Precise spraying control method and device based on combination of laser scanning and neural network

Abstract

The application relates to the technical field of spray control, and discloses a precise spray control method based on the combination of laser scanning and a neural network, which comprises the following steps: scanning an environment through a laser radar, dynamically adjusting the scanning line bundle density based on the real-time speed of a device, and generating sparse sampling point cloud data; performing compressed sensing reconstruction on the sparse sampling point cloud data, inputting the reconstructed three-dimensional point cloud into a neural network model for target positioning, and outputting the three-dimensional coordinates of the target; and calculating the spray error of the current position of the device according to the three-dimensional coordinates; and the application also discloses a precise spray control device based on the combination of laser scanning and a neural network, which comprises a laser scanning module, a data processing module, a control calculation module, an actuating mechanism module and a feedback closed loop module. Through laser radar linkage regulation and control, neural network dynamic distribution and fractional order control, the application realizes the improvement of spray precision, the acceleration of response and the optimization of resources.

Description

Technical Field

[0001] This invention relates to the field of spray control technology, specifically to a precise spray control method and device based on the combination of laser scanning and neural networks. Background Technology

[0002] Modern firefighting operations place increasingly stringent demands on sprinkler precision and resource efficiency; traditional sprinkler equipment, lacking dynamic environmental perception and adaptive control capabilities, struggles to cope with complex terrain distributions.

[0003] In existing technologies, fixed-beam-density lidar scanning schemes can stably acquire target coordinates during low-speed operations, with simple hardware structures and easy maintenance; traditional PID controllers, due to their linear operation characteristics, exhibit good control stability under constant speed conditions. Independent modular design allows the positioning, control, and execution systems to be debugged separately, reducing initial deployment difficulty.

[0004] Existing technologies exhibit significant limitations in dynamic usage scenarios: fixed scanning modes generate data redundancy or insufficient sampling when the equipment changes speed, resulting in processing delay fluctuations of up to 200%; the integer-order model of traditional PID cannot adapt to the nonlinear dynamic characteristics of spray error, and the control deviation increases by 3 times under rapid speed change conditions; timing mismatch between independent modules causes action lag, and the spray landing point deviation exceeds 10cm. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a precise spray control method and device based on the combination of laser scanning and neural networks, which solves the problems of inaccurate sensing, delayed control response, and large spraying errors in existing spray equipment in dynamic operating environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a precise spray control method based on a combination of laser scanning and neural networks, comprising the following steps:

[0007] The environment is scanned by lidar, and the scanning beam density is dynamically adjusted based on the real-time speed of the device to generate sparse sampling point cloud data.

[0008] Compressed sensing reconstruction is performed on the sparse sampled point cloud data, and the reconstructed 3D point cloud is input into a neural network model for target localization, outputting the 3D coordinates of the target;

[0009] The spraying error at the current position of the device is calculated based on the three-dimensional coordinates. Based on the time-domain rate of change of the spraying error, combined with the acceleration data collected by the inertial measurement unit and the historical motion trajectory, the future acceleration is predicted through a stochastic process prediction model.

[0010] The predicted future acceleration and spray error are input into a fractional-order PID controller to generate nozzle control commands.

[0011] Based on the absolute value of the real-time spraying error, a non-smooth correction term is superimposed on the control command to output a control signal with stability correction.

[0012] The high-frequency piezoelectric nozzle is driven to perform spraying action according to the modified control signal, and the spray coverage is fed back to the lidar to dynamically adjust the scanning frequency and sampling rate.

[0013] Preferably, the dynamic adjustment of the scan beam density is as follows:

[0014] When the device moves faster than a preset first speed threshold, the laser radar scanning beam density is reduced to a first beam density; when the device moves slower than the first speed threshold, the scanning beam density is increased to a second beam density, and the beam density adjustment is synchronized with the sampling rate of compressed sensing reconstruction.

[0015] Preferably, the compressed sensing reconstruction includes:

[0016] An iterative shrinkage threshold algorithm is used to reconstruct sparse sampled point cloud data, wherein the measurement matrix dynamically adjusts the sparsity constraint weights according to the device's moving speed;

[0017] The neural network model is a lightweight PointNet++ architecture. Its input layer receives the reconstructed 3D point cloud coordinates, and its output layer extracts the centroid position of the target through max pooling operation.

[0018] The compressed sensing reconstruction and neural network model inference share computing resources. When the point cloud density is lower than a preset threshold, computing power is preferentially allocated to the reconstruction algorithm.

[0019] Preferably, the stochastic process prediction model includes:

[0020] The Iton process model based on device acceleration has a drift term that is a linear decay function of acceleration and a diffusion term that is a random perturbation term proportional to the square root of acceleration.

[0021] By calculating the Malliavin derivative to quantify the path space sensitivity, and combining the acceleration change trend in historical motion trajectories, the expected acceleration value in the future time period is predicted.

[0022] The prediction model is coupled to the input channel of the fractional-order PID controller. When the predicted acceleration changes abruptly beyond a preset threshold, an emergency correction mode of the control command is automatically triggered.

