process detection report; The preprocessing module is connected to the excitation and acquisition module through seed matrix coding information, the excitation and acquisition module is connected to the signal processing module through multi-channel time domain waveform data, and the signal processing module is connected to the classification and sorting module through acoustic feature vectors.

[0005] Preferably, the preprocessing module comprises: The vibration spreading and grid positioning unit is used to pour the grass seeds to be tested into the vibration platform and evenly distribute them through the groove grid. The pressure sensor embedded in the bottom detects overlapping seeds and automatically removes them, and outputs a matrix sample with a single seed occupying an independent grid; Optical coordinate calibration and ID allocation unit, used to capture seed distribution images and locate seed center coordinates, generate unique IDs and establish coordinate-ID mapping tables, and output seed matrices and coordinate databases with position codes; An environmental parameter control unit is used to maintain stable environmental parameters through temperature and humidity sensors and light adjustment equipment, and output detection samples with stable environmental parameters; The output of the vibration spreading and grid positioning unit is connected to the input of the optical coordinate calibration and ID allocation unit, and the output of the optical coordinate calibration and ID allocation unit is connected to the input of the environmental parameter control unit.

[0006] Preferably, the excitation and acquisition module comprises: A frequency division multiplexing sweep signal design unit is used to divide the 10-50 kHz frequency band into N sub-frequency bands according to the seed matrix encoding information and allocate them to each seed, and output N groups of frequency band-seed binding relationship tables; The time-division multiplexing excitation unit is used to transmit the sweeping signals of each sub-band in a polling manner according to the time slice, dynamically adjust the transmission power, and output the excitation signal sequence with precise timing; A multi-channel signal acquisition unit is used to collect mixed signals using a circular microphone array, ensure that the clock deviation is less than 1μs through a hardware synchronization circuit, and output multi-channel time domain waveform data; The output of the frequency division multiplexing sweep signal design unit is connected to the input of the time division multiplexing excitation unit, and the output of the time division multiplexing excitation unit is connected to the input of the multi-channel signal acquisition unit.

[0007] Preferably, the signal processing module includes: A digital filtering and signal separation unit is used to apply a FIR bandpass filter group to the multi-channel mixed signal and use a beamforming algorithm to enhance the target seed signal, and output a separated single seed time domain signal set; GPU accelerated spectrum analysis unit, used to call CUDA library to execute FFT in parallel and select Blackman-Harris window to reduce sidelobe interference, and output high-precision frequency domain spectrum set; A multi-dimensional feature calculation unit is used to calculate the main resonance peak, Q factor, attenuation coefficient and other features based on the spectrum and attenuation segment fitting, and output the acoustic feature vector of each seed; The output of the digital filtering and signal separation unit is connected to the input of the GPU accelerated spectrum analysis unit, and the output of the GPU accelerated spectrum analysis unit is connected to the input of the multi-dimensional feature calculation unit.

[0008] Preferably, the classification and sorting module comprises: The feature standardization and enhancement unit is used to load the pre-stored standardization parameters by species, perform Z-score normalization, add interactive features, and output the enhanced feature matrix; The principal component analysis dimensionality reduction unit is used to calculate the universal PCA model across varieties and project it into a 3D principal component space, outputting a low-dimensional feature subset; The parallel SVM classification unit is used to call the pre-trained multi-classification SVM model to batch output category probabilities and set confidence thresholds, and output a set of vitality level labels with confidence levels; A sorting instruction generation unit, used to convert the activity level label and coordinate data into a driving signal for the nozzle array and encode it, and output a nozzle trigger instruction queue; An air circuit control unit, used to dynamically adjust the air pressure and control the air circuit response time according to the seed weight data, and output a controlled air flow pulse sequence; The sorting execution and data archiving unit is used to control the nozzle array to execute sorting and write the test data to the SQL database, and output the sorted seed set and the full process test report; Among them, the output of the feature normalization and enhancement unit is connected to the input of the principal component analysis dimensionality reduction unit, the output of the principal component analysis dimensionality reduction unit is connected to the input of the parallel SVM classification unit, the output of the parallel SVM classification unit is connected to the input of the sorting instruction generation unit, the output of the sorting instruction generation unit is connected to the input of the gas path control unit, and the output of the gas path control unit is connected to the input of the sorting execution and data archiving unit.

[0009] Preferably, in the vibration spreading and grid positioning unit, the amplitude of the vibration platform is 0.5 mm, the frequency is 30 Hz, the size of the groove grid is 3×3 mm, the range of the pressure sensor is 0-10 g, and when the pressure is detected to be greater than 0.5 g, an alarm is triggered and overlapping seeds are automatically removed.

