Crop picking shake prediction method, readable storage medium and equipment
By combining the shaking trajectory prediction model of the physical prediction model and the Informer model, the problem of the existing technology that predicts shaking trajectory prone to deviations during crop picking is solved, and a fast and accurate picking effect is achieved.
Patent Information
- Application Number
- CN202411255544.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-08-30
- Filing Date
- 2024-09-09
- Publication Date
- 2025-04-04
AI Technical Summary
The prior art is prone to large deviations when predicting the shaking trajectory during crop picking for a long time, making it difficult to accurately capture complex nonlinear relationships.
Combining the shaking trajectory prediction model of the physical prediction model and the Informer model, by obtaining the shaking data of the crop and the direction of the picking force, the trained model is input to determine the first predicted trajectory and the second predicted trajectory, and sum it to obtain the predicted shaking trajectory.
It realizes the rapid and accurate picking of the shaking fruits of crops, solves the deviation problem of the existing technology in long-term prediction, and improves the picking accuracy.
Smart Images
Figure CN119760622A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a crop picking shaking prediction method, a readable storage medium and a device, belonging to the technical field of agricultural picking. Background Art
[0002] Currently, harvesting crop fruits by robots has become the norm. During the picking process, the fruits are often in an oscillating state, so accurately and quickly predicting the coordinates of the fruits becomes the key to improving picking efficiency.
[0003] The traditional simplified second-order system model ignores the structure of crop fruits and the limitation of parameter accuracy, so long-term predictions are prone to large deviations or even misalignment. The ARIMA model in the stock market, finance and other fields is built based on linear assumptions, which makes it difficult to capture complex nonlinear relationships and is difficult to apply to crop vibration prediction. The trajectory prediction data in the field of unmanned driving is relatively flat and is not suitable for vibration prediction needs with fast speed changes. Although deep learning methods are widely used in multivariate time series prediction due to their automatic feature extraction and nonlinear modeling capabilities, the use of pure deep learning methods (such as RNN and LSTM) is prone to overfitting. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a crop picking shaking prediction method, a readable storage medium and a device. By combining the physical prediction model and the shaking trajectory prediction model of the Informer model, the shaking trajectory of crops with obvious second-order system characteristics during picking is predicted, and the shaking fruits can be picked quickly and accurately, which solves the problem that the current machine picking prediction model is prone to large deviations after long-term prediction work.
[0005] In order to achieve the above purpose / solve the above technical problems, the present invention is implemented by adopting the following technical solutions: The present invention provides a crop picking shaking prediction method on the one hand, comprising: Acquire shaking data of the target crop when it is shaking and the corresponding picking force direction; wherein the shaking data includes horizontal direction data and vertical direction data; Inputting the shaking data and the picking force direction into a trained shaking trajectory prediction model to determine a first predicted trajectory and a second predicted trajectory; Determining a predicted shaking trajectory according to the first predicted trajectory and the second predicted trajectory; Wherein, the sway trajectory prediction model includes a physical prediction model and an Informer model; The input of the physical prediction model is the horizontal direction data of the target crop and the corresponding picking force direction, and the output is the prediction residual and the first prediction trajectory; The input of the Informer model is the vertical direction data of the target crop, the corresponding picking force direction and the prediction residual, and the output is the second prediction trajectory.
[0006] Further, determining the predicted shaking trajectory according to the first predicted trajectory and the second predicted trajectory includes: The first predicted trajectory and the second predicted trajectory are summed to obtain a predicted shaking trajectory.
[0007] Furthermore, the training of the sway trajectory prediction model includes: Acquire historical shaking data of harvested crops and corresponding historical picking force directions; wherein the historical shaking data includes historical horizontal direction data and historical vertical direction data; Inputting historical horizontal direction data and corresponding historical picking force directions into a target physical prediction model to determine optimal system parameters of the target physical prediction model; Based on the optimal system parameters, determine the optimal physical prediction model; Based on the optimal physical prediction model, obtain the prediction residual; The historical vertical direction data, the corresponding historical picking force direction and the prediction residual are used to train the target Informer model to obtain the trained Informer model.