[0023] Preferably, the fractional-order PID controller includes:

[0024] Differential operators are defined using the first-order definition, while integral operators are defined using the second-order definition.

[0025] The control parameters are tuned using a frequency distribution algorithm, with phase margin and gain margin as optimization objectives;

[0026] The controller's output command is linked to the equipment's kinematic model. When the spraying error direction is opposite to the equipment's movement direction, the differential term weight is dynamically increased to suppress overshoot.

[0027] Preferably, the superposition logic of the non-smoothness correction term is as follows:

[0028] When the absolute value of the real-time spraying error exceeds the preset first error threshold, a non-smooth compensation term in the same direction as the error sign is superimposed on the control command, and the compensation amount is the product of the preset gain coefficient and the absolute value of the error.

[0029] The activation state of the non-smooth correction term is related to the device acceleration;

[0030] The convergence of the corrected control signal is verified by the Lyapunov function to ensure that the error decays to below the second error threshold within a preset time threshold.

[0031] Preferably, the driving parameters of the high-frequency piezoelectric nozzle include:

[0032] The pulse width is fixed and has a linear mapping relationship with the preset time length and the driving voltage;

[0033] The nozzle triggering timing is delayed and compensated based on the predicted acceleration, and the compensation time is calculated based on the predicted acceleration and the square of the control period;

[0034] The array of piezoelectric nozzles is orthogonal to the laser scanning direction, and the driving signal of each nozzle is independently adjustable to match the local density distribution of the target.

[0035] Preferably, the spray coverage feedback includes:

[0036] The sprayed area is segmented using the HSV color space by acquiring environmental images after spraying using a multispectral camera, and the coverage percentage is calculated.

[0037] When the coverage rate is lower than the preset coverage rate threshold, the LiDAR scanning frequency is increased to the first frequency, and the sampling rate of compressed sensing is increased to the first sampling rate.

[0038] The feedback mechanism is linked to the training data of the neural network model, adding the coordinates of the uncovered areas to the next round of training dataset to optimize positioning accuracy.

[0039] Preferably, the sparse sampled point cloud data includes:

[0040] The density range is a preset density range, the signal-to-noise ratio is greater than a preset signal-to-noise ratio threshold, and the point cloud distribution follows a Laplace distribution in the direction of device motion.

[0041] The sparsity of the point cloud data is dynamically correlated with the device speed. When the speed changes, the sparsity is dynamically adjusted according to the rate of change of the device speed.

[0042] The point cloud attributes serve as prior constraints for compressed sensing reconstruction and are used to calculate the regularization term coefficients in the iterative shrinkage threshold algorithm.

[0043] The present invention also provides a precision spray control device based on a combination of laser scanning and neural networks, comprising:

[0044] The laser scanning module includes a lidar with adjustable beam density and an inertial measurement unit, used to acquire three-dimensional point clouds of the environment and equipment motion data in real time.

[0045] The data processing module is connected to the laser scanning module via a high-speed data interface and includes an FPGA accelerator and an embedded processor. The FPGA is configured to execute a compressed sensing reconstruction algorithm, and the embedded processor is configured to run a lightweight PointNet++ neural network model.

[0046] The control calculation module is connected to the data processing module via the PCIe bus. It includes a fractional-order PID controller and a non-smooth correction unit. The fractional-order PID controller receives the coordinate and device position error and the predicted acceleration to generate an initial control command. The non-smooth correction unit adds a stability compensation term according to the error threshold.

[0047] The actuator module is connected to the control calculation module via a PWM interface and includes a piezoelectric nozzle array and a high-voltage drive circuit. The triggering delay of the piezoelectric nozzle is less than 15 microseconds.

[0048] The feedback closed-loop module, wherein the actuator module is configured to transmit the spray coverage data back to the data processing module via the wireless communication module, triggering dynamic adjustment of the lidar scanning frequency and compressed sensing sampling rate to form closed-loop control.

[0049] This invention provides a precise spray control method and device based on a combination of laser scanning and neural networks. It possesses the following features:

[0050] Beneficial effects:

[0051] 1. This invention employs a technical solution that links the lidar beam density with the equipment speed, compressive sensing reconstruction, and dynamic allocation of neural network computing power, achieving an optimal balance between point cloud processing efficiency and positioning accuracy. Compared to existing technologies where fixed scanning parameters lead to wasted computing power or insufficient accuracy, this invention solves the problem of balancing efficiency and accuracy.

[0052] 2. A technical solution that combines acceleration feedforward generation using a stochastic process prediction model with nonlinear dynamic characteristics fused by a fractional-order PID controller achieves advance compensation for spraying errors and suppression of dynamic delays. Compared with traditional integer-order PID control and static feedforward compensation methods, this solution overcomes the lag in response to rapid speed changes in equipment.

[0053] 3. This invention employs a technical solution that combines error amplitude-triggered non-smooth correction term superposition with Lyapunov stability verification, ensuring control stability under extreme conditions. Compared to linear correction or fixed threshold protection strategies, this resolves the contradiction between sluggish response and excessive conservatism.

