Intelligent corn tassel removing system

Abstract

The invention discloses an intelligent corn tassel removing system, and belongs to the technical field of corn castration, the intelligent corn tassel removing system comprises a castration mechanism used for removing corn tassels, two castration wheels of each castration wheel set are coaxially provided with spiral wheels, and the corn tassels entering a feeding area between two shafts can be forcibly and accurately fed between the castration wheels, so that the corn tassels can be accurately removed. The tassel identification module is used for carrying out tassel identification and height calculation, separating corn tassels sequentially entering a feeding area to prevent the multiple tassels from being fed into a tassel removal wheel at the same time, successfully pulling out the tassels, effectively protecting corn leaves, reducing damage and remarkably improving the complete tassel extraction rate, and the control mechanism is used for adjusting the height of the tassel in real time; according to the corn tassel emasculation device, the positions of corn tassels are dynamically recognized in the operation process, the height displacement of the tassels is output, accurate servo control is achieved, the emasculation accuracy and efficiency are improved, effective support is provided for mechanical production of corn for seed production, and good application prospects are shown.

Description

Technical Field

[0001] The invention relates to the technical field of corn detasseling, and in particular to an intelligent corn tassel removal system. Background Art

[0002] There are two main ways to detassel corn: cutting and extraction. Cutting detasseling uses the cutting action of a tool to remove the corn tassels and the accompanying stems and leaves; extraction detasseling uses the rotation of a roller to clamp the corn tassels and separate them from the corn plant. The existing seed corn detasseling machines have a low detasseling rate and cause great damage to corn leaves, resulting in the failure of corn detasseling machines to be widely used, and manual detasseling is still the main method. Therefore, an intelligent corn tassel removal system is urgently needed to address the problems of low tassel removal efficiency and great damage to corn leaves during mechanized seed production. Summary of the invention

[0003] The purpose of the present invention is to provide a high-efficiency, low-damage corn tassel intelligent removal system, which can effectively solve the problems existing in the above-mentioned prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: a corn tassel intelligent removal system, comprising a tassel extraction mechanism for corn tassel removal, the tassel extraction mechanism comprising a bracket, and a plurality of tassel removal wheel groups mounted on the bracket, each of the tassel removal wheel groups comprising two tassel removal wheels, each of the tassel removal wheels being configured with an independent drive assembly, the drive assembly controlling the relative rotation of the two tassel removal wheels, and the system further comprising:

[0005] The tassel recognition module includes a tassel recognition model and a depth camera. The tassel recognition model executes the following instructions based on the image captured by the depth camera:

[0006] Construct a target box with tassels in the image;

[0007] Extract the spatial coordinates and orientation information of the top of the tassel in the target frame to form the tassel position data;

[0008] Output the height displacement of the tassel and feed back the height displacement data to the control mechanism;

[0009] The control mechanism adjusts the height of the male wheel in real time based on the feedback data.

[0010] Preferably, the detasseling wheel is an inflatable member and the drive assembly is a motor drive assembly.

[0011] Preferably, the bracket is a parallelepiped structure, and the control mechanism includes a hydraulic cylinder, which is located below the parallelepiped structure, and the hydraulic cylinder is configured to drive the parallelepiped structure to move the detasseling wheel up and down while ensuring that the angle between the detasseling wheel axis and the ground remains unchanged.

[0012] Preferably, a spiral wheel is coaxially mounted on one end of each detasseling wheel facing the traveling direction, and the two spiral wheels cooperate to drive the male ears to be fed into the corresponding two detasseling wheels one by one and at intervals.

[0013] Preferably, an independent bearing support structure is provided on the bracket at each detasseling wheel, and the bearing support structure is used to assemble and support the wheel axle. One end of the wheel axle is connected to the detasseling wheel, and the other end is connected to the driving assembly. A V-shaped plate is installed between the two bearing support structures corresponding to each group of detasseling wheels.

[0014] Preferably, the tassel recognition model is trained using an established data set and evaluated using a cross-validation method, and the parameters of the tassel recognition model are dynamically optimized based on the evaluation results.