[0010] Preferably, in the optical coordinate calibration and ID allocation unit, the resolution of the CMOS camera is 1280×960, and the Canny edge detection algorithm is used to locate the seed center coordinates with an accuracy of ±0.05 mm. A unique ID is generated according to the grid row and column numbers and a coordinate-ID mapping table is established.

[0011] Preferably, in the frequency division multiplexing sweep signal design unit, the 10-50 kHz frequency band is divided into N sub-bands, each sub-band width is 40 kHz / N, and the sweep rate is adaptively adjusted according to the seed density, and the formula is: , is the seed density of the ith sub-band, is the sweep rate of the ith sub-band, is the proportionality constant.

[0012] Preferably, in the multi-dimensional feature calculation unit, the main resonance peak recognition searches for the value corresponding to the maximum amplitude value in the sub-frequency band. , the Q factor is calculated based on the 3 dB bandwidth method with an accuracy of ±0.1. The attenuation coefficient fitting performs nonlinear least square fitting on the attenuation segment of the time domain signal, and the iteration termination condition is that the residual is less than 1e-6.

[0013] Preferably, in the sorting execution and data archiving unit, the nozzle array sprays synchronously within a 10 ms time window, the airflow cone angle is 30°, and the coverage area is 3 mm in diameter. The detection data (feature value, classification result) of each seed is written into the SQL database, and the batch number and timestamp are associated.

[0014] Beneficial effects of the present invention: 1. The present invention significantly improves the mechanized harvesting efficiency of forage grass seeds by introducing an integrated system of combined harvesting and intelligent detection. Compared with traditional manual harvesting, the efficiency of mechanized harvesting is significantly improved, which can effectively reduce labor costs and labor intensity, solve the current problem of only 5 mu / person / day in labor efficiency, and effectively improve labor productivity.

[0015] 2. The loss rate caused by traditional harvesting methods is as high as 30%. The present invention can accurately distinguish high-quality and low-quality seeds through the real-time monitoring and screening capabilities of the intelligent detection system, thereby reducing the loss rate. The system can effectively identify and exclude unqualified seeds, ensure the quality of the harvested seeds, and in turn promote the sustainable development of grassland and animal husbandry.

[0016] 3. The non-destructive testing technology used in the present invention can avoid the destructive effects of traditional chemical testing methods such as Kjeldahl nitrogen determination, while maintaining the accuracy and reliability of the test. Through non-destructive testing, seeds can be kept in their natural state, avoiding waste of resources.

[0017] 4. In view of the characteristics of high-altitude and mountainous environments, the system is designed to operate stably in complex geographical environments in special areas such as Tibet, thereby ensuring the feasibility of harvesting. The system has excellent environmental adaptability and can effectively overcome the impact of the external environment on harvesting efficiency.

[0018] 5. Through optimized image processing and acoustic detection technology, the system can reduce the misjudgment associated with traditional detection technology, solve the problem of up to 15% misjudgment rate of image segmentation technology, ensure the accuracy and reliability of data results, and improve the accuracy of evaluation and screening.

[0019] 6. The intelligent data processing and monitoring module in the system can analyze and feedback the collected data in real time, which helps to make quick decisions and improve management efficiency. At the same time, the association of batch number and timestamp facilitates subsequent tracking and data management. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 Schematic diagram of the system framework of the system of the present invention; Figure 2 is a schematic diagram of a pre-processing module in the system of the present invention; Figure 3 Schematic diagram of the classification and sorting module in the system of the present invention. DETAILED DESCRIPTION

[0022] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0023] See also Figure 1-Figure 3 The embodiment of the present invention provides a combined harvest quality intelligent detection and screening system for forage grass seeds, which is composed of multiple modules, and each module is connected through a specific information flow and data interface to achieve efficient seed detection and sorting functions. The following is a description of the connection relationship, detailed steps and implementation methods of each module and its beneficial effects.

[0024] First, the main function of the pre-processing module is to load grass seeds in batches, perform position calibration, and maintain the stability of environmental parameters. The module is equipped with a vibration spreading and grid positioning unit and an optical coordinate calibration unit. Through this system, grass seeds are evenly distributed into independent grids and each seed is assigned unique position information (position coding). In this way, the output "seed matrix" not only ensures the independence of each seed, but also provides the necessary position information for use by subsequent modules.

[0025] Furthermore, the excitation and acquisition module performs frequency-band acoustic excitation on each grass seed based on the seed matrix encoding output by the preprocessing module. This excitation process dynamically maps each seed to a specific acoustic frequency band, collects the returned mixed signal through the circular microphone array, and outputs multi-channel time domain waveform data. This process ensures that each seed can receive excitation under optimal conditions through reasonable timing control and frequency allocation, thereby improving the quality and accuracy of the collected signal.