[0008] Furthermore, the obtaining of historical shaking data of harvested crops and corresponding historical harvesting force directions includes: A Cartesian coordinate system is established with the center of mass of the part of the crop to be picked as the origin, in which the shaking data is expressed as rectangular coordinates X , Y , Z The time series of the axis position coordinates, defined as Z The axis is vertical; The picking force direction is represented by the force direction of the end of the picking machine on the part of the crop to be picked and the X The angle between the axis and the
[0009] Furthermore, the physical prediction model includes: X Axis Direction: ; Y Axis Direction: ; in: X ( t ) indicates that the horizontal data is X The time series in the axis direction, Y ( t ) indicates that the horizontal data isY The time series in the axis direction, t Indicates time, Indicates the picking force direction, represents the damping ratio of the second-order system, is the undamped natural angular frequency, A represents the amplitude, represents the standard deviation, represents the mean, represents the proportionality coefficient, represents the proportionality coefficient, b represents the bias, e Represents a natural constant.
[0010] Furthermore, the inputting of the historical horizontal direction data and the corresponding historical picking force direction into the target physical prediction model to determine the optimal system parameters of the target physical prediction model includes: Select multiple time points corresponding to historical horizontal data; Obtaining actual observation values corresponding to a plurality of the time points; Calculating the predicted output values corresponding to the target physical prediction model at the plurality of time points; Calculate the error value corresponding to the actual observation value and the predicted output value; Construct an objective function based on the error value; With the goal of minimizing the objective function, the system parameters of the target physical prediction model are optimized to determine the optimal system parameters.
[0011] Furthermore, obtaining the prediction residual based on the optimal physical prediction model includes: Inputting historical horizontal direction data and corresponding historical picking force directions into the optimal physical prediction model to calculate a first prediction trajectory; The difference between the first predicted trajectory and the historical horizontal data is calculated to obtain the prediction residual.
[0012] Furthermore, the historical vertical direction data, the corresponding historical picking force direction and the prediction residual are used to train the target Informer model to obtain the trained Informer model, including: The historical vertical direction data, the corresponding historical picking force directions and the prediction residuals are divided into a training set and a label set in time sequence; Input the training set into the target Informer model to generate a prediction sequence; Calculate the loss value of the predicted sequence and the label set and back-propagate to update the model parameters of the target Informer model; Use the optimization algorithm to adjust the parameters of the target Informer model until the loss value meets the preset value; Get the trained Informer model.
[0013] A second aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned crop picking shake prediction method is implemented.
[0014] A third aspect of the present invention provides a computer device, comprising: A memory for storing instructions; The processor is used to execute the instructions so that the device implements the crop picking shaking prediction method as described above.
[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention inputs the sway data of the target crops and the corresponding picking force direction into a sway trajectory prediction model that combines a physical prediction model and an Informer model, predicts the sway trajectory of crops with obvious second-order system characteristics during picking, and can quickly and accurately pick the swaying fruits, solving the problem that the current machine picking prediction model is prone to large deviations after long-term prediction work.
[0016] 2. The present invention inputs the historical shaking data of picked crops and the corresponding historical picking force direction into the target physical prediction model, selects multiple time points corresponding to the historical horizontal direction data, obtains the actual observation values corresponding to the multiple time points and the predicted output values of the target physical prediction model corresponding to the multiple time points, takes the error corresponding to the predicted output value and the actual observation value as the objective function, and determines the optimal parameters of the target physical prediction model with minimizing the objective function as the optimization goal, thereby obtaining the optimal physical prediction model, which can capture the nonlinearity and complexity of crop shaking.
[0017] 3. The present invention utilizes the prediction residual output by the optimal physical prediction model and uses historical vertical direction data and corresponding historical picking force direction to train the target Informer model, so that the trained Informer model can make full use of the swaying data of crops in multiple directions, thereby improving the accuracy of the swaying trajectory prediction and greatly improving the picking accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 It is a flowchart of a method for predicting crop picking shaking according to an embodiment of a method, a readable storage medium and a device of the present invention; Figure 2 It is a training flow chart of a shaking trajectory prediction model in a crop picking shaking prediction method according to an embodiment of the present invention; Figure 3It is a schematic diagram of the prediction effect of an optimal physical prediction model of a crop picking shaking prediction method of the present invention; Figure 4 It is a schematic diagram of the prediction residual of the optimal physical prediction model of a crop picking shaking prediction method of the present invention; Figure 5 It is a schematic diagram of the prediction effect of a crop picking shaking prediction method of the present invention; Figure 6 It is a comparison diagram of the prediction effects of a shaking trajectory prediction model of a crop picking shaking prediction method of the present invention, a physical prediction model, and an Informer model. DETAILED DESCRIPTION
[0019] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention. Example
[0020] like Figure 1 As shown, a crop picking shaking prediction method includes: Obtain shaking data of target crops and corresponding picking force direction when shaking; Inputting the shaking data and the picking force direction into the trained shaking trajectory prediction model to determine a first predicted trajectory and a second predicted trajectory; The first predicted trajectory and the second predicted trajectory are summed to obtain a predicted shaking trajectory; The sway trajectory prediction model includes a physical prediction model and an informer model, and the sway data includes horizontal direction data and vertical direction data; The input of the physical prediction model is the horizontal direction data of the target crop and the corresponding picking force direction, and the output is the prediction residual and the first prediction trajectory; The input of the Informer model is the vertical direction data of the target crop, the corresponding picking force direction and the prediction residual, and the output is the second prediction trajectory.