[0015] Preferably, the data set includes corn images at different growth stages and under various environmental conditions in the agricultural image database, corn images taken on-site in corn fields, and tassel images in corn images annotated using LabelMe.

[0016] Preferably, the marking includes marking the tassel boundary box, and setting tassel labels including unobstructed, semi-obstructed and fully-obstructed.

[0017] Preferably, in the process of evaluating the tassel recognition model, the predicted frame coordinate loss is established, and the loss is calculated by the overlapping area, center point distance and aspect ratio of the predicted frame and the real frame, specifically:

[0018]

[0019]

[0020] Among them, B gt represents the real box, B pred represents the prediction box, ρ 2 (B gt ,B pred ) represents the Euclidean distance between the center points of the predicted box and the true box, C represents the diagonal length of the minimum bounding rectangle that can contain the predicted box and the true box, α is the balance parameter, and v is the aspect ratio metric function used to measure the consistency of the height-to-width ratio.

[0021] Beneficial effects: In the present invention, the two detasseling wheels of each detasseling wheel group are coaxially mounted with a spiral wheel, which can force and accurately feed the corn tassels entering the feeding area between the two shafts into the detasseling wheels, and at the same time separate the corn tassels that enter the feeding area successively to prevent multiple tassels from being fed into the detasseling wheels at the same time, so that the tassels can be successfully plucked out, and at the same time effectively protect the corn leaves, reduce damage, and significantly improve the extraction rate of the tassels. The detasseling wheel is an inflatable component, which reduces damage to the corn plant, and the detasseling wheel adopts a motor drive assembly, which more effectively drives the detasseling wheel to achieve the required pulling force of the corn tassels.

[0022] In the present invention, a tassel recognition module is provided, which can not only accurately and effectively identify the tassels in different situations through the optimized tassel recognition model, but also cooperate with the depth camera and the control mechanism to capture corn images at high speed, ensuring that the tassels can be clearly identified and the position information can be obtained during fast movement; the top spatial coordinates and orientation information of the tassels are extracted to form posture data; the posture information obtained by the identification is fed back to the tassel removal device control mechanism in real time, so that the position of the corn tassels can be dynamically identified during the operation process, and the height displacement of the tassels can be output, thereby realizing precise servo control and improving the accuracy and efficiency of emasculation; it provides effective support for the mechanized production of seed corn and shows good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0024] In the attached picture:

[0025] Figure 1 It is a flow chart of the intelligent corn tassel removal system of the present invention;

[0026] Figure 2 It is a physical structural schematic diagram of the male extraction mechanism and the control mechanism of the present invention;

[0027] Figure 3 It is a structural schematic diagram of the male extraction mechanism and the control mechanism of the present invention;

[0028] Figure 4 is a schematic diagram of a corn image of the present invention;

[0029] Figure 5 is a schematic diagram of a corn image labeled by the present invention;

[0030] Figure 6 This is a diagram of the corn test set recognition result of the present invention;

[0031] Figure 7 It is a confusion matrix diagram of corn image training of the present invention;

[0032] Figure 8 It is the F1-confidence curve diagram of the present invention;

[0033] Fig. 9 It is the labels diagram of the present invention;

[0034] Fig.10 It is the PR curve diagram of the present invention;

[0035] Fig.11 is a Results graph of the present invention;

[0036] Fig.12 This is a simulation result diagram of the male extraction mechanism of the present invention;

[0037] Numbers in the figure: 1. parallel four-bar structure; 2. hydraulic cylinder; 3. bearing support structure; 4. wheel axle; 5. detasseling wheel; 6. reducer; 7. drive assembly; 8. V-plate; 9. spiral wheel. DETAILED DESCRIPTION

[0038] The embodiments of the present invention are described below in conjunction with the drawings in the embodiments of the present invention. The terms used in the implementation mode of the present invention are only used to explain the specific embodiments of the present invention, and are not intended to limit the present invention. The embodiments of the present application are described below in conjunction with the drawings.