[0026] On this basis, the signal processing module separates the independent signal of each seed from the multi-channel time domain waveform data obtained by the excitation and acquisition module, and uses highly optimized digital filtering technology and spectrum analysis to extract high-discrimination features. Finally, the acoustic feature vector is output, which carries the unique acoustic characteristics of each seed. In this way, the system can accurately reflect the differences in quality of each seed and provide basic data for subsequent classification.

[0027] Finally, the classification and sorting module uses the acoustic feature vector output by the signal processing module to determine the vitality level. Through machine learning algorithms (such as support vector machines SVM), the system classifies the vitality level of each seed and generates corresponding nozzle drive instructions. The instructions guide the nozzle to perform precise sorting and separate seeds of different grades. Finally, the sorted seed collection and full-process detection report output by the system provide a basis for subsequent operations such as fertilization and sowing.

[0028] The main benefit of this information flow and data processing method between modules is that it improves the overall detection efficiency and accuracy, enabling the system to quickly and effectively process large quantities of grass seeds, and has good flexibility and scalability to adapt to the detection needs of grass seeds of different varieties and characteristics. At the same time, precise control of each link also effectively reduces errors and improves the overall quality and harvesting efficiency of seeds.

[0029] In a possible implementation, each subunit of the preprocessing module is connected through a specific output and input relationship to ensure the efficiency and accuracy of the grass seed detection process. The following is a detailed description of the connection relationship between each unit in the preprocessing module, a specific analysis of the steps and their beneficial effects.

[0030] First of all, the main responsibility of the vibration spreading and grid positioning unit is to evenly distribute the grass seeds to be tested on the vibration platform. Through the design of the groove grid, the seeds can be automatically arranged under the action of vibration to avoid stacking and mutual occlusion. The embedded pressure sensor can monitor the overlap of seeds in real time to ensure a stable single-grain distribution. Once overlapping seeds are found, the system will automatically start the rejection mechanism to remove the overlapping seeds and finally output a matrixed sample. Each grass seed occupies an independent grid position and its position is marked.

[0031] Next, the output of the vibration spreading and grid positioning unit is directly connected to the input of the optical coordinate calibration and ID assignment unit. This unit is responsible for taking images of the distribution of grass seeds and automatically identifying and locating the center coordinates of each seed using image processing technology. For each seed, the system generates a unique identification ID and establishes this information into a coordinate-ID mapping table. In this way, the output seed matrix containing position coding and the corresponding coordinate database ensure that each seed has a unique identity tag, which is convenient for further processing by subsequent modules.

[0032] Subsequently, the output of the optical coordinate calibration and ID allocation unit is connected to the input of the environmental parameter control unit. The main function of this unit is to maintain the stability of the detection environment. The current environmental conditions are monitored by temperature and humidity sensors, and the light adjustment device is used to provide the seeds with appropriate light to ensure that the grass seeds are in the best detection state. Finally, the unit outputs the detection sample with stable environmental parameters to ensure that the subsequent excitation, acquisition and signal processing can be carried out under suitable conditions.

[0033] Through step-by-step processing and dynamic monitoring, we ensure that each grass seed is carefully screened and marked before entering the subsequent process, reducing classification errors. The flow of the environmental parameter control unit enables the system to maintain the reliability of its test results in different operating environments and adapt to a variety of laboratory conditions. The automated design of the overall process reduces manual intervention, improves the efficiency of grass seed detection and screening, and meets the needs of large-scale testing.

[0034] In one possible implementation, first, the main task of the frequency division multiplexing sweep signal design unit is to divide the frequency band (10-50 kHz) into N sub-bands according to the seed matrix encoding information. Each sub-band is assigned to a specific grass seed, which ensures that each seed receives a unique frequency signal during the excitation process. The unit outputs N sets of frequency band-seed binding relationship tables in the form of a clear mapping to ensure that each seed corresponds to a specific frequency.

[0035] This output is then connected to the input of the time-division multiplexing excitation unit. This unit is responsible for polling and transmitting the sweeping signals of each sub-band according to the time slice. In this process, the time-division multiplexing excitation unit can not only excite each seed at a fixed time interval according to the binding relationship, but also dynamically adjust the transmission power to adapt to the characteristics and requirements of different seeds. The output excitation signal sequence ensures the timing accuracy during the excitation process and avoids interference between different signals.

