Transgenic rice seed sorting method and system based on fluorescence reaction

Abstract

The invention discloses a transgenic rice seed sorting method and system based on fluorescence reaction, relates to the technical field of sorting of seeds with seed coats, and solves the technical problem that the sorting precision is affected due to the fact that rice seed glumes are uneven in thickness and fixed pulse energy cannot effectively break walls of all rice seeds. According to the method, the glume thickness of the rice seeds in the seed queue is detected, and the corresponding pulse energy is matched to adjust the laser generator for wall breaking, so that the problem that the separation precision is affected due to the fact that the glume thickness of the rice seeds is not uniform and the fixed pulse energy cannot effectively break the wall of all the rice seeds is solved. Different from a fixed pulse energy wall breaking mode in the prior art, the method can adjust laser pulse energy according to the actual glume thickness of each rice seed, it is ensured that effective wall breaking can be achieved on rice seeds with glumes of different thicknesses, fluorescence labels are fully exposed for detection, and the transgenic rice seed sorting precision based on fluorescence reaction is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of seed coat sorting, and in particular to a transgenic rice seed sorting method and system based on fluorescence reaction. Background Art

[0002] Third-generation hybrid rice seeds are often harvested in a 1:1 ratio of red fluorescent transgenic seeds to colorless non-GMO seeds. The red fluorescent transgenic seeds are fertile and can be used for self-crossing and breeding; the colorless non-GMO seeds are male-sterile and can be used for seed production and planting. Third-generation hybrid rice seed sorting requires 100% high-purity sorting to completely separate transgenic and non-GMO seeds, ensuring the safety of non-GMO seed preparation and planting.

[0003] In existing seed sorting technologies, laser wall breaking technology is widely used to overcome the interference of rice husk on the accuracy of fluorescence detection. This technology creates a window in the husk of rice seeds, allowing the excitation light source to illuminate the fluorescent protein inside the seeds without obstruction. At the same time, it also prevents the fluorescence from being blocked by the husk, greatly improving the subsequent recognition of the fluorescence image, making seed sorting based on fluorescence reaction possible and more accurate. However, it faces many difficulties in practical application. Due to the complexity of the seed growth environment, such as differences in environmental factors such as light intensity, temperature, humidity, and soil fertility, the thickness of rice husk is not uniform. When using a fixed pulse energy for laser wall breaking, it is impossible to achieve effective wall breaking for all rice seeds.

[0004] The present invention provides a transgenic rice seed sorting method and system based on fluorescence reaction to solve the above technical problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method and system for sorting transgenic rice seeds based on fluorescence reaction. The present invention detects the hull thickness of rice seeds in the seed queue and adjusts the laser generator to break the hulls according to the corresponding pulse energy. This solves the problem that due to the uneven thickness of the rice hulls, the fixed pulse energy cannot effectively break the hulls of all rice seeds, thereby affecting the sorting accuracy. Unlike the fixed pulse energy breaking method in the prior art, this method can adjust the laser pulse energy according to the actual thickness of the hull of each rice seed, ensuring that rice seeds with different hull thicknesses can effectively break the hulls, allowing the fluorescent markers to be fully exposed for detection, greatly improving the sorting accuracy of transgenic rice seeds based on fluorescence reaction.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for sorting transgenic rice seeds based on fluorescence reaction, comprising:

[0007] Detecting the hull thickness of rice seeds in a seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on a conveying module;

[0008] Matching pulse energy corresponding to husk thickness; adjusting a laser generator based on the matched pulse energy to break rice seed walls; wherein the matched pulse energy is the optimal energy value for laser wall breaking;

[0009] The rice seeds after wall breaking are subjected to fluorescence excitation, and the rice seeds showing fluorescence reaction and those not showing fluorescence reaction are separated.

[0010] Preferably, the hull thickness of the rice seeds is detected by a laser displacement sensor, comprising:

[0011] The laser displacement sensor is placed perpendicular to the direction of rice seed movement and aimed at the rice seed surface at the guide rail exit at an incident angle of 45°. Each rice seed is tested several times to obtain a certain amount of light spot offset.

[0012] Formula based on triangle similarity principle Calculate the rice hull thickness d; where φ is the reflection angle, and the optical axis of the laser displacement sensor's receiver coincides with the normal to the rice seed surface, i.e., φ - θ; L is the baseline distance, θ is the laser incident angle; f is the receiver focal length, and Δx is the spot offset;

[0013] The average value of the rice hull thickness corresponding to several tests was taken as the rice hull thickness.

[0014] Preferably, the hull thickness of the rice seeds is detected by a laser displacement sensor and a hyperspectral camera, including:

[0015] The thickness of rice hulls was obtained by inversion using a hyperspectral camera, marked as d g The thickness of rice hull detected by the laser displacement sensor is marked as d c ;

[0016] By formula Fusion calculation of target rice hull thickness; where d is the hull thickness calculated by fusion, d c is the single point thickness value detected by the laser displacement sensor, d g is the global average thickness value inverted by the hyperspectral camera, w c is the weight coefficient of the laser displacement sensor data, d g is the weight coefficient of the hyperspectral camera data, and w c +w g =1,w c ≥w g .