[0039] Example: Figure 1 As shown, a corn tassel intelligent removal system includes a tassel extraction mechanism for corn tassel removal, a tassel identification module for tassel identification and height calculation, and a control mechanism for real-time adjustment of the height of the tassel wheel;

[0040] For male extraction mechanisms, refer to Figure 2-Figure 3 As shown, it includes a bracket, and a plurality of detasseling wheel groups mounted on the bracket, each of which includes two detasseling wheels, and each detasseling wheel is equipped with an independent driving assembly 7, which controls the relative rotation of the two detasseling wheels 5;

[0041] As for the detasseling wheel 5, in this embodiment, the detasseling wheel is an inflatable component, which can reduce the damage to the corn plants;

[0042] As for the driving assembly 7, the driving assembly used in this embodiment is a motor driving assembly, which is more effective in driving the detasseling wheel to achieve the required pulling force of the corn tassel compared with the existing pneumatic drive and hydraulic drive;

[0043] Based on the above-mentioned drive components, in a specific case, the experimental materials were samples of seed corn plants taken from the Shihu Kiln test site. The corn plants were in the emasculation period, and the average height was about 1550mm. For seed corn plants, the tassel removal force was tested at three growth stages: the tassel had not yet emerged (stage I), the tassel had just emerged (stage II), and the tassel had emerged but had not yet shed powder (stage III). In each growth stage, 10 corn plants (excluding corn in the field area) were randomly selected for testing. The results of the removal force and cross-sectional area are shown in the following table:

[0044]

[0045] According to the test data in the table, the maximum pulling force Fmax of corn tassels is 95N; the initial selected wheelbase of the two-wheel tassel extraction is 250mm, and the force arm L when extracting tassels is 125mm; therefore, the maximum extraction torque required is:

[0046]

[0047] Considering the difference of maize mechanical parameters and the uncertainty of friction, the working condition coefficient K is taken as 1.5 and the safety factor S is taken as 2; therefore, the calculation formula for calculating the torque T is:

[0048] T 计算 =T·K·S=11.9×1.5×2=35.7N·m;

[0049] Each detasseling module is equipped with two motors, each of which provides a torque of about 18Nm; the speed of the detasseling wheel is 500r / min, so a servo motor with a speed of 2000r / min is selected, and a planetary gear reducer with a reduction ratio of 4:1 is equipped. The output shaft of the reducer 6 can output a torque of 24N·m;

[0050] For the bracket: the bracket is a parallelogram structure 1, and an independent bearing support structure 3 is provided on the bracket at each detasseling wheel. The bearing support structure is used to assemble and support the wheel shaft 4. One end of the wheel shaft 4 is connected to the detasseling wheel 5, and the other end is connected to the driving assembly 7. A V-shaped plate 8 is installed between the two bearing support structures corresponding to each group of detasseling wheels; the strength of the bracket is improved, and the V-shaped plate 8 can guide and protect the male ear to swing toward the detasseling wheel.

[0051] Based on the above, further, in reference Figure 3As shown, a spiral wheel 9 is coaxially installed at one end of each detasseling wheel 5 facing the travel direction, and the two spiral wheels 9 work together to drive the male ears to be fed one by one and at intervals into the corresponding two detasseling wheels 5; the corn male ears entering the feeding area between the two axes can be forced to be accurately fed into the detasseling wheels 5, and at the same time, the corn male ears that enter the feeding area successively are separated to prevent multiple male ears from being fed into the detasseling wheels 5 at the same time, so that the male ears are successfully pulled out, and at the same time, the corn leaves are effectively protected, the damage is reduced, and the extraction rate of the male ears is significantly improved.