[0036] Furthermore, the output of the time-division multiplexing excitation unit is connected to the input of the multi-channel signal acquisition unit. The unit collects mixed signals from each seed in real time through a circular microphone array. During the signal acquisition process, the design of the hardware synchronization circuit can ensure that the clock deviation of the acquisition is less than 1μs, thereby ensuring high accuracy and synchronization of signal acquisition. The multi-channel signal acquisition unit finally outputs multi-channel time domain waveform data, which contains signal information from different frequency bands, making subsequent analysis and processing more reliable and efficient.

[0037] The frequency division multiplexing design ensures that different seeds can receive unique excitation signals, enhancing the pertinence and accuracy of the detection. The dynamic adjustment function of the time division multiplexing excitation unit can adapt to the response characteristics of different seeds, avoid interference between signals, and make the excitation process more efficient. The multi-channel signal acquisition unit has high synchronization accuracy, ensuring the integrity and accuracy of the signal, and providing a reliable foundation for subsequent data analysis. The design and implementation of the entire module greatly improves the automation level and operating efficiency of the system, and realizes efficient and accurate grass seed excitation and signal acquisition. Through this modular design, the system can flexibly respond to different experimental needs and provide better experimental plans and results.

[0038] In a possible implementation, first, the digital filtering and signal separation unit is the front-end processing part of the system, which is responsible for receiving the mixed signal output from the multi-channel signal acquisition unit. This unit uses a FIR (finite impulse response) bandpass filter group for preliminary signal processing, which can effectively filter out noise and interference signals that are not in the frequency band of interest. In this process, combined with the beamforming algorithm, the target seed signal of a specific direction and frequency can be enhanced, and a separated single seed time domain signal set can be output. Such signal enhancement can improve the accuracy of subsequent analysis, so that the target signal can be extracted more clearly.

[0039] Furthermore, the output of the digital filtering and signal separation unit is connected as input to the GPU accelerated spectrum analysis unit. The core function of this unit is to call the CUDA library and use modern GPUs for fast parallel processing to achieve fast Fourier transform (FFT). Through FFT, the time domain signal is converted into a frequency domain signal, and the Blackman-Harris window function is used to reduce the sidelobe interference in the spectrum. This process ensures the clarity and accuracy of the spectrum. Finally, the GPU accelerated spectrum analysis unit outputs a high-precision frequency domain spectrum set.

[0040] Furthermore, the output of the GPU-accelerated spectrum analysis unit is connected to the multi-dimensional feature calculation unit. Based on the previously obtained spectrum data, this unit calculates the acoustic features such as the main resonance peak, Q factor and attenuation coefficient by fitting the attenuation segment. These features are key parameters for characterizing the acoustic characteristics of each seed. Finally, the output acoustic feature vector of each seed will be used for subsequent analysis and classification.

[0041] Through digital filtering and beamforming, the target signal is fully enhanced, reducing the impact of background noise, so that subsequent analysis can focus more on useful signals. Using GPU acceleration technology, the execution time of FFT is greatly shortened, and frequency domain features can be quickly acquired to meet real-time or semi-real-time application requirements. The multi-dimensional feature calculation unit ensures that the obtained acoustic feature vector is credible and has certain physical meaning through precise fitting, providing a reliable basis for subsequent intelligent analysis and pattern recognition. The clear functional division and connection relationship between each unit make the system have good scalability and maintainability, which is convenient for future optimization and upgrading.

[0042] In a possible implementation, the feature standardization and enhancement unit is the first step of the entire classification and sorting module. It is responsible for loading the pre-stored standardization parameters and applying Z-score normalization to the input feature data. This process can effectively eliminate the differences between different varieties due to different dimensions, making the feature data comparable. At the same time, the unit also further enhances the expressive power of the features by adding interactive features, ensuring that the system can better capture the true characteristics of each seed. The output of this unit is the enhanced feature matrix.

[0043] The output of the feature normalization and enhancement unit is connected to the input of the principal component analysis dimensionality reduction unit. The purpose of principal component analysis (PCA) is to project the high-dimensional feature matrix into a low-dimensional space, usually a three-dimensional principal component space, to achieve dimensionality reduction and feature compression, thereby reducing the computational burden while retaining as much information as possible. The output is a low-dimensional feature subset, which simplifies the subsequent calculation process.

[0044] The output of the principal component analysis dimensionality reduction unit is used as input to the parallel SVM classification unit. The parallel SVM classification unit uses a pre-trained multi-classification support vector machine (SVM) model to classify the low-dimensional feature subset and output the probability of each category. By setting the confidence threshold, it can be ensured that only vitality level labels with confidence higher than the preset value are returned, which helps to improve the accuracy and reliability of classification.

[0045] The vitality level label and the corresponding coordinate data are input to the sorting instruction generation unit, which generates and encodes the drive signal for controlling the nozzle array based on this information and outputs the nozzle trigger instruction queue. This ensures that each seed can be efficiently sorted according to its classification results.