[0017] Preferably, the weight coefficients of the laser displacement sensor data and the hyperspectral camera data are dynamically adjusted, including:

[0018] Extract the variance of the data collected by the laser displacement sensor and the hyperspectral camera, marked as σ c and σ g ;

[0019] By formula Calculate the weight coefficient w of laser displacement sensor data c , through the formula Calculate the weight coefficient w of hyperspectral camera data g , is a constant.

[0020] Preferably, obtaining the hull thickness of rice seeds by inverting with a hyperspectral camera includes:

[0021] Divide the glume range of rice seeds into several glume areas;

[0022] Controlling the hyperspectral camera to collect spectral reflectance data in line scanning mode, and extracting characteristic band spectral data from the spectral reflectance data; wherein the characteristic bands include cellulose sensitive bands: 1450nm and 2300nm, and lignin sensitive bands: 1650nm and 2050nm;

[0023] The spectral reflectance of the characteristic bands corresponding to several hull areas is input into a machine learning algorithm to obtain the thickness value, and the average of the thickness values of several hull areas is used as the hull thickness of the rice seeds; wherein the machine learning algorithm includes a partial least squares regression model.

[0024] Preferably, matching the pulse energy corresponding to the hull thickness includes:

[0025] The pulse energy with the best wall breaking effect for different rice varieties at different hull thicknesses was obtained through experimental data; the rice variety type, hull thickness, and pulse energy were associated into a set of data, thereby obtaining several sets of data;

[0026] Using several sets of data to train a machine learning algorithm, and using the trained machine learning algorithm as an energy matching model; wherein the machine learning algorithm includes a deep convolutional neural network model and a BP neural network model;

[0027] When matching pulse energy, the rice seed type and hull thickness are pre-processed and then input into the energy matching model to obtain the pulse energy.

[0028] Preferably, matching the pulse energy corresponding to the hull thickness includes:

[0029] The pulse energy with the best wall breaking effect for different rice varieties at different hull thicknesses was obtained through experimental data; the rice variety type, hull thickness, and pulse energy were associated into a set of data, thereby obtaining several sets of data;

[0030] A correlation matching table of rice seed type, husk thickness, and pulse energy is established based on several sets of data; when the pulse energy needs to be matched, the husk thickness is used to search in the correlation matching table to obtain the corresponding pulse energy.

[0031] Preferably, the rice seeds are broken based on a matched pulse energy modulated laser generator, comprising:

[0032] When the rice seeds in the seed queue enter the laser pre-aiming area;

[0033] Laser pulses are emitted to break the rice seed walls; wherein the pulse energy corresponding to the wall breaking is the pulse energy matched with the thickness of the rice seed husk.

[0034] Preferably, after matching the pulse energy corresponding to the hull thickness, the method further comprises:

[0035] The next rice seed to be broken is marked as the reference rice seed, and the pulse energy corresponding to the reference rice seed is used as the reference energy;

[0036] selecting, from candidate rice varieties, rice varieties whose pulse energies are closest to the reference energies as target rice varieties; wherein the candidate rice varieties are rice varieties subsequent to the reference rice varieties;

[0037] The target rice seeds are broken after the reference rice seeds, and when the reference rice seeds are broken, the target rice seeds are used as new reference rice seeds.

[0038] A second aspect of the present invention provides a rice seed sorting system based on fluorescence reaction, comprising:

[0039] Thickness detection module: used to detect the thickness of the hulls of rice seeds in the seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on the conveying module;

[0040] Sorting control module: used to match the pulse energy corresponding to the husk thickness; adjust the laser generator based on the matched pulse energy to break the rice seed wall; and,

[0041] It is used to excite fluorescence of rice seeds after wall breaking and separate rice seeds that show fluorescent reaction from those that do not.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. This invention detects the hull thickness of rice seeds in a seed array and adjusts the laser generator to break the hulls accordingly, addressing the problem of uneven hull thickness, which prevents fixed pulse energy from effectively breaking the hulls of all seeds and thus affects sorting accuracy. Unlike existing fixed pulse energy methods for hull breaking, this method adjusts the laser pulse energy based on the actual hull thickness of each rice seed, ensuring effective hull breaking for seeds of varying hull thicknesses and fully exposing the fluorescent marker for detection. This significantly improves the accuracy of fluorescence-based transgenic rice sorting, meeting the stringent requirement for 100% high-purity sorting of third-generation hybrid rice seeds.