[0052] The tassel recognition module includes a tassel recognition model and a depth camera. The tassel recognition model executes the following instructions based on the image taken by the depth camera:

[0053] Construct a target box with tassels in the image;

[0054] Extract the spatial coordinates and orientation information of the top of the tassel in the target frame to form the tassel position data;

[0055] Output the height displacement of the tassel and feed back the height displacement data to the control mechanism;

[0056] The control mechanism adjusts the height of the male wheel in real time based on the feedback data;

[0057] For the tassel recognition model: the tassel recognition model uses the established data set for model training and is evaluated through a cross-validation method, and the parameters of the tassel recognition model are dynamically optimized based on the evaluation results;

[0058] Establish a data set: The data set includes corn images in different growth periods and under various environmental conditions in the agricultural image database, as well as corn images taken on-site in corn fields. The images of corn in actual environments with different lighting, angles, and occlusions greatly enhance the diversity and practicality of the data, and the male ear images in corn images annotated with LabelMe; Reference Figure 4 Shown are images of corn from two sources;

[0059] refer to Figure 5 As shown in the figure, the tassel image in the corn image annotated with LabelMe: In addition to annotating the bounding box, the pixels on the image are also annotated, which can be used to create datasets required for tasks such as semantic segmentation and instance segmentation, allowing users to draw complex polygonal areas by clicking points on the image for more accurate annotation, and providing a dataset browser where users can view and manage annotated data;

[0060] The annotation includes annotating the tassel boundary box and setting tassel labels including unobstructed, semi-obstructed and fully obstructed. This not only realizes large-scale data collection, but also can effectively identify and measure the tassels that have not yet fully grown.

[0061] Convert the processed data set into txt format through python for model training:

[0062] Constructing the corn target detection experimental operating environment, in a specific case: based on the cloud computing technology experimental platform built by Anaconda, GPU is used to accelerate the training speed of the model. The specific software and hardware environment is shown in the following table:

[0063]

[0064] Based on the YOLOv8 network, the loss function used for corn tassel target detection is determined: CIoU (complete intersection over union) Loss is used; by establishing the predicted bounding box coordinate loss, the loss is calculated by the overlapping area, center point distance and aspect ratio of the predicted box and the real box, specifically:

[0065]

[0066] Among them, B gt represents the real box, B pred represents the prediction box, ρ 2 (B gt ,B pred ) represents the Euclidean distance between the center points of the predicted box and the true box, C represents the diagonal length of the minimum bounding rectangle that can contain the predicted box and the true box, α is the balance parameter, and v is the aspect ratio metric function used to measure the consistency of the height-to-width ratio;

[0067] Based on the data set, the model is trained by dividing it into a training set and a validation set, and the accuracy and robustness of the model are evaluated by the cross-validation method;

[0068] In a specific case: Based on the above, open the file named "coco128.yaml" in YOLOv8 in pycharm. This file is a data file, which mainly records the location of the data set and the names of all trained objects; change the paths of the training set and validation set to the paths where the images are located;

[0069] Open the "train" file in pycharm and set the configuration of the training model: use "YOLOv8s" (Weight) for training weight; set the number of training rounds (Epoch) to 300 rounds;

[0070] The YOLOv8 target detection model is used to detect corn targets. The data set format is YOLOv8, with the suffix txt. The data set has a total of 947 images, including 805 training sets and 142 validation sets. A "weight" weight folder is obtained by training the YOLOv8 model. "best.pt" is used to perform target detection tests on 200 corn images in the test set. The python program in YOLOv8 is used and the parameters are changed to the path of the test set.

[0071] Recognition is performed based on the above trained model; the results are as follows Figure 6 As shown in the figure, the detection time of corn tassels does not exceed 0.343s. The results show that the YOLOv8 model has a good detection effect on corns of different shapes and light conditions.

[0072] Analysis of parameters obtained from training corn tassel images:

[0073] The confusion matrix is ​​a graphical summary of the predictions and actual detection results of the classification of image target detection. Compared with the classification accuracy, the confusion matrix and the normalized confusion matrix are more accurate, more efficient, and more convenient and intuitive to discover the key points and features. Figure 7 As shown in the figure, the columns of the matrix represent the predicted categories, the rows represent the actual categories, and the gradient color on the far right corresponds to the color ratio of each cell in the matrix. The higher the color ratio, the darker the color. The values ​​on the diagonal represent the correct prediction ratio, and the values ​​on the off-diagonal represent the wrong prediction ratio.