[0046] The output of the sorting instruction generating unit is connected to the air path control unit, which is responsible for dynamically adjusting the air pressure according to the weight data of the seeds and controlling the response time of the air path to generate an air flow pulse sequence suitable for seeds of different weights and types.

[0047] The output of the gas circuit control unit is connected to the sorting execution and data archiving unit, which controls the actual sorting process of the nozzle array and records the relevant test data into the SQL database after sorting. Finally, it outputs the sorted seed set and the full process test report to ensure the tracking and recording of the entire sorting action.

[0048] The output of each unit provides data in a suitable format for the next unit, ensuring smooth data flow. Through the functional sharing of each unit, the entire system can not only respond quickly to input, but also provide higher efficiency in high-dimensional data processing. Through standardization, dimensionality reduction, classification and other steps, the accuracy and operability of classification are improved, so that the final sorting can better meet actual needs. The combination of airflow control and nozzle command generation can respond to classification results in real time, realize fast sorting operations, and adapt to the stringent requirements of modern agriculture for efficiency.

[0049] In one possible implementation, the vibration platform is the core component of the system, which is set to an amplitude of 0.5 mm and a frequency of 30 Hz. Its function is to produce a uniform spreading effect through vibration, evenly distributing the input seeds on the groove grid. The control system of the vibration platform receives data from upstream modules (such as classification and sorting modules) to ensure that the seeds can enter the vibration spreading unit in time for the next step of processing after being classified.

[0050] Furthermore, a grid of grooves measuring 3×3 mm is placed above the vibration platform, which is designed to accommodate the spread seeds. The size of the grooves is precisely calculated to ensure that each groove can only accommodate one seed, thereby reducing overlap. Due to the operation of the vibration platform, the seeds will naturally slide into the corresponding grooves at this stage according to gravity and vibration forces.

[0051] Furthermore, the pressure sensor has a range of 0-10 g. When a seed enters a groove and triggers the sensor, the sensor monitors the pressure data in each groove in real time. If the pressure is greater than 0.5 g, it means there are duplicate or overlapping seeds. At this time, the sensor immediately triggers an alarm signal and sends a command to the vibration platform through the control system to start the rejection mechanism.

[0052] When the pressure sensor detects overlapping seeds, the system automatically starts the rejection mechanism, driving a set of well-designed mechanical arms or pneumatic devices to automatically remove the overlapping seeds. This mechanism ensures the separation of high-quality seeds and the improvement of the yield rate, avoiding the problems caused by overlap in subsequent processing.

[0053] This combination of vibration spreading and grid positioning reduces manual intervention, making seed spreading more efficient, ensuring that each seed can be accurately positioned, and reducing human errors. The design and implementation of the vibration spreading and grid positioning unit helps to improve the efficiency and accuracy of the overall system in seed processing, and is an indispensable and important part of modern seed sorting and processing equipment. Through the real-time monitoring and alarm mechanism of the pressure sensor, overlapping seeds can be detected and removed in time, avoiding the impact of overlap on the accuracy of classification and subsequent operations. The system greatly improves the degree of automation of operations through the automatic rejection function, reducing the workload of operators.

[0054] In one possible implementation, a CMOS camera with a resolution of 1280×960 is used in the system to collect images of seeds. The camera is mounted above the vibration platform and can capture images of the doorway area in real time. Its high resolution ensures that the details of the image are captured, thereby improving the ability to accurately locate the seeds.

[0055] The acquired image is sent to the image processing module, where the Canny edge detection algorithm is used. This algorithm can effectively extract edge information from the image and identify the outline of the seed. By detecting the edge of the seed, the center coordinates of each seed can be accurately located, and the processed data can reach an accuracy of ±0.05 mm. This accuracy ensures the accuracy of subsequent operations and avoids processing problems caused by positioning errors.

[0056] Furthermore, after edge detection, the system calculates the row and column number of each seed in the grid from the center coordinates. These row and column numbers form a unique index, and combined with the specific coordinates, the system generates a unique ID for each seed. This ID includes not only the row and column information of the seed, but also other identification information, such as timestamps or batch numbers, to reduce confusion in subsequent processing.

[0057] Furthermore, the corresponding coordinates and the generated unique ID will be recorded in the coordinate-ID mapping table, which plays a fundamental role in the system. The mapping table allows the subsequent processing modules to quickly find the ID information of each seed, facilitating data tracking, quality traceability, and subsequent sorting and classification.