[0021] like Figure 2 As shown, the training of the sway trajectory prediction model includes: The historical shaking data of the harvested crops and the corresponding historical picking force direction are obtained. Specifically, the historical shaking data includes historical horizontal direction data and historical vertical direction data. The center of mass of the part of the crop to be harvested is taken as the origin to establish a Cartesian rectangular coordinate system, where: The sway data is expressed as rectangular coordinates X , Y , Z The time series of the axis position coordinates, defined as Z The axis is vertical; The picking force direction is the direction of the force applied by the end of the picking machine to the part of the crop to be picked. X The angle between the axes; Physics prediction models include: X Axis Direction: ; Y Axis Direction: ; X ( t ) indicates that the horizontal data is X The time series in the axis direction, Y ( t ) indicates that the horizontal data is Y The time series in the axis direction, t Indicates time, Indicates the picking force direction, represents the damping ratio of the second-order system, is the undamped natural angular frequency, A represents the amplitude, represents the standard deviation, represents the mean, represents the proportionality coefficient, represents the proportionality coefficient, b represents the bias, e Represents a natural constant.
[0024] The historical horizontal direction data is low-pass filtered to eliminate high-frequency noise interference, and the historical horizontal direction data and the corresponding historical picking force direction are input into the target physical prediction model to determine the optimal system parameters of the target physical prediction model. Specifically: Select multiple time points corresponding to historical horizontal data; Get the actual observation values corresponding to multiple time points; Calculate the predicted output values corresponding to the target physical prediction model at multiple time points; Calculate the error value corresponding to the actual observation value and the predicted output value; Construct an objective function based on the error value, including:
[0025] represents the objective function constructed based on the error value, N Indicates the number of selected time points, Indicates i A point in time, represents the actual observation value corresponding to the i-th time point, Represents the error value corresponding to the actual observation value and the predicted output value; Indicates that the target physical prediction model is i The predicted output value at each time point, A vector representing the target physics prediction model parameters. The target physics prediction model parameters include , , A , , , , , b。
[0026] With the goal of minimizing the objective function, optimize the system parameters of the target physical prediction model and determine the optimal system parameters; Based on the optimal system parameters, the optimal physical prediction model is determined.
[0027] Based on the optimal physical prediction model, the prediction residual is obtained, specifically: The historical horizontal direction data and the corresponding historical picking force direction are input into the optimal physical prediction model, the first prediction trajectory is calculated, and the difference between the first prediction trajectory and the historical horizontal direction data is calculated to obtain the prediction residual. X Take the first predicted trajectory in the axis direction as an example: The historical horizontal data of the previous two cycles of crop shaking ; M Indicates the time point when the first two cycles of crop shaking end. Indicates i Historical horizontal data corresponding to each time point; The historical horizontal data of the previous two periods , compared with the historical horizontal data of the previous two periods Corresponding picking force direction Input the optimal physical prediction model and get X First predicted trajectory in the axis direction ; Representation and M Corresponding X The number of horizontal data of the first predicted trajectory in the axis direction; Representation and The corresponding first predicted trajectory; Calculate the difference between the first predicted trajectory and the historical horizontal data: ; express X The first predicted trajectory in the axial direction X The prediction residuals of the historical horizontal data in the axis direction.