[0054] 4. A technical solution based on dynamic adjustment of lidar scanning frequency and compressed sensing sampling rate according to spray coverage feedback forms a closed-loop optimization of resource consumption and detection accuracy. Compared with existing open-loop control and fixed parameter scanning schemes, this overcomes the technical bottleneck of their inability to adapt to operational results. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0056] Figure 2 This is a schematic diagram of the device construction of the present invention. Detailed Implementation

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see the appendix Figure 1 This invention provides a precise spray control method based on a combination of laser scanning and neural networks, comprising the following steps:

[0059] S1. The environment is scanned by lidar, and the scanning beam density is dynamically adjusted based on the real-time speed of the device to generate sparse sampling point cloud data.

[0060] The lidar scanning module dynamically adjusts the scanning beam density based on the real-time moving speed of the spray equipment, generating sparsely sampled point cloud data. This process is achieved through a speed sensing mechanism and hardware-coordinated control, resolving the conflict between point cloud data redundancy and processing latency under high-speed movement, while ensuring target positioning accuracy. The dynamic adjustment strategy, along with subsequent compressed sensing reconstruction and neural network positioning modules, forms a closed-loop data flow, providing highly timely input for spray control.

[0061] In some embodiments, the lidar uses an OusterOS1-64 line scanning device with a vertical angular resolution of 0.33° and a horizontal scanning frequency of 10Hz. The inertial measurement unit (IMU) is an XsensMTi-670, which acquires real-time triaxial acceleration and angular velocity data. The lidar and IMU are rigidly connected via a support frame, with a coordinate system alignment error of less than 2mm.

[0062] Specifically, the scanning beam density adjustment of the LiDAR is achieved through a MOSFET array in the control circuit. The MOSFET drive signal is dynamically generated by the main control board based on the speed data, with a response delay of less than 1ms. The beam shutdown logic follows a "top-down" principle, prioritizing the shutdown of the upper beam transmitters while reserving the bottom beams for low-profile identification.

[0063] In one possible implementation, the harness density adjustment rule is defined as:

[0064]

[0065] Where, N lines To activate the number of harnesses, v represents the real-time moving speed of the device, measured by the chassis encoder and smoothed by a Kalman filter.

[0066] Specifically, when the speed exceeds the threshold, the LiDAR scanning mode switches to a low-density state, shutting down the top 48 lines out of the 48 beams, leaving only the bottom 16 lines operational. At this point, the vertical resolution of the point cloud decreases to 0.99°, but the horizontal scanning frequency is increased to 20Hz to compensate for the loss of spatial information.

[0067] As an alternative, the sparsity of sparsely sampled point cloud data is dynamically correlated with device speed. The point cloud density ρ satisfies:

[0068]

[0069] Where ρ0 = 200 points / m³ is the baseline density, v is the real-time moving speed of the equipment, ρ(v) is the dynamic adjustment of the wire harness density when the equipment moves at speed v, and ΔN = 0.15ρ0, v max =5m / s is the maximum design speed of the equipment.

[0070] In some embodiments, sparsely sampled point cloud data satisfy the following characteristics:

[0071] Point cloud distribution: follows a Laplace distribution in the direction of device motion, with the probability density function as follows:

[0072]

[0073] Where p(x,y,z) is the probability density function of the point cloud coordinates in three-dimensional space, describing the positional distribution characteristics of the device's motion direction (x-axis), x-μ x The absolute deviation of the current x-coordinate from the mean, quantized as the position offset, μ. x The coordinates of the equipment's direction of motion are the average values, and b = 0.2m is the scale parameter.

[0074] Signal-to-noise ratio: The original point cloud signal-to-noise ratio (SNR) is greater than 20dB, and ambient light noise is suppressed by hardware filtering circuit;

[0075] Data compression: Run-length encoding (RLE) is used to compress point cloud storage space, achieving a compression ratio of 6:1.

[0076] In one possible implementation, the dynamic harness density adjustment is linked to the measurement matrix parameters of the compressed sensing reconstruction module. When the harness density decreases, the sparsity constraint weight λ of the compressed sensing measurement matrix ΦΦ is adjusted synchronously: λ(v)=λ0+k·v;

[0077] Wherein, λ(v) is the speed-related comprehensive evaluation coefficient, which quantifies the dynamic balance between equipment moving speed and operational accuracy; λ0 is the static benchmark evaluation coefficient, which characterizes the basic accuracy weight when the equipment is stationary or at low speed; k is the speed influence gain coefficient, which describes the linear contribution intensity of speed to the comprehensive evaluation value; and v is the real-time moving speed of the equipment.

[0078] In some embodiments, the lidar scanning parameters are optimized through feedback via the following mechanism:

[0079] Data quality monitoring: When the reconstruction error exceeds the threshold At that time, the wire harness density was forcibly increased to 64 lines;

[0080] Energy balance: The scanning frequency is dynamically adjusted according to the remaining battery power. When the battery power is below 30%, it switches to energy-saving mode and the wire harness density is fixed at 32 lines.

[0081] Speed ​​acquisition: The encoder pulse signal is captured by the STM32F407 main control board, and the real-time speed v is calculated;

[0082] Threshold determination: Compare v with the 1.5m / s threshold and generate a MOS transistor control signal;

[0083] Harness switching: The MOS transistor array turns off / on the corresponding laser emitter, with a switching time of less than 2ms;

[0084] Data verification: Verify the integrity of the generated point cloud data. If the packet loss rate exceeds 5%, trigger the retransmission mechanism.