[0074] F1-Confidence Curve: Reference Figure 8 As shown, the F1 score is the harmonic mean function of precision and recall, between 0 and 1. The closer the number is to "1", the better the training effect of the model is, and the closer the number is to "0", the worse the training effect of the model is. The value of F1 can be changed by adjusting the confidence interval. The calculation formula of F1 is: F1 = 2 × (precision × recall) / (precision + recall). The F1-confidence curve of the corn tassel image training;

[0075] Labels: Reference Fig. 9 The following is a Labels diagram, which represents the category and label information of each detected target. It corresponds to the data in txt obtained after being annotated with LabelMe;

[0076] PR curve: reference Fig.10As shown in the figure, the PR curve describes the relationship between precision and recall. The area formed by the curve and the x-axis and y-axis is the mAP value. The larger the area, the higher the detection accuracy. In the PR curve, the larger the area formed by the precision of the detection target and the x-axis and y-axis, the better the detection effect.

[0077] Results curve: Fig.11 In the Results curve, the most important ones are mAP50, mAP50-95, metrics / precision, and metrics / recall. They represent mAP at 50% IoU threshold, mAP at 50-95% IoU threshold, training accuracy, and recall respectively; in the loss function, the numerical changes of box_loss, cls_loss, dfl_loss, precision, recall, and mAP in multiple training of the model, and the difference between the true value and the predicted value are plotted; after 100 trainings, the data changes of all losses trained in the "Anaconda prompt" command from 0 to 100 are plotted as coordinates.

[0078] After the above analysis, the hyperparameters were optimized to improve the model performance and obtain a male tassel recognition model with high accuracy and efficiency;

[0079] For the depth camera, in one specific case, the Intel RealSense Depth Camera D435 was selected, which has high-speed image capture capabilities to ensure that the tassels can be clearly identified and position information can be obtained even in fast movements;

[0080] Regarding the control mechanism: the control mechanism includes a hydraulic cylinder 2. Compared with the pneumatic lifting mechanism and the screw lifting mechanism, the oil cylinder lift has the advantages of strong carrying capacity, stable lifting and lowering, strong self-locking ability, controllable lifting height and speed, stable power source, etc. The hydraulic cylinder is located below the parallel four-bar structure, and the hydraulic cylinder is configured to drive the parallel four-bar structure to ensure that the angle between the detasseling wheel shaft and the ground remains unchanged while the detasseling wheel moves up and down;

[0081] In a specific case: assemble the components and set the material properties for each component; set the base to a fixed state, add a motion analysis case and configure the gravity direction to the negative direction of the Y axis; in the simulation, introduce a linear motor, set its motion direction and speed to 15 mm / s, and set the simulation time to 13.6 seconds; according to Fig.12The simulation results show that the maximum motor force during operation does not exceed 17000N, and a hydraulic cylinder with a rated thrust of 2 tons is selected; according to the actual working range, the working stroke of the hydraulic cylinder is determined to be 450mm, and the hydraulic cylinder model is HSG40-450; the whole process ensures the effective operation and accurate analysis of the mechanism, and provides reliable data support for subsequent engineering applications; through the above steps, the operating efficiency and stability of the detasseling mechanism can be effectively improved, thereby meeting the actual work needs.

[0082] Based on the above, during the detasseling work, the image of the corn plant during the operation is captured by a depth camera, and the image is transmitted to the industrial computer through a data cable; then, the industrial computer uses a pre-trained tassel recognition model to identify the corn tassels in the image and their three-dimensional coordinates; finally, the industrial computer sends the recognized coordinate information to the profiling mechanism control unit, i.e., the control mechanism, which outputs a control signal according to the difference between the set height and the current height, thereby driving the hydraulic cylinder to adjust the height of the tassel-pulling wheel module; unlike the traditional detasseling mechanism that uses a photoelectric sensor to detect the plant height (whose detection range is small), a deep learning-based tassel recognition model is used to establish a special corn tassel data set, and three labels including unobstructed, semi-obstructed and fully obstructed are set; this not only realizes large-scale data collection, but also can effectively identify and measure the tassels that have not yet fully grown, significantly improving the quality and efficiency of mechanical detasseling of seed corn, reducing production costs, and playing an important role in promoting the development of the corn seed production industry.