[0028] The historical vertical direction data, the corresponding historical picking force direction and the prediction residual are used to train the target Informer model to obtain the trained Informer model. Specifically, the historical vertical direction data, the corresponding historical picking force direction and the prediction residual are divided into a training set and a label set in time series. The historical vertical direction data, the corresponding historical picking force direction and the prediction residual of the first two cycles of crop shaking are taken as the training set:
[0029] The remaining historical vertical direction data, the corresponding historical picking force direction, and the prediction residual are used as the label set: ,
[0030] M represents the time point at which the first two cycles of crop shaking end, and D represents the time point at which crop shaking ends; Indicates the first two cycles X The prediction residuals of the historical horizontal data in the axial direction, Indicates the first two cycles X Axis direction M The difference between the horizontal direction data of the first predicted trajectory at a time point and the historical horizontal direction data; Indicates the first two cycles Y The prediction residuals of the historical horizontal data in the axial direction, Indicates the first two cycles Y Axis direction M The difference between the horizontal direction data of the first predicted trajectory at a time point and the historical horizontal direction data; Represents the historical vertical data of the previous two periods, Indicates M Historical vertical data at each time point; Indicates the historical picking force direction corresponding to the first two cycles, Indicates M The historical picking force direction at each time point; Indicates the remaining X The prediction residuals of the historical horizontal data in the axial direction, Indicates the remaining X Axis direction D The difference between the horizontal direction data of the first predicted trajectory at a time point and the historical horizontal direction data; Indicates the remaining Y The prediction residuals of the historical horizontal data in the axial direction, Indicates the remaining Y Axis directionD The difference between the horizontal direction data of the first predicted trajectory at a time point and the historical horizontal direction data; Represents the remaining historical vertical data, Indicates D Historical vertical data at each time point; represents the remaining corresponding historical picking force direction, Indicates D The historical picking force direction at a time point, .
[0031] Input the training set into the target Informer model to generate a prediction sequence , , ; Calculate the loss value of the predicted sequence and the label set and back-propagate to update the model parameters of the target Informer model, including:
[0032]
[0033] Represents the loss value, and the loss function is the mean square error function; represents the L2 norm.
[0034] Calculate the gradient of the loss function with respect to the model parameters of the target Informer model and update the model parameters; Use the gradient descent optimization algorithm to adjust the parameters of the target Informer model until the loss value meets the preset value; Get the trained Informer model.
[0035] like Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, the shaking trajectory prediction model adopted by the present invention can accurately predict the shaking trajectory of crops during picking, and compared with the existing models, the shaking trajectory prediction model adopted by the present invention has better prediction effect. Example
[0036] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the following method: Acquire shaking data of the target crop when it is shaking and the corresponding picking force direction; wherein the shaking data includes horizontal direction data and vertical direction data; Inputting the shaking data and the picking force direction into a trained shaking trajectory prediction model to determine a first predicted trajectory and a second predicted trajectory; Determining a predicted sway trajectory according to the first predicted trajectory and the second predicted trajectory; Wherein, the sway trajectory prediction model includes a physical prediction model and an Informer model; The input of the physical prediction model is the horizontal direction data of the target crop and the corresponding picking force direction, and the output is the prediction residual and the first prediction trajectory; The input of the Informer model is the vertical direction data of the target crop, the corresponding picking force direction and the prediction residual, and the output is the second prediction trajectory. Example
[0037] A computer device comprising: A memory for storing instructions; The processor is used to execute the instructions so that the device performs the steps of implementing the following method: Acquire shaking data of the target crop when it is shaking and the corresponding picking force direction; wherein the shaking data includes horizontal direction data and vertical direction data; Inputting the shaking data and the picking force direction into a trained shaking trajectory prediction model to determine a first predicted trajectory and a second predicted trajectory; Determining a predicted sway trajectory according to the first predicted trajectory and the second predicted trajectory; Wherein, the sway trajectory prediction model includes a physical prediction model and an Informer model; The input of the physical prediction model is the horizontal direction data of the target crop and the corresponding picking force direction, and the output is the prediction residual and the first prediction trajectory; The input of the Informer model is the vertical direction data of the target crop, the corresponding picking force direction and the prediction residual, and the output is the second prediction trajectory.
[0038] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0039] The present application is described with reference to the flowcharts of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0040] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.
[0041] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.
[0042] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, which all fall within the protection of the present invention.