[0044] 2. The present invention utilizes a laser displacement sensor and a hyperspectral camera to jointly detect the thickness of rice hulls, and calculates the thickness value by dynamically adjusting the weight coefficient fusion. The laser displacement sensor has high single-point detection accuracy, and the hyperspectral camera can achieve global distribution detection. The combination of the two has complementary advantages. When the surface of the rice seed is intact, the laser displacement sensor can accurately detect it; when the edge of the rice seed is thickened or the seed coat has cracks, the hyperspectral camera can make up for the shortcomings of the laser displacement sensor and correct the detection error. At the same time, the weight coefficient is dynamically adjusted according to the data variance, which can flexibly adapt to the thickness detection of different types of rice seeds, significantly enhancing the accuracy of hull thickness detection, and providing reliable data support for the subsequent precise matching of pulse energy and efficient sorting of rice seeds. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 Schematic diagram of the process of sorting transgenic rice seeds in the present invention;

[0047] Figure 2 This is a schematic diagram of the process of detecting the thickness of rice hulls based on the laser displacement sensor and the hyperspectral camera in the present invention;

[0048] Figure 3 Schematic diagram of the process of matching pulse energy based on machine learning algorithm in the present invention. DETAILED DESCRIPTION

[0049] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] Example 1:

[0051] See also Figure 1 The first embodiment of the present invention provides a method for sorting transgenic rice seeds based on fluorescence reaction, comprising:

[0052] S100: Detecting the hull thickness of rice seeds in a seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on a conveying module;

[0053] S200: Matching pulse energy corresponding to husk thickness; adjusting the laser generator based on the matched pulse energy to break the rice seed wall;

[0054] S300: Fluorescence excitation is performed on the rice seeds after the cell wall is broken, and the rice seeds showing fluorescence reaction and those not showing fluorescence reaction are separated.

[0055] The seed queue is formed by single rice seeds moving continuously and at intervals on the conveying module. The purpose is to ensure that the detection of husk thickness, laser wall breaking and laser fluorescence is not affected by seed overlap, thereby improving processing accuracy.

[0056] In S100 , the formation of the seed queue may be achieved through a vibration module.

[0057] The rice seeds fall from the discharge port into the starting section of the vibrating guide rail. The size of the discharge port is 1.5 times the average length of the rice seeds to avoid rice seed accumulation.

[0058] The rice seeds are pre-dispersed by a vibrator, so that a low-speed sliding flow is formed at the front end of the vibrating guide rail, which initially increases the distance between the rice seeds.

[0059] The vibration guide rail is a multi-stage structure, including a first-stage vibration guide rail, a second-stage vibration guide rail, etc. The first-stage vibration guide rail is mainly used for coarse separation of rice seeds, and the second-stage vibration guide rail is used for fine separation of rice seeds after coarse separation.

[0060] The working principle of the first-stage vibrating guide is to use the center of gravity of the overlapping seeds to offset, and when jumping, they are blocked by the side wall of the vibrating guide and rearranged in the return flow. The parameters of the first-stage vibrating guide are as follows:

[0061] 1. Guide rail parameters:

[0062] 1) Length: 1-1.5m, inclination angle α1 = 10°-15° (angle with the horizontal plane);

[0063] 2) Width: Width W1 = average seed width + 0.5 mm; for example, if the rice seeds are 2.5 mm wide, then W1 = 3 mm. The purpose is to force the rice seeds to be arranged horizontally.

[0064] 2. Vibration parameters: frequency f1 = 50-60 Hz, amplitude A1 = 1-2 mm; these vibration parameters can make the rice seeds jump forward, with each jump distance of 1-2 cm.

[0065] The secondary vibration guide removes overlapping seeds based on the processing of the primary vibration guide, forming a single rice seed queue with uniform spacing. The parameter settings of the secondary vibration guide are as follows:

[0066] 1. Guide rail parameters:

[0067] 1) Length: 0.5-0.8m, inclination angle α2 = 8°-10° (angle with the horizontal plane);

[0068] 2) Width: Width W2 = average seed width + (0.2-0.3) mm;

[0069] 3) Surface treatment: Surface roughness Ra = 0.8-1.6 μm, which can be achieved by polytetrafluoroethylene coating, in order to ensure low-speed sliding of single rice seeds.

[0070] 2. Key components:

[0071] 1) Elastic sorting teeth: These are set about 10mm apart on the surface of the secondary vibration guide rail, with a height of 1.5mm (the distance between the lower end and the secondary vibration guide rail surface) and an inclination of 45° (the angle with the secondary vibration guide rail). When rice seeds overlap, the tips of these teeth push back the upper layer of rice seeds.

[0072] 2) Height limit baffle: A height limit baffle is set 2.0-2.5 mm away from the surface of the secondary vibration guide rail; the average thickness of rice seeds is 1.5 mm, and the height limit baffle can only allow single rice seeds to pass through.

[0073] It should be noted that key components are generally installed on the secondary vibrating guide rail, but can also be installed on the subsequent guide rail or conveyor belt connected to the secondary vibrating guide rail. At least one of the elastic sorting teeth and the height limiting baffle is provided to limit the passage of overlapping rice seeds. Moreover, the heights of the elastic sorting teeth and the height limiting baffle are both set to prevent the passage of overlapping rice seeds. Therefore, the set height corresponds to the height between the lowest end of the elastic sorting teeth and the plane on which the rice seeds are located, or the height between the lower end of the height limiting baffle and the plane on which the rice seeds are located.

[0074] In S100, the detection of hull thickness can be achieved by a laser displacement sensor.