[0085] S2. Perform compressed sensing reconstruction on the sparse sampled point cloud data, input the reconstructed 3D point cloud into the neural network model for target localization, and output the 3D coordinates of the target.

[0086] The compressed sensing reconstruction module receives sparsely sampled point cloud data generated by dynamically adjusting the wire bundle density, reconstructs the complete 3D point cloud through an iterative optimization algorithm, and inputs the reconstruction result into a lightweight neural network model to output the target's 3D coordinates. This process is linked to the velocity sensing mechanism of the laser scanning module, using a dynamic hardware resource allocation strategy to balance reconstruction accuracy and real-time performance, providing high-precision positioning input for subsequent spray control.

[0087] In some embodiments, compressed sensing reconstruction employs the Iterative Shrink Thresholding (ISTA) algorithm, the mathematical expression of which is:

[0088]

[0089] Where y represents sparsely sampled point cloud data, Φ is the measurement matrix, and S λ / L Here, λ is the soft thresholding function, λ is the sparsity constraint coefficient, and x is the sparseness constraint coefficient. (k) The signal estimate obtained in the k-th iteration represents the sparse target signal in compressed sensing reconstruction. The transpose of the measurement matrix Φ, L is the step size parameter of the gradient descent algorithm, which is related to the spectral norm of the measurement matrix, y is the sparsely sampled observation data, generated by the lidar beam dynamic adjustment module, and k is the iteration number index.

[0090] In one possible implementation, the rules for constructing the measurement matrix Φ include:

[0091] When the lidar beam density is 16 lines, Φ adopts a partial Fourier matrix, and its row vectors are randomly drawn from the standard Fourier basis.

[0092] When the wire density is 64 lines, Φ is switched to a Gaussian random matrix.

[0093] Specifically, the sparsity constraint coefficient λ is dynamically related to the equipment's moving speed v, satisfying:

[0094]

[0095] Wherein, λ(v) is the speed-related comprehensive evaluation coefficient, which quantifies the dynamic balance between equipment movement speed and operational accuracy, and v is the real-time movement speed of the equipment. max Measured by the encoder and processed by moving average filtering.

[0096] In some embodiments, the neural network employs an improved PointNet++ architecture, whose structural optimizations include:

[0097] Layer compression: The original network's third-level Set Abstraction (SA) layer is deleted, while the first two SA layers and Feature Propagation (FP) layer are retained; Channel number adjustment: The number of output channels of the first-level SA layer is set to 64, and the number of output channels of the second-level SA layer is compressed to 128;

[0098] Input normalization: The reconstructed point cloud coordinates (x, y, z) are normalized to the interval [-1, 1] by the maximum and minimum values.

[0099] In one possible implementation, the model training data is generated in the following way:

[0100] Simulation data: Three-dimensional models of corn and wheat were constructed using Blender, with plant height (0.5-2.0m) and spacing (0.3-0.8m) randomly set.

[0101] Data augmentation: Adding Gaussian noise And random occlusion (maximum occlusion rate 30%);

[0102] Loss function: Weighted centroid loss Where c is the centroid coordinate and IoU is the intersection-union ratio between the predicted and the actual point cloud.

[0103] As an alternative, neural network inference is accelerated through the following techniques:

[0104] Model quantization: Floating-point weights are converted to INT8 format, and piecewise linear approximation is used for the activation function;

[0105] Memory optimization: Layer fusion is achieved using the TensorRT engine, reducing the number of GPU memory swaps;

[0106] Post-processing: Retrieving the coarse-grained centroid coordinates c from the network output. coarse Perform Gaussian weighted optimization:

[0107]

[0108] Among them, c final The target coordinates, after Gaussian weighted optimization, have a higher accuracy than the coarse-grained coordinates directly output by the network. coarse The coarse-grained target coordinates output by the neural network are used as the initial center for weighted optimization, p. i Let K be the point cloud coordinates, and K be the maximum number of cluster points after point cloud clustering.

[0109] S3. Calculate the spraying error of the current position of the equipment based on the three-dimensional coordinates, and combine the acceleration data collected by the inertial measurement unit and the historical motion trajectory based on the time domain change rate of the spraying error.

[0110] Based on the target's three-dimensional coordinates output by a neural network, the spraying error at the current position of the equipment is calculated in real time. Combined with acceleration data collected by an inertial measurement unit (IMU) and historical motion trajectories, future acceleration is predicted through a stochastic process model. This process addresses the control delay problem under high-speed movement through time-domain error rate of change analysis and path space sensitivity quantification, providing forward-looking input for a fractional-order PID controller.

[0111] The spraying error e(t) is defined as the target centroid coordinate c target Compared with the current nozzle position p of the device current Euclidean distance:

[0112] e(t) = ||c target -p current ||2;

[0113] Where e(t) is the spraying error, c target Let p be the coordinates of the target centroid. current This indicates the nozzle position.