Claims
1. A method for predicting crop picking vibration, characterized in that: include: Obtain shaking data of target crops and corresponding picking force direction when shaking; The shaking data includes horizontal direction data and vertical direction data; Inputting the shaking data and the picking force direction into a trained shaking trajectory prediction model to determine a first predicted trajectory and a second predicted trajectory; Determining a predicted sway trajectory according to the first predicted trajectory and the second predicted trajectory; Wherein, the sway trajectory prediction model includes a physical prediction model and an Informer model; The input of the physical prediction model is the horizontal direction data of the target crop and the corresponding picking force direction, and the output is the prediction residual and the first prediction trajectory; The input of the Informer model is the vertical direction data of the target crop, the corresponding picking force direction and the prediction residual, and the output is the second prediction trajectory.
2. The crop picking shaking prediction method according to claim 1, characterized in that: The step of determining the predicted shaking trajectory according to the first predicted trajectory and the second predicted trajectory includes: The first predicted trajectory and the second predicted trajectory are summed to obtain a predicted shaking trajectory.
3. The crop picking shaking prediction method according to claim 1, characterized in that: The training of the sway trajectory prediction model includes: Acquire historical shaking data of harvested crops and corresponding historical picking force directions; wherein the historical shaking data includes historical horizontal direction data and historical vertical direction data; Inputting historical horizontal direction data and corresponding historical picking force directions into a target physical prediction model to determine optimal system parameters of the target physical prediction model; Based on the optimal system parameters, determine the optimal physical prediction model; Based on the optimal physical prediction model, determine the prediction residual; The historical vertical direction data, the corresponding historical picking force direction and the prediction residual are used to train the target Informer model to obtain the trained Informer model.
4. The crop picking shaking prediction method according to claim 3 is characterized in that: The obtaining of historical shaking data of harvested crops and corresponding historical harvesting force directions includes: A Cartesian coordinate system is established with the center of mass of the part of the crop to be picked as the origin, in which the shaking data is expressed as rectangular coordinates X , Y , Z The time series of the axis position coordinates, defined as Z The axis is vertical; The picking force direction is represented by the force direction of the end of the picking machine on the part of the crop to be picked and the X The angle between the axis and the 5. The crop picking shaking prediction method according to claim 4, characterized in that: The physical prediction model includes: X Axis Direction: ; Y Axis Direction: ;in: X ( t ) indicates that the horizontal data is X The time series in the axis direction, Y ( t ) indicates that the horizontal data is Y The time series in the axis direction, t Indicates time, Indicates the picking force direction, represents the damping ratio of the second-order system, is the undamped natural angular frequency, A represents the amplitude, represents the standard deviation, represents the mean, represents the proportionality coefficient, represents the proportionality coefficient, b Indicates bias, e Represents a natural constant.
6. The crop picking shaking prediction method according to claim 3, characterized in that: The step of inputting historical horizontal direction data and corresponding historical picking force directions into the target physical prediction model to determine the optimal system parameters of the target physical prediction model includes: Select multiple time points corresponding to historical horizontal data; Obtaining actual observation values corresponding to a plurality of the time points; Calculating the predicted output values corresponding to the target physical prediction model at the plurality of time points; Calculate the error value corresponding to the actual observation value and the predicted output value; Construct an objective function based on the error value; With the goal of minimizing the objective function, the system parameters of the target physical prediction model are optimized to determine the optimal system parameters.
7. The crop picking shaking prediction method according to claim 3, characterized in that: The step of determining the prediction residual based on the optimal physical prediction model includes: Inputting historical horizontal direction data and corresponding historical picking force directions into the optimal physical prediction model to calculate a first prediction trajectory; The difference between the first predicted trajectory and the historical horizontal data is calculated to obtain the prediction residual.
8. The crop picking shaking prediction method according to claim 3, characterized in that: The method of using historical vertical direction data, corresponding historical picking force directions and prediction residuals to train a target informer model to obtain a trained informer model includes: The historical vertical direction data, the corresponding historical picking force directions and the prediction residuals are divided into a training set and a label set in time sequence; Input the training set into the target Informer model to generate a prediction sequence; Calculate the loss value of the predicted sequence and the label set and back-propagate to update the model parameters of the target Informer model; Use the optimization algorithm to adjust the parameters of the target Informer model until the loss value meets the preset value; Get the trained Informer model.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the crop picking shake prediction method as described in any one of claims 1-8 is implemented.
10. A computer device, characterized in that: include: A memory for storing instructions; A processor is used to execute the instructions so that the device implements the crop picking shake prediction method as described in any one of claims 1-8.