[0075] The laser displacement sensor detects the thickness of rice husks. The specific process is as follows:

[0076] The laser displacement sensor is set perpendicular to the direction of rice seed movement and aimed at the rice seed surface at the guide rail exit at an incident angle of 45°, ensuring that the laser spot falls on the flat area in the middle of the husk; and the height of the laser displacement sensor from the rice seed surface is 50-100 mm.

[0077] Before testing, the laser displacement sensor is zero-calibrated and its linearity verified. During testing, it is linked to the conveyor belt encoder, triggering a test at fixed intervals to ensure that each rice seed is tested at least 3-5 times.

[0078] Formula based on triangle similarity principle Calculate the rice hull thickness d; where φ is the reflection angle. Usually, the optical axis of the laser displacement sensor's receiver coincides with the normal line of the rice seed surface, i.e., φ - θ; L is the baseline distance, θ is the laser incident angle; f is the receiver focal length, and Δx is the spot offset.

[0079] The average thickness of a single rice seed is obtained by multiple measurements and calculations, and this average thickness is used as the hull thickness of the rice seed, marked as d c .

[0080] It should be noted that the baseline distance L and the receiver focal length f are typically determined by the laser displacement sensor manufacturer based on the device's optical design. Calibration is required before actual use. A calibration plate of standard thickness can be used, accurately measured using high-precision measuring equipment (such as an atomic force microscope or coordinate measuring machine).

[0081] In some other preferred embodiments, the detection of hull thickness can be achieved by combining a laser displacement sensor and a hyperspectral camera. In the aforementioned method, the hull thickness d is obtained by detecting the laser displacement sensor. c Based on the data from the literature, a hyperspectral camera was used to detect the thickness of the hulls of the same rice variety.

[0082] The hyperspectral camera detects the thickness of rice hulls. The specific process is as follows:

[0083] Divide the glume range of rice seeds into several glume areas;

[0084] The hyperspectral camera was set to a wavelength range of 900-1700 nm, covering the absorption peaks of cellulose and lignin; the spectral resolution was ≤5 nm, the spatial resolution was 6.5 μm / pixel, the exposure time was 50-100 ms, the gain was 10-20 dB, and the spectral signal-to-noise ratio was ≥20 dB.

[0085] As rice seeds pass through the camera's field of view on a conveyor belt, line scan mode is triggered to collect spectral reflectance data (size: 1024 × 512 pixels × 256 bands). After preprocessing, characteristic band spectral data is extracted from the spectral reflectance data. The characteristic bands include the cellulose-sensitive bands of 1450 nm (OH stretching vibration) and 2300 nm (CH bending vibration); and the lignin-sensitive bands of 1650 nm (aromatic ring vibration) and 2050 nm (CO stretching vibration). The preprocessing here mainly includes abnormal data detection and elimination.

[0086] The spectral reflectance of the characteristic band corresponding to each hull area is input into the machine learning algorithm to obtain the thickness value; the machine learning algorithm can predict the thickness value of each hull area, and then calculate the average thickness within each hull area as the hull thickness of the rice seed.

[0087] Taking the partial least squares regression model to predict the thickness of each hull area as an example, the following is the training process of the model;

[0088] 1. Data collection and integration:

[0089] 1) Experimental data were collected using a hyperspectral camera with a wavelength range of 900–1700 nm, covering the absorption peaks of cellulose and lignin, and a spectral resolution of 5 nm or less, which was used to capture details of characteristic peaks such as 1450 nm and 1650 nm.

[0090] 2) Perform radiometric calibration and noise correction on the experimental data, mainly including whiteboard correction, bad line repair, and spectral smoothing. These processing methods are well disclosed in existing solutions and will not be detailed here;

[0091] 3) Feature extraction:

[0092] Filter cellulose / lignin sensitive bands from 256 bands:

[0093] 1450nm (OH stretching vibration, cellulose)

[0094] 2300nm (CH bending vibration, cellulose)

[0095] 1650nm (aromatic ring vibration, lignin)

[0096] 2050 nm (CO stretching vibration, lignin);

[0097] Integrate the extracted features into a feature matrix: X∈R n×4 , n is the number of samples, and each row corresponds to the reflectance of 4 characteristic bands.

[0098] 2. Image segmentation and regional feature calculation:

[0099] 1) Glume ROI Extraction: The 900 nm (near-infrared) band of the hyperspectral data was converted to a grayscale image. The Otsu thresholding method was used to distinguish the glume from the background (stage). Internal holes were filled using dilation (kernel 3×3) and edge noise was removed using erosion (kernel 2×2) to obtain a binary mask. 10% of the samples were manually annotated, and the Dice coefficient (required to be ≥0.92) was calculated for verification.

[0100] 2) For each hull ROI, calculate the following parameters:

[0101] Average reflectivity: μ 1450 ,μ 1650 ,μ 2050 ,μ 2300 , a total of 4 dimensions;

[0102] Reflectance standard deviation: σ 1450 ,σ 1650 ,σ 2050 ,σ 2300 , a total of 4 dimensions;

[0103] Combine the mean reflectance and the standard deviation of reflectance into an 8-dimensional feature vector: [μ 1450 ,σ 1450 ,…,μ 2300 ,σ 2300 ].