[0058] The introduction of the Canny edge detection algorithm, combined with a high-resolution camera, greatly enhances the ability to accurately identify seed positions, providing reliable data support for subsequent operations. By establishing a coordinate-ID mapping table, the system can achieve more efficient management and scheduling, reduce the burden on operators, and improve the automation of the entire system. The unique ID design for each seed enables easy data tracking in the subsequent process, enhances product traceability, and helps quality control. Automatic identification and ID allocation reduce manual operations, reduce errors caused by human factors, and ensure accurate processing of each seed.

[0059] In a possible implementation, the implementation process of the frequency division multiplexing sweeping signal design unit is divided into the following steps, the main purpose of which is to achieve efficient identification and precise control of seeds through the division of frequency bands and adaptive adjustment of the sweeping rate.

[0060] Specifically, the frequency band division module is responsible for dividing the 10-50 kHz frequency band into N sub-bands, and the width of each sub-band is In practical applications, this module uses digital signal processing (DSP) technology to implement frequency band division by performing fast Fourier transform (FFT) on the input signal and generate corresponding frequency band signals.

[0061] The seed density measurement module monitors the seed density x in real time through an integrated sensor. Commonly used sensors include piezoelectric sensors or laser range sensors, which can collect density data at a high frequency. These data will be transmitted to the control module in real time to provide a basis for adjusting the sweep rate.

[0062] This module uses the formula based on the measurement results of seed density. ; is the seed density of the ith sub-band, is the sweep rate of the ith sub-band, is the proportionality constant, (where ) Dynamically adjust the scanning rate. This module receives the density data from the sensing module and calculates the required scanning rate.

[0063] Further, the frequency signal output by the module is adjusted according to the sweep rate, and this module is responsible for generating and transmitting the adaptive sweep signal. The signal generator uses a high-frequency clock source to ensure the accuracy of the frequency conversion.

[0064] Furthermore, the final signal is received and spectrum analysis is performed, and the analysis results are sent back to the feedback module to adjust the parameters of signal transmission to reduce interference and improve recognition accuracy.

[0065] After dividing the frequency band into multiple sub-bands, the system can send and receive multiple signals simultaneously within the same spectrum range, thereby improving spectrum utilization efficiency and reducing interference between signals. The system adjusts the scanning rate according to the seed density measured in real time. This flexibility enables it to maintain optimal working performance in different environments and seed types, ensuring rapid signal recognition and processing. Through real-time monitoring and adjustment, the system can better adapt to environmental changes and improve the accuracy and reliability of seed detection, especially in cases of large density changes. The system optimizes parameters through a feedback mechanism, which can identify potential problems in advance, reduce errors, and improve the stability of the overall system.

[0066] In a possible implementation, the dynamic characteristics of the system are accurately extracted by identifying the main resonance peak and calculating the Q factor, as well as fitting the attenuation coefficient.

[0067] Specifically, the signal input module is responsible for receiving signal data from the signal receiving and processing module, and the signal is usually a modulated and frequency-division-multiplexed signal. This module sends the signal to the multi-dimensional feature calculation unit.

[0068] Within the sub-band, the main resonance peak identification module converts the time domain signal into the frequency domain signal through the fast Fourier transform (FFT), thereby identifying the resonance frequency corresponding to the maximum amplitude. This process includes a peak detection algorithm that can accurately find the main resonance peak in each sub-band.

[0069] The Q factor calculation module calculates the Q factor based on the 3 dB bandwidth method. It first determines the main resonant frequency The amplitude at the point where the 3 dB bandwidth is measured is the frequency range where the amplitude drops to half of the maximum value. The Q factor is calculated with an accuracy of ±0.1. This module is closely connected with the main formant identification module to ensure timely update of frequency data.

[0070] The attenuation coefficient fitting module uses the nonlinear least squares method to fit the attenuation segment of the time domain signal to determine the attenuation coefficient. By analyzing the attenuation characteristics of the signal, a mathematical model is established for fitting. During iteration, the termination condition is set to a residual less than 1e-6 to ensure the accuracy and reliability of the fitting results.

[0071] Finally, all the calculation results (resonance frequency, Q factor and attenuation coefficient) will be summarized in the output module and passed to the subsequent data processing or decision module. In addition, the output results can also be used to adjust the system parameters to achieve adaptive optimization.

[0072] By finding peaks within the sub-band and accurately calculating the Q factor, the accuracy of signal characteristic analysis can be significantly improved, helping to identify and distinguish different types of signals. The analysis of the signal attenuation segment by nonlinear least squares fitting enables the system to extract dynamic features, thereby gaining an in-depth understanding of the various influencing factors in the signal transmission process, such as environmental changes or material properties. By setting the iteration termination conditions, the system can adapt to different signal characteristics and changes, making the calculation process more flexible and adaptive. The construction of this multi-dimensional feature calculation unit improves the overall efficiency of signal processing and accelerates the speed of feature extraction through the division of labor among multiple modules, and is suitable for real-time systems.