[0114] Current location of the device p current The positioning accuracy is less than 2cm, obtained through a GNSS module.

[0115] In one possible implementation, the rate of change of the time-domain error Calculated using first-order difference:

[0116]

[0117] Where e(t) is the spraying error, e(t-Δt) is the instantaneous rate of change of the error signal, which characterizes the rate of change of the error in the control system over time. e(t-Δt) is the historical error amount of the previous control cycle, which is stored in the circular buffer. Δt is the time step of the control system, that is, the interval between two adjacent error samples.

[0118] Specifically, the stochastic process model uses the Itō diffusion process to describe the dynamic characteristics of the device's acceleration, and its differential equation is expressed as:

[0119] da(t)=μ(a)dt+σ(a)dW t ;

[0120] Where da(t) is the instantaneous change in equipment acceleration, characterizing the stochastic dynamic characteristics of acceleration, a(t) is the equipment acceleration, μ(a) is the drift term function, determining the deterministic evolution trend of acceleration, and W t For the Wiener process, σ(a) is the coefficient of the diffusion term, quantifying the intensity of random perturbations on acceleration, dW t It is the differential of the Wiener process (standard Brownian motion).

[0121] In some embodiments, the path space sensitivity index D is calculated based on the Malliavin derivative. s a(t) is used to correct the prediction results:

[0122]

[0123] Among them, D s a(t) is the path space sensitivity index, a(t+δt) is the predicted acceleration value at the future time t+δt, a(t-δt) is the measured acceleration value at the historical time t-δt, and δt is the differential time interval, which controls the calculation resolution of the sensitivity index.

[0124] S4. Input the predicted future acceleration and spray error into the fractional-order PID controller to generate nozzle control commands; the fractional-order PID controller receives the future acceleration from the prediction module. Sprinkler control commands are generated through fractional calculus operations. This process combines the equipment's kinematic model with the dynamic characteristics of errors to achieve nonlinear delay compensation and stability control, ensuring precise matching between sprinkler actions and the equipment's motion trajectory.

[0125] In this embodiment, the mathematical expression for the fractional-order PID controller is:

[0126]

[0127] in, This represents a fractional integral operator of order α. K is a fractional differential operator of order β. p K i K d These are the proportional, integral, and derivative gain coefficients, respectively; u(t) is the controller's output command; and e(t) is the real-time control error, defined as the deviation between the setpoint and the feedback value.

[0128] Specifically, fractional calculus operators are implemented using time-domain approximation algorithms:

[0129] Integral operator: The Riemann-Liouville definition is adopted, and the weighted sum of historical errors is calculated by Grünwald-Letnikov difference formula after discretization;

[0130] Differential operator: It adopts the Caputo definition and is implemented by convolving the error rate of change within a short time window with an exponentially decaying kernel function.

[0131] Controller parameters (K) p K i K d (α, β) are tuned using a frequency distribution algorithm, the specific steps of which include:

[0132] Frequency domain response test: Inject a frequency sweep signal into the device and collect nozzle position response data;

[0133] Phase margin optimization: Adjust parameters to make the phase margin of the open-loop transfer function at the cutoff frequency greater than a preset threshold;

[0134] Robustness verification: Introduce perturbation signals to verify the tolerance of parameters to model uncertainties.

[0135] The controller output commands are dynamically linked with the device's kinematic model. The specific implementation method is as follows:

[0136] Directional correlation compensation: When the spray error direction is opposite to the equipment movement direction, the weight of the differential term is increased to suppress overshoot; Acceleration feedforward compensation: The predicted acceleration is... As a feedforward quantity injection control command.

[0137] The controller's output commands are transmitted to the nozzle drive circuit via the D / A conversion module, while closed-loop stability is ensured through the following mechanisms:

[0138] Command limiting: Limits the output voltage within the hardware-allowed range to prevent actuator saturation;

[0139] Anti-integral saturation: When the error continues to exceed the limit, the accumulation of the integral term is paused to avoid control drift;

[0140] The design of the fractional-order PID controller fully considers the data coupling characteristics with the upstream module:

[0141] Predictive acceleration fusion: the output of the prediction module The phase lag caused by equipment inertia is compensated by injection through the feedforward channel; error dynamic characteristics are adapted: the fractional-order calculus operator is more suitable for the non-exponential convergence characteristics of spraying error than the integer-order PID.

[0142] Hardware resource collaboration: The control algorithm is implemented in parallel on the FPGA, and the operation cycle is strictly synchronized with the data acquisition cycle.

[0143] S5. Based on the absolute value of the real-time spraying error, a non-smooth correction term is superimposed on the control command, and the control signal after stability correction is output.

[0144] For the initial control command output by the fractional-order PID controller, a non-smooth correction term is superimposed based on the absolute value of the real-time spray error to generate the final control signal with optimized stability. This process solves the problem of sluggish response to sudden errors in traditional linear correction by using an error amplitude triggering mechanism and a dynamic gain adjustment strategy. At the same time, it forms a closed-loop adaptation with the fractional-order arithmetic logic of the upstream control module and the physical constraints of the downstream actuator.