[0104] 3. Calculate the true value of hull thickness and generate label data:

[0105] Using a laser confocal microscope (LCM) or a scanning electron microscope (SEM), the thickness of the hull of each rice seed was collected at five sites: site distribution: one point in the center and four points equally spaced around it (to avoid edge effects); true value calculation: y = mean(d1, d2, …, d5).

[0106] 4. The 8-dimensional feature vector corresponding to each hull ROI area and the true value of the hull thickness y are combined into a training data. Multiple hull areas of multiple rice varieties can correspond to several training data; these training data can be used to train the partial least squares regression model, and the model that meets the requirements can be saved.

[0107] It should be noted that when the input is 8-dimensional reflectance features (such as reflectance in the cellulose- and lignin-sensitive bands) and the output is the thickness of the central region of the rice hull, the model can output the thickness of each hull ROI during the prediction phase. The core logic lies in the spatial correspondence between features and labels and the model's generalization ability. Each hull ROI is an independent sample in the training data, and its label is the measured thickness of the central region of the ROI.

[0108] When combining a laser displacement sensor and a hyperspectral camera to measure rice hull thickness, the accuracy of both must be considered. Laser displacement sensors are generally considered more accurate for single-point detection, and hyperspectral camera-derived hull thickness data is used as a supplement to correct for errors in laser displacement sensor detection of specific rice varieties.

[0109] See also Figure 2 , based on the joint detection of rice seed hull thickness by laser displacement sensor and hyperspectral camera, the calculation process is as follows:

[0110] The thickness of the target rice hull was detected by laser displacement sensor and hyperspectral camera, marked as d. c and d g ; At the same time, the variance of the data collected by the laser displacement sensor and the hyperspectral camera is extracted, marked as σ c and σ g ;

[0111] By formula Fusion calculation of target rice hull thickness; where d is the hull thickness calculated by fusion, d c is the single point thickness value detected by the laser displacement sensor, d g is the global average thickness value inverted by the hyperspectral camera, w c is the weight coefficient of the laser displacement sensor data, w g is the weight coefficient of the hyperspectral camera data, And w c +w g =1; It is a constant, and its purpose is to prevent the calculation result of the weight coefficient from being invalid. Therefore, it is determined according to the fluctuation range and accuracy requirements of the actual data. like ).

[0112] Laser displacement sensors offer high single-point detection accuracy and strong real-time performance, but they can only detect local thickness and are significantly affected by the curvature of the rice seed surface. Hyperspectral cameras can achieve global distribution detection, but they obtain hull thickness through indirect inversion, resulting in lower detection accuracy and spectral noise. If the rice seed surface is intact, direct detection with a laser position sensor is theoretically sufficient. However, if the rice seed's edge is thickened, the single-point thickness detected by the laser displacement sensor will be significantly lower than the overall thickness of the rice seed. If the rice seed coat has cracks, the laser displacement sensor's detection at the crack will significantly decrease, while the hyperspectral camera's spectral reflectance data at the crack will show a sudden change. These issues can be addressed by calculating the rice seed hull thickness using the weighted fusion formula described above.

[0113] Because the single-point detection accuracy of the laser displacement sensor is higher, the hull thickness detected by it is used as the benchmark in the fusion calculation process, and the hull thickness detected by the hyperspectral camera is used as an auxiliary calibration. c In general, it is greater than w g , but exceptions are not excluded. Moreover, in the above fusion calculation formula, w c and w g It is dynamically adjusted through data variance, and the weight coefficient can be flexibly adjusted to adapt to the thickness detection of various types of rice seeds.

[0114] During the joint detection process of laser displacement sensors and hyperspectral cameras, the following points need to be noted:

[0115] 1. Both devices must be precisely synchronized in time to ensure that the same rice seed is detected and photographed at the same moment. This can be achieved through hardware triggering or software synchronization signals. For example, using the signal from a conveyor belt encoder to simultaneously trigger the laser displacement sensor for detection and the hyperspectral camera for imaging, ensuring that data from each rice seed is captured by both devices at the same location. Furthermore, the trigger frequency should be appropriately set based on the rice seed's movement speed and detection requirements. Each rice seed must be detected at least 3-5 times by the laser displacement sensor, while the hyperspectral camera also captures clear and representative images of the rice seed to avoid missing or duplicate data.

[0116] 2. During installation, ensure that the laser displacement sensor's detection point accurately overlaps with the hyperspectral camera's imaging area. This ensures that the hyperspectral camera captures the exact location of the rice seed where the laser displacement sensor measures husk thickness, typically the flat area in the middle of the husk. This ensures that both data correspond to the same rice seed location. Furthermore, the laser displacement sensor should be aimed at the rice seed surface at the guide rail exit at a 45° angle of incidence. The hyperspectral camera's imaging angle should avoid reflections and shadows that may affect image quality. The hyperspectral camera's position relative to the laser displacement sensor should also be considered to prevent mutual obstruction, ensuring accurate and reliable images and thickness data.