[0073] In a possible implementation, the system first receives data from the front-end detection module. These data include the characteristic values ​​of each seed (such as size, shape, color, etc.) and the classification results (such as whether the seed is qualified). The signal receiving module prepares these data for subsequent processing.

[0074] The nozzle array consists of multiple nozzles that spray synchronously within a 10 ms time window. The cone angle of the airflow is 30°, which can accurately spray within a diameter area of ​​about 3 mm to ensure that every seed is covered. The control module adjusts the nozzle's spray frequency and spray intensity in real time based on sensor feedback information to ensure accuracy and efficiency.

[0075] After receiving the feature values ​​and classification results of each seed, the module will perform data processing, including data validation and cleaning, to ensure that the collected data is complete and reliable. After processing, the data will be ready to be associated with the batch number and timestamp.

[0076] The processed seed test data will be sent to the SQL database. The data archiving module is responsible for writing the feature values, classification results, batch numbers and timestamps into the database. This step usually involves SQL INSERT operations, which can easily retrieve and analyze data by establishing a clear database structure.

[0077] This module monitors the real-time status of the entire jetting and data writing process to ensure that every link of the system can operate efficiently. At the same time, if an error or abnormality occurs, an alarm will be issued in time to prevent data loss or inconsistency.

[0078] Through the synchronous injection of the nozzle array, the system can process multiple seeds quickly and effectively within a very short time window, significantly improving the classification efficiency and adapting to the needs of high-throughput processing. The design of the airflow cone angle and coverage diameter can ensure that every seed can be within the injection range while reducing waste. This design increases the flexibility of system processing, especially when facing seeds of different sizes and shapes. Associating the feature values, classification results with batch numbers and timestamps will help with subsequent data management and traceability. This design ensures the integrity and consistency of seed classification data, facilitating subsequent analysis and decision-making. The system can be dynamically adjusted based on real-time feedback from data processing and classification, which improves the stability and adaptability of the overall system and can better meet the ever-changing market needs.

[0079] Through reasonable module design and efficient workflow, the sorting execution and data archiving unit can achieve efficient and accurate seed classification and reliable data storage, providing important support for agricultural production and scientific research.

[0080] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0081] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An intelligent detection and screening system for the quality of combined harvesting of forage grass seeds, characterized in that: The system comprises: The pre-processing module is used to realize batch loading, position calibration and environmental stabilization of grass seeds, and output matrix samples of single seeds occupying independent grids and seed matrices with position coding; The excitation and acquisition module is used to perform frequency-band acoustic excitation on the matrixed samples and collect mixed signals, and output multi-channel time-domain waveform data; The signal processing module is used to separate the single seed data from the mixed signal and extract the high-discrimination features, and output the acoustic feature vector of each seed; The classification and sorting module is used to determine the vitality level of multiple seeds based on feature vectors and perform accurate sorting, and output the sorted seed collection and full-process detection report; The preprocessing module is connected to the excitation and acquisition module through seed matrix coding information, the excitation and acquisition module is connected to the signal processing module through multi-channel time domain waveform data, and the signal processing module is connected to the classification and sorting module through acoustic feature vectors.

2. The system according to claim 1, characterized in that The preprocessing module comprises: The vibration spreading and grid positioning unit is used to pour the grass seeds to be tested into the vibration platform and evenly distribute them through the groove grid. The pressure sensor embedded in the bottom detects overlapping seeds and automatically removes them, and outputs a matrix sample with a single seed occupying an independent grid; Optical coordinate calibration and ID allocation unit, used to capture seed distribution images and locate seed center coordinates, generate unique IDs and establish coordinate-ID mapping tables, and output seed matrices and coordinate databases with position codes; An environmental parameter control unit is used to maintain stable environmental parameters through temperature and humidity sensors and light adjustment equipment, and output detection samples with stable environmental parameters; The output of the vibration spreading and grid positioning unit is connected to the input of the optical coordinate calibration and ID allocation unit, and the output of the optical coordinate calibration and ID allocation unit is connected to the input of the environmental parameter control unit.