[0145] In some embodiments, the activation condition of the non-smooth correction term is strongly correlated with the absolute value of the real-time spray error and the device motion state. When the absolute value of the spray error exceeds a preset threshold, a compensation amount in the same direction as the error sign is injected into the control command, and the magnitude of the compensation amount is determined by the product of the error magnitude and the dynamic gain coefficient.

[0146] Specifically, the logic for superimposing correction terms includes the following core rules:

[0147] Threshold triggering mechanism: The absolute value threshold of the error is set to 0.01 meters. When the real-time error exceeds this threshold, the non-smooth correction term is activated.

[0148] Sign-direction compensation: The direction of the correction term is always consistent with the direction of the error, that is, the compensation amount is positive when the error is positive and negative when the error is negative;

[0149] Dynamic gain adjustment: The base gain coefficient is set to 0.05 V·s / m. When the device acceleration exceeds 3 m / s², the gain coefficient is increased to 0.08 V·s / m.

[0150] In one possible implementation, the dynamic adjustment of the gain coefficient is linked to the device acceleration and the rate of change of error. Acceleration data is acquired in real time by the inertial measurement unit and filtered through a sliding window to eliminate high-frequency noise. The rate of change of error is calculated as a weighted average of the error differences from the most recent three control cycles, with weighting coefficients allocated at 0.5, 0.3, and 0.2.

[0151] Preferably, when the device is in a state of rapid acceleration or deceleration, an additional feedforward compensation factor is introduced. This factor is proportional to the square root of the predicted acceleration and is used to compensate for insufficient correction caused by inertial lag.

[0152] In some embodiments, the convergence of the modified control signal is verified using Lyapunov stability theory. An energy function is constructed that includes a combination of error terms and their rates of change. By determining whether the time derivative of the energy function is negative definite, it is ensured that the system converges to equilibrium within a finite time.

[0153] Specifically, when the derivative of the energy function does not satisfy the negative definite condition, the following protection mechanism is triggered:

[0154] Force the gain factor to be reduced to a safe threshold;

[0155] Temporarily freeze the integral term calculation to prevent error accumulation;

[0156] Send a speed reduction request to the upstream path planning module until stability is restored.

[0157] In this embodiment, the modified control signal works in conjunction with other modules of the system in the following way:

[0158] Command limiting protection: The final output signal is limited to the voltage tolerance range of the nozzle drive circuit to prevent hardware overload; Timestamp synchronization: The correction item superposition process is strictly synchronized with the controller operation cycle, with a timing deviation of less than 50 microseconds;

[0159] Feedback closed-loop linkage: When the number of consecutive triggers of the correction item exceeds the preset number, a parameter reset command is sent to the laser scanning module to recalibrate the target positioning reference.

[0160] The superposition logic of non-smooth correction terms is deeply coupled with the output characteristics of the upstream control module and the physical constraints of the downstream actuator:

[0161] Fractional PID compatibility: The correction term is superimposed after the controller completes the fractional calculus operation to avoid interfering with the long memory effect of the fractional operator;

[0162] Nozzle drive adaptation: The voltage amplitude of the correction is matched with the response slope of the nozzle piezoelectric ceramic to prevent mechanical resonance caused by step jumps;

[0163] Real-time performance guarantee: The calculation cycle of the correction item is compressed to less than 10% of the main control cycle, and parallel processing is achieved through hardware acceleration circuits.

[0164] Multiple anomaly detection mechanisms are set up during the non-smooth correction term stacking process:

[0165] Error mutation protection: When the error change rate exceeds the critical value, the correction term superposition is paused and the system switches to pure proportional control mode; Hardware status monitoring: The temperature and current data of the nozzle drive circuit are read in real time. If an overload risk is detected, the correction amount is reduced exponentially.

[0166] Data integrity verification: Perform CRC check on the received control commands, and use the data from the previous cycle to replace the data if the check fails.

[0167] S6. Drive the high-frequency piezoelectric nozzle to perform spraying action according to the corrected control signal, and feed back the spray coverage to the lidar to dynamically adjust the scanning frequency and sampling rate.

[0168] The high-frequency piezoelectric nozzle array is driven by a stability-corrected control signal to perform spraying actions. A multispectral imaging module collects real-time spray coverage data and feeds it back to the lidar, dynamically adjusting the scanning frequency and sampling rate. This process forms a closed-loop control architecture, solving the problems of spray blind zone compensation and resource optimization, while also being deeply coupled with the instruction generation logic of the upstream control module and the physical characteristics of the downstream actuators.

[0169] The generation of the drive signal for the piezoelectric nozzle follows these rules:

[0170] Fixed pulse width: The pulse width of the control signal is set to 10 microseconds, and the rise time is less than 0.5 microseconds to match the mechanical response characteristics of piezoelectric ceramics;

[0171] Delay compensation: The nozzle triggering timing is fed forward based on the predicted acceleration, and the compensation time is equal to 0.5 times the product of the predicted acceleration and the square of the control period.

[0172] Specifically, when the equipment is in an accelerated state, the nozzle trigger time is advanced to compensate for the positional difference caused by inertia.