[0117] In S200, after the hull thickness of each rice seed in the seed queue is detected, the pulse energy of the laser generator can be adjusted according to the hull thickness. When the laser generator uses this pulse energy to break the rice seed wall, it can ensure that the wall breaking effect is just right.

[0118] The pulse energy of the laser generator is dynamically adjusted according to the thickness of the hull. The specific process is as follows:

[0119] After determining the hull thickness of the rice seeds, the corresponding pulse energy is matched with the hull thickness; when the rice seeds are within the working area of the laser generator, a laser with the corresponding pulse energy is emitted to break the rice seeds for subsequent fluorescence detection.

[0120] The hull thickness of rice varieties is not a standard thickness and fluctuates greatly. The hull thickness range of different rice varieties can be referred to in Table 1 below:

[0121] Table 1: Hull thickness range of different rice varieties

[0122] Rice type Husk thickness (unit: micrometer) Glume characteristics Japonica rice 10～30 Thick husk and high hardness Indica 5～20 Glumes are thin and soft glutinous rice 8～25 The thickness of the husk is between that of japonica rice and indica rice, and some varieties are slightly thicker. hybrid rice 8～22 Glume thickness is affected by parental genetics and has a wide range of variation.

[0123] As can be seen from the table above, different rice varieties have different hull thickness ranges, and there is some overlap between different rice varieties. Most importantly, each rice variety has distinct hull characteristics, so even with the same hull thickness, different pulse energies may be required.

[0124] The first method provided by the present invention for matching pulse energy according to hull thickness is as follows:

[0125] The pulse energy with the best wall breaking effect for different rice varieties at different hull thicknesses was obtained through experimental data; the rice variety type, hull thickness, and pulse energy were associated into a set of data, thereby obtaining several sets of data;

[0126] A correlation matching table of rice seed type, husk thickness, and pulse energy is established based on several sets of data; when the pulse energy needs to be matched, the husk thickness is used to search in the correlation matching table to obtain the corresponding pulse energy.

[0127] The pulse energy corresponding to the rice seed type and the hull thickness is matched by establishing an associated matching table. This does not require a large amount of experimental data and can quickly match the pulse energy.

[0128] It should be noted that the distance between the thickness of adjacent hulls of the same type of rice varieties in the experimental data may be large, which will make it difficult to match the pulse energy of the target thickness during the association matching table search process. At this time, the association matching table can be expanded by interpolation so that the accurate pulse energy can be found.

[0129] Optimal cell wall breaking performance refers to effective cell wall breaking to ensure excitation of the fluorescent marker while also ensuring no substantial damage to the rice seeds during the cell wall breaking process. Of course, in some cases, multiple pulse energies may meet this requirement. In such cases, any pulse energy can be selected, or the lowest pulse energy can be chosen to minimize the possibility of damage to the rice seeds.

[0130] See also Figure 3 The second method of matching pulse energy according to hull thickness is as follows:

[0131] Using the data obtained in the first method to train a machine learning algorithm, the trained machine learning algorithm is used as an energy matching model; wherein the machine learning algorithm includes a deep convolutional neural network model and a BP neural network model, and the model structure and nodes are set according to the data type;

[0132] When pulse energy matching is required, the rice seed type and hull thickness are pre-processed and then input into the energy matching model to obtain the pulse energy. The pre-processing here mainly involves model adaptation, including normalization.

[0133] Matching pulse energy with a machine learning algorithm is to learn the mapping relationship between rice seed types, husk thickness and their corresponding optimal pulse energy in experimental data through a machine learning algorithm. Once the model training meets the requirements, it is applicable to pulse energy matching for all rice seed types and all husk thickness ranges.

[0134] In S300 , the rice seeds after cell wall breaking are subjected to fluorescence excitation, and the rice seeds showing fluorescence reaction and the rice seeds without fluorescence reaction are separated.

[0135] When the rice seeds in the seed queue enter the laser pre-aiming area, the laser generator is controlled to emit laser pulses according to the pulse energy matching the thickness of the rice seed husk to break the rice husk; then the excitation light source is used to excite the fluorescent markers in the rice seeds, and the rice seeds are sorted according to the fluorescent markers.

[0136] The wall-breaking process of the laser generator is precisely controlled by a control unit that matches it. It also requires optical elements such as reflectors and lenses. Their function is to focus and guide the laser beam generated by the laser generator so that the laser can accurately act on specific parts of the rice seeds to improve the wall-breaking effect.

[0137] The laser generator needs to adjust the pulse energy in a short time, so a picosecond laser can be selected.

[0138] The fiber optic positioning sensor can be used to determine whether the rice seeds have entered the laser pre-aiming area. Of course, image recognition can also be used to determine whether the rice seeds have entered the laser pre-aiming area.

[0139] After the fluorescent markers on the broken rice seeds are excited by an excitation light source, only the fluorescently labeled seeds will show a fluorescent reaction; seeds without fluorescent markers remain unaffected. The image detection module collects image data of the excited rice seeds and determines whether they are fluorescent. This, combined with the sorting module, enables the sorting of fluorescent rice seeds. The sorting module can use either an air-blowing or a clamping method, and a red filter is required during the sorting process.