3. The system according to claim 1, characterized in that The excitation and acquisition module comprises: A frequency division multiplexing sweep signal design unit is used to divide the 10-50 kHz frequency band into N sub-frequency bands according to the seed matrix encoding information and allocate them to each seed, and output N groups of frequency band-seed binding relationship tables; The time-division multiplexing excitation unit is used to transmit the sweeping signals of each sub-band in a polling manner according to the time slice, dynamically adjust the transmission power, and output the excitation signal sequence with precise timing; A multi-channel signal acquisition unit is used to collect mixed signals using a circular microphone array, ensure that the clock deviation is less than 1μs through a hardware synchronization circuit, and output multi-channel time domain waveform data; The output of the frequency division multiplexing sweep signal design unit is connected to the input of the time division multiplexing excitation unit, and the output of the time division multiplexing excitation unit is connected to the input of the multi-channel signal acquisition unit.

4. The system according to claim 1, characterized in that The signal processing module comprises: A digital filtering and signal separation unit is used to apply a FIR bandpass filter group to the multi-channel mixed signal and use a beamforming algorithm to enhance the target seed signal, and output a separated single seed time domain signal set; GPU accelerated spectrum analysis unit, used to call CUDA library to execute FFT in parallel and select Blackman-Harris window to reduce sidelobe interference, and output high-precision frequency domain spectrum set; A multi-dimensional feature calculation unit is used to calculate the main resonance peak, Q factor, and attenuation coefficient characteristics based on the spectrum and attenuation segment fitting, and output the acoustic feature vector of each seed; The output of the digital filtering and signal separation unit is connected to the input of the GPU accelerated spectrum analysis unit, and the output of the GPU accelerated spectrum analysis unit is connected to the input of the multi-dimensional feature calculation unit.

5. The system according to claim 1, characterized in that The classification and sorting module includes: The feature standardization and enhancement unit is used to load the pre-stored standardization parameters by species, perform Z-score normalization, add interactive features, and output the enhanced feature matrix; The principal component analysis dimensionality reduction unit is used to calculate the universal PCA model across varieties and project it into a 3D principal component space, outputting a low-dimensional feature subset; The parallel SVM classification unit is used to call the pre-trained multi-classification SVM model to batch output category probabilities and set confidence thresholds, and output a set of vitality level labels with confidence levels; A sorting instruction generation unit, used to convert the activity level label and coordinate data into a driving signal for the nozzle array and encode it, and output a nozzle trigger instruction queue; An air circuit control unit, used to dynamically adjust the air pressure and control the air circuit response time according to the seed weight data, and output a controlled air flow pulse sequence; The sorting execution and data archiving unit is used to control the nozzle array to execute sorting and write the test data to the SQL database, and output the sorted seed set and the full process test report; Among them, the output of the feature normalization and enhancement unit is connected to the input of the principal component analysis dimensionality reduction unit, the output of the principal component analysis dimensionality reduction unit is connected to the input of the parallel SVM classification unit, the output of the parallel SVM classification unit is connected to the input of the sorting instruction generation unit, the output of the sorting instruction generation unit is connected to the input of the gas path control unit, and the output of the gas path control unit is connected to the input of the sorting execution and data archiving unit.

6. The system according to claim 2, characterized in that In the vibration spreading and grid positioning unit, the amplitude of the vibration platform is 0.5 mm, the frequency is 30 Hz, the size of the groove grid is 3×3 mm, the range of the pressure sensor is 0-10g, and when the pressure is detected to be greater than 0.5 g, an alarm is triggered and overlapping seeds are automatically removed.

7. The system according to claim 2, characterized in that In the optical coordinate calibration and ID allocation unit, the resolution of the CMOS camera is 1280×960, and the Canny edge detection algorithm is used to locate the seed center coordinates with an accuracy of ±0.05mm. A unique ID is generated according to the grid row and column numbers and a coordinate-ID mapping table is established.

8. The system according to claim 3, characterized in that In the frequency division multiplexing sweep signal design unit, the 10-50 kHz frequency band is divided into N sub-bands, each sub-band width is 40 kHz / N, and the sweep rate is adaptively adjusted according to the seed density, and the formula is: , is the seed density of the ith sub-band, is the sweep rate of the ith sub-band, is the proportionality constant.

9. The system according to claim 4, characterized in that In the multi-dimensional feature calculation unit, the main resonance peak recognition searches for the value corresponding to the maximum amplitude in the sub-frequency band. , the Q factor is calculated based on the 3 dB bandwidth method with an accuracy of ±0.

1. The attenuation coefficient fitting performs nonlinear least square fitting on the attenuation segment of the time domain signal, and the iteration termination condition is that the residual is less than 1e-6.

10. The system according to claim 5, characterized in that In the sorting execution and data archiving unit, the nozzle array sprays synchronously within a 10 ms time window, the airflow cone angle is 30°, and the coverage area is 3 mm in diameter. The detection data (feature value, classification result) of each seed is written into the SQL database, and the batch number and timestamp are associated.

Images