[0173] In one possible implementation, the piezoelectric nozzle array is arranged orthogonally to the lidar scanning direction, and each nozzle independently receives a control signal. The nozzle driving voltage is dynamically allocated based on the local point cloud density.

[0174] When the point cloud density is higher than 150 points / cubic meter, the corresponding nozzle drive voltage is increased to 8 volts;

[0175] When the point cloud density is less than 50 points / cubic meter, the driving voltage drops to below 3 volts;

[0176] The voltage gradient between adjacent nozzles is limited to less than 5 volts per second.

[0177] Spray coverage is calculated by acquiring post-spray environmental images using a multispectral camera.

[0178] Image preprocessing: Near-infrared band images are extracted, and Gaussian filtering is used to eliminate illumination interference;

[0179] Region segmentation: In the HSV color space, set the hue threshold range to 100-140, the saturation threshold to be greater than 0.3, and the lightness threshold to be greater than 0.4;

[0180] Coverage calculation: Calculate the percentage of pixels that meet the threshold. When the coverage is below 85%, parameter adjustment is triggered.

[0181] Preferably, the multispectral camera is rigidly connected to the nozzle array, and the optical field of view covers more than 120% of the spray area, ensuring the integrity of edge area detection.

[0182] The adjustment rules for lidar scanning frequency and compressed sensing sampling rate include:

[0183] Frequency enhancement: When the coverage is below the threshold, the LiDAR scanning frequency is increased from 10 Hz to 20 Hz, and the compressed sensing sampling rate is increased from 50% to 80%.

[0184] Resource optimization: In low-density areas (coverage greater than 95%), switch to energy-saving mode, reduce the scanning frequency to 5 Hz, and reduce the sampling rate to 30%;

[0185] Data closure: The coordinate information of the uncovered areas is added to the neural network training dataset, and the target positioning accuracy is optimized in the next cycle through incremental learning.

[0186] Specifically, the nozzle drive circuit incorporates multiple protection mechanisms:

[0187] Temperature monitoring: When the temperature of the piezoelectric ceramic exceeds 60 degrees Celsius, the frequency reduction operation mode is triggered, and the driving frequency is reduced by 50%; Current limitation: The driving current is monitored in real time. If it exceeds 2 amps for 3 consecutive cycles, the power supply to the corresponding nozzle is cut off and a fault code is reported.

[0188] Waterproof design: The circuit board is coated with conformal coating, achieving an IP67 protection rating, making it suitable for high humidity environments.

[0189] In this embodiment, the actuator works in collaboration with other modules of the system in the following ways:

[0190] Timing synchronization: The nozzle trigger signal is aligned with the lidar scanning cycle, with the deviation controlled within 1 millisecond;

[0191] Data verification: The feedback coverage data includes a timestamp and device location information, and is spatially aligned with the point cloud data;

[0192] Resource preemption: When the coverage detection module issues an emergency adjustment command, the LiDAR immediately interrupts the current scanning task and prioritizes high-frequency sampling.

[0193] The nozzle drive and feedback adjustment mechanism are fully adapted to the overall system architecture characteristics:

[0194] Control signal compatibility: The corrected control signal voltage range matches the input impedance of the drive circuit to avoid signal reflection loss; Real-time performance guarantee: The nozzle response delay is less than 15 microseconds to ensure synchronization with the movement trajectory of high-speed mobile equipment;

[0195] Energy efficiency optimization: Dynamically adjusting the scanning frequency reduces the power consumption of the lidar by 40%, extending the outdoor operation time.

[0196] The precision spray control device based on the combination of laser scanning and neural networks described below can be referred to in correspondence with the precision spray control method based on the combination of laser scanning and neural networks described above.

[0197] Please see the appendix Figure 2 The present invention also provides a precision spray control device based on the combination of laser scanning and neural networks, comprising:

[0198] The laser scanning module includes a lidar with adjustable beam density and an inertial measurement unit, used to acquire three-dimensional point clouds of the environment and equipment motion data in real time.

[0199] The data processing module, which is connected to the laser scanning module via a high-speed data interface, includes an FPGA accelerator and an embedded processor. The FPGA is configured to execute a compressed sensing reconstruction algorithm, and the embedded processor is configured to run a lightweight PointNet++ neural network model.

[0200] The control calculation module is connected to the data processing module via the PCIe bus. It includes a fractional PID controller and a non-smooth correction unit. The fractional PID controller receives the coordinate and device position error and the predicted acceleration to generate the initial control command. The non-smooth correction unit adds a stability compensation term according to the error threshold.

[0201] The actuator module, connected to the control calculation module via a PWM interface, includes a piezoelectric nozzle array and a high-voltage drive circuit. The trigger delay of the piezoelectric nozzle is less than 15 microseconds.

[0202] The feedback closed-loop module and the actuator module are configured to transmit the sprinkler coverage data back to the data processing module via the wireless communication module, triggering dynamic adjustments to the lidar scanning frequency and compressed sensing sampling rate to form a closed-loop control.

[0203] The device in this embodiment can be used to execute the above method embodiments, and its principle and technical effects are similar, so they will not be described again here.