[0140] The excitation process of the excitation light source is also precisely controlled by the control unit that matches it, and data is exchanged with the control units of other devices to achieve precise excitation of rice seeds.

[0141] Example 2: Based on Example 1, this example performs a small-scale rearrangement of rice seeds to be broken to avoid frequent large-span adjustment of pulse energy by the laser generator, thereby extending the life of the laser generator.

[0142] Because rice husk thickness varies, it's necessary to detect the husk thickness and then adjust the pulse energy accordingly. The laser generator is then adjusted based on the pulse energy to emit laser pulses to break the rice husks. During this process, the pulse energy required to break the rice husks of the previous and next rice seeds may vary widely, such as 50mJ and 5mJ, respectively. In this case, the laser pulse energy generated by the laser generator needs to be quickly adjusted from 50mJ to 5mJ. This large-scale pulse energy adjustment can affect the lifespan of the laser generator.

[0143] After the hull thickness of each rice seed in the seed queue is detected, the hull thickness of all rice seeds is stored. During the laser wall breaking process, the next rice seed to be broken is used as the reference rice seed, and the corresponding pulse energy is used as the reference energy.

[0144] At least several rice seeds are selected from the seed queue (no rice seeds are stored in the temporary storage module at the beginning) as candidate rice seeds according to the rice wall breaking order, and the rice seed with pulse energy closest to the reference energy is selected from the candidate rice seeds as the target rice seed, and the target rice seed is used as the rice seed for rice wall breaking next to the reference rice seed.

[0145] Before the laser generator's cell-breaking work area, a temporary storage module is required to store rice seeds between the target and base rice seeds. When the base rice seeds undergo cell-breaking, the target rice seeds are automatically updated to the base rice seeds, and the base energy is also updated.

[0146] After the benchmark rice seeds and benchmark energy are updated, the rice seeds in the temporary wall-breaking module need to be taken into consideration when determining the alternative rice seeds, and the next target rice seeds should be comprehensively selected. This cycle can ensure that the adjustment amplitude of the pulse energy is as small as possible.

[0147] It should be noted that the temporary storage module can be installed on the side of the guide rail, such as by installing a temporary platform on the side of the guide rail. When rice seeds need to make way for target rice seeds, they are temporarily placed on the temporary platform. When the rice seeds on the temporary platform are selected as target rice seeds, they are then transferred from the temporary platform to the guide rail. The temporary platform is suitable for storing multiple rice seeds. If the number of rice seeds to be stored is small, such as a single seed, a robotic arm can be used instead of the temporary platform to pick up the rice seeds for temporary storage.

[0148] In addition to setting up a temporary storage module to rearrange the rice seeds, the rice seeds can also be directly rearranged using a robotic arm or other equipment.

[0149] In order to provide enough operating time for temporary storage of rice seeds, multiple rice seeds can be selected as target rice seeds at the same time. When these multiple target rice seeds are broken into pieces, enough time can be left to determine the next batch of target rice seeds.

[0150] It is worth noting that the rice seeds in the temporary storage module will be returned to the seed queue. If the distance between adjacent rice seeds in the seed queue is too small, overlap may occur when the rice seeds return to the seed queue.

[0151] If the temporary storage module only stores one rice seed, the spacing between adjacent rice seeds in the seed queue must be at least greater than the length of one rice seed (this length is the maximum length counted). If the temporary storage module can store multiple target rice seeds, the spacing between adjacent rice seeds in the seed queue must be increased. In extreme cases, all rice seeds in the temporary storage module must be returned to the seed queue at the same time.

[0152] A second embodiment of the present invention provides a transgenic rice seed sorting system based on fluorescence reaction, comprising:

[0153] Thickness detection module: used to detect the thickness of the hulls of rice seeds in the seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on the conveying module;

[0154] Sorting control module: used to match the pulse energy corresponding to the husk thickness; adjust the laser generator based on the matched pulse energy to break the rice seed wall; and,

[0155] It is used to excite fluorescence of rice seeds after wall breaking and separate rice seeds that show fluorescent reaction from those that do not.

[0156] The thickness detection module collects data using a laser displacement sensor and a hyperspectral camera, then processes this data to determine the hull thickness of each rice seed in the seed queue. The sorting control module interacts with the thickness detection module, matching pulse energy to the hull thickness of the rice seeds to break the rice hulls. The excitation light source is then controlled to excite fluorescence in the broken rice seeds. Finally, images are captured and used in conjunction with the sorting module to separate the rice seeds with and without fluorescent markers.

[0157] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for sorting transgenic rice seeds based on fluorescence reaction, characterized in that: include: Detecting the hull thickness of rice seeds in a seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on a conveying module; Matching the pulse energy corresponding to the hull thickness; adjusting the laser generator based on the matched pulse energy to break the rice seed wall; wherein the matched pulse energy is the optimal energy value for laser wall breaking; The rice seeds after wall breaking are subjected to fluorescence excitation and the rice seeds are separated according to whether a fluorescence reaction occurs.

2. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: The hull thickness of the rice seeds is detected by a laser displacement sensor, comprising: The laser displacement sensor is placed perpendicular to the direction of rice seed movement and aimed at the rice seed surface at the guide rail exit at an incident angle of 45°. Each rice seed is tested several times to obtain a certain amount of light spot offset. Formula based on triangle similarity principle Calculate the rice hull thickness d; where φ is the reflection angle, and the optical axis of the laser displacement sensor's receiver coincides with the normal to the rice seed surface, i.e., φ - θ; L is the baseline distance, θ is the laser incident angle; f is the receiver focal length, and Δx is the spot offset; The average value of the rice hull thickness corresponding to several tests was taken as the rice hull thickness.

3. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: The hull thickness of the rice seeds is detected by a laser displacement sensor and a hyperspectral camera, including: The thickness of rice hulls was obtained by inversion using a hyperspectral camera, marked as d g The thickness of rice hull detected by the laser displacement sensor is marked as d c ; By formula Fusion calculation of target rice hull thickness; where d is the hull thickness calculated by fusion, d c is the single point thickness value detected by the laser displacement sensor, d g is the global average thickness value inverted by the hyperspectral camera, W c is the weight coefficient of the laser displacement sensor data, d g is the weight coefficient of the hyperspectral camera data, and W c +w g =1,w c ≥w g .

4. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 3, characterized in that: The weight coefficients of the laser displacement sensor data and the hyperspectral camera data are dynamically adjusted, including: Extract the variance of the data collected by the laser displacement sensor and the hyperspectral camera, marked as σ c and σ g ; By formula Calculate the weight coefficient w of laser displacement sensor data c , through the formula Calculate the weight coefficient w of hyperspectral camera data g ,∈―is a constant.

5. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 3, characterized in that: The method of obtaining the hull thickness of rice seeds by inverting the hull thickness of rice seeds through a hyperspectral camera includes: Divide the glume range of rice seeds into several glume areas; Controlling the hyperspectral camera to collect spectral reflectance data in line scanning mode, and extracting characteristic band spectral data from the spectral reflectance data; wherein the characteristic bands include cellulose sensitive bands: 1450nm and 2300nm, and lignin sensitive bands: 1650nm and 2050nm; The spectral reflectance of the characteristic bands corresponding to several hull areas is input into a machine learning algorithm to obtain the thickness value, and the average of the thickness values of several hull areas is used as the hull thickness of the rice seeds; wherein the machine learning algorithm includes a partial least squares regression model.

6. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: Matching the pulse energy corresponding to the hull thickness includes: The pulse energy with the best wall breaking effect for different rice varieties at different hull thicknesses was obtained through experimental data; the rice variety type, hull thickness, and pulse energy were associated into a set of data, thereby obtaining several sets of data; Using several sets of data to train a machine learning algorithm, and using the trained machine learning algorithm as an energy matching model; wherein the machine learning algorithm includes a deep convolutional neural network model and a BP neural network model; When matching pulse energy, the rice seed type and hull thickness are pre-processed and then input into the energy matching model to obtain the pulse energy.

7. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: Matching the pulse energy corresponding to the hull thickness includes: The pulse energy with the best wall breaking effect for different rice varieties at different hull thicknesses was obtained through experimental data; the rice variety type, hull thickness, and pulse energy were associated into a set of data, thereby obtaining several sets of data; A correlation matching table of rice seed type, husk thickness, and pulse energy is established based on several sets of data; when the pulse energy needs to be matched, the husk thickness is used to search in the correlation matching table to obtain the corresponding pulse energy.

8. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: Based on the matching pulse energy modulation laser generator, the rice seed wall is broken, including: When the rice seeds in the seed queue enter the laser pre-aiming area, laser pulses are emitted to break the rice seed walls; wherein the pulse energy corresponding to the wall breaking is the pulse energy matched with the thickness of the rice seed husk.

9. The method for sorting transgenic rice seeds based on fluorescence reaction according to claim 1, characterized in that: After matching the pulse energy corresponding to the hull thickness, the method further includes: The next rice seed to be broken is marked as a reference rice seed, and the pulse energy corresponding to the reference rice seed is used as the reference energy; selecting, from candidate rice varieties, rice varieties whose pulse energies are closest to the reference energies as target rice varieties; wherein the candidate rice varieties are rice varieties subsequent to the reference rice varieties; The target rice seeds are broken after the reference rice seeds, and when the reference rice seeds are broken, the target rice seeds are used as new reference rice seeds.

10. A transgenic rice seed sorting system based on fluorescence reaction, used to implement the transgenic rice seed sorting method based on fluorescence reaction according to any one of claims 1 to 9, characterized in that: It includes a thickness detection module and a sorting control module, and the thickness detection module and the sorting control module exchange data; Thickness detection module: used to detect the thickness of the hulls of rice seeds in the seed queue; wherein the seed queue is formed by single rice seeds moving continuously and at intervals on the conveying module; A sorting control module is used to match the pulse energy corresponding to the husk thickness; adjust the laser generator based on the matched pulse energy to break the rice seed wall; and It is used to excite fluorescence of rice seeds after wall breaking and separate rice seeds that show fluorescent reaction from those that do not.

Images