[0083] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. For ordinary technicians in this technical field, after knowing the contents recorded in the present invention, they can make several equivalent changes and substitutions without departing from the principle of the present invention. These equivalent changes and substitutions should also be regarded as belonging to the protection scope of the present invention.

Claims

1. A corn tassel intelligent removal system, comprising a tassel extraction mechanism for corn tassel removal, the tassel extraction mechanism comprising a bracket, and a plurality of tassel removal wheel groups mounted on the bracket, each of the tassel removal wheel groups comprising two tassel removal wheels, each of the tassel removal wheels being provided with an independent drive assembly, the drive assembly controlling the relative rotation of the two tassel removal wheels, characterized in that: The system also includes: The tassel recognition module includes a tassel recognition model and a depth camera. The tassel recognition model executes the following instructions based on the image captured by the depth camera: Construct a target box with tassels in the image; Extract the spatial coordinates and orientation information of the top of the tassel in the target frame to form the tassel position data; Output the height displacement of the tassel and feed back the height displacement data to the control mechanism; The control mechanism adjusts the height of the male wheel in real time based on the feedback data.

2. A corn tassel intelligent removal system according to claim 1, characterized in that: The detasseling wheel is an inflatable member and the drive assembly is a motor drive assembly.

3. The intelligent corn tassel removal system according to claim 1, characterized in that: The bracket is a parallelogram structure, and the control mechanism includes a hydraulic cylinder, which is located below the parallelogram structure and is configured to drive the parallelogram structure to move the detasseling wheel up and down while ensuring that the angle between the detasseling wheel shaft and the ground remains unchanged.

4. A corn tassel intelligent removal system according to claim 3, characterized in that: A spiral wheel is coaxially mounted on one end of each detasseling wheel facing the traveling direction, and the two spiral wheels cooperate to drive the male ears to be fed into the corresponding two detasseling wheels one by one and at intervals.

5. The corn tassel intelligent removal system according to claim 1, characterized in that: An independent bearing support structure is arranged on the bracket at each detasseling wheel, and the bearing support structure is used to assemble and support the wheel axle. One end of the wheel axle is connected to the detasseling wheel, and the other end is connected to the driving assembly. A V-shaped plate is installed between the two bearing support structures corresponding to each group of detasseling wheels.

6. The corn tassel intelligent removal system according to claim 1, characterized in that: The tassel recognition model is trained using an established data set and evaluated using a cross-validation method, and the parameters of the tassel recognition model are dynamically optimized based on the evaluation results.

7. The intelligent corn tassel removal system according to claim 6, characterized in that: The dataset includes corn images at different growth stages and under various environmental conditions in the agricultural image database, corn images taken on-site in corn fields, and tassel images in corn images annotated using LabelMe.

8. The intelligent corn tassel removal system according to claim 7, characterized in that: The annotation includes annotating the tassel boundary box and setting tassel labels including unoccluded, semi-occluded and fully-occluded.

9. The intelligent corn tassel removal system according to claim 7, characterized in that: In the process of evaluating the male tassel recognition model, the predicted frame coordinate loss is established, and the loss is calculated by the overlapping area, center point distance and aspect ratio of the predicted frame and the true frame, specifically: Among them, B gt represents the real box, B pred represents the prediction box, ρ 2( B gt ,B pred ) represents the Euclidean distance between the center points of the predicted box and the true box, C represents the diagonal length of the minimum enclosing rectangle that can contain the predicted box and the true box, α is the balance parameter, and v is the aspect ratio metric function used to measure the consistency of the aspect ratio.

Images