[0204] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A precise spray control method based on the combination of laser scanning and neural networks, characterized in that, Includes the following steps: The environment is scanned by lidar, and the scanning beam density is dynamically adjusted based on the real-time speed of the device to generate sparse sampling point cloud data. Compressed sensing reconstruction is performed on the sparse sampled point cloud data, and the reconstructed 3D point cloud is input into a neural network model for target localization, outputting the 3D coordinates of the target; The spraying error at the current position of the device is calculated based on the three-dimensional coordinates. Based on the time-domain rate of change of the spraying error, combined with the acceleration data collected by the inertial measurement unit and the historical motion trajectory, the future acceleration is predicted through a stochastic process prediction model. The predicted future acceleration and spray error are input into a fractional-order PID controller to generate nozzle control commands. Based on the absolute value of the real-time spraying error, a non-smooth correction term is superimposed on the control command to output a control signal with stability correction. The high-frequency piezoelectric nozzle is driven to perform spraying action according to the modified control signal, and the spray coverage rate is fed back to the lidar to dynamically adjust the scanning frequency and sampling rate. The stochastic process prediction model includes: The Iton process model based on device acceleration has a drift term that is a linear decay function of acceleration and a diffusion term that is a random perturbation term proportional to the square root of acceleration. By calculating the Malliavin derivative to quantify the path space sensitivity, and combining the acceleration change trend in historical motion trajectories, the expected acceleration value in the future time period is predicted. The prediction model is coupled to the input channel of the fractional-order PID controller. When the predicted acceleration changes abruptly beyond a preset threshold, an emergency correction mode of the control command is automatically triggered.

2. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The dynamic adjustment of scan beam density: When the device moves faster than a preset first speed threshold, the laser radar scanning beam density is reduced to the first beam density. When the device moves at a speed less than the first speed threshold, the scanning beam density is increased to the second beam density, and the beam density adjustment is synchronized with the sampling rate of the compressed sensing reconstruction.

3. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The compressed sensing reconstruction includes: An iterative shrinkage threshold algorithm is used to reconstruct sparse sampled point cloud data, wherein the measurement matrix dynamically adjusts the sparsity constraint weights according to the device's moving speed; The neural network model is a lightweight PointNet++ architecture. Its input layer receives the reconstructed 3D point cloud coordinates, and its output layer extracts the centroid position of the target through max pooling operation. The compressed sensing reconstruction and neural network model inference share computing resources. When the point cloud density is lower than a preset threshold, computing power is preferentially allocated to the reconstruction algorithm.

4. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The fractional-order PID controller includes: Differential operators are defined using the first-order definition, while integral operators are defined using the second-order definition. The control parameters are tuned using a frequency distribution algorithm, with phase margin and gain margin as optimization objectives; The output command of the fractional-order PID controller is linked with the kinematic model of the equipment. When the direction of the spraying error is opposite to the direction of equipment movement, the weight of the derivative term is dynamically increased to suppress overshoot.

5. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The superposition logic of the non-smooth correction term is as follows: When the absolute value of the real-time spraying error exceeds the preset first error threshold, a non-smooth compensation term in the same direction as the error sign is superimposed on the control command, and the compensation amount is the product of the preset gain coefficient and the absolute value of the error. The activation state of the non-smooth correction term is related to the device acceleration; The corrected control signal is convergent through a Lyapunov function to ensure that the error decays to below the second error threshold within a preset time threshold.

6. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The driving parameters of the high-frequency piezoelectric nozzle include: The pulse width is fixed to a preset time length and has a linear mapping relationship with the driving voltage; The nozzle triggering timing is delayed and compensated based on the predicted acceleration, and the compensation time is calculated based on the predicted acceleration and the square of the control period; The array of piezoelectric nozzles is orthogonal to the laser scanning direction, and the driving signal of each nozzle is independently adjustable to match the local density distribution of the target.

7. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The spray coverage feedback includes: The sprayed area was captured by a multispectral camera, and the coverage percentage was calculated by segmenting the sprayed area using the HSV color space. When the coverage rate is lower than the preset coverage rate threshold, the LiDAR scanning frequency is increased to the first frequency, and the sampling rate of compressed sensing is increased to the first sampling rate. The feedback mechanism is linked to the training data of the neural network model, adding the coordinates of the uncovered areas to the next round of training dataset to optimize positioning accuracy.

8. The precise spray control method based on laser scanning and neural network according to claim 1, characterized in that, The sparse sampled point cloud data includes: The density range is a preset density range, the signal-to-noise ratio is greater than a preset signal-to-noise ratio threshold, and the point cloud distribution follows a Laplace distribution in the direction of device motion. The sparsity of the point cloud data is dynamically correlated with the device speed. When the speed changes, the sparsity is dynamically adjusted according to the rate of change of the device speed. Point cloud attributes serve as prior constraints for compressed sensing reconstruction and are used to calculate the regularization term coefficients in the iterative shrinkage threshold algorithm.

9. A precision spray control device based on laser scanning and neural network, used to execute the precision spray control method based on laser scanning and neural network as described in any one of claims 1-8, characterized in that, include: The laser scanning module includes a lidar with adjustable beam density and an inertial measurement unit, used to acquire three-dimensional point cloud data of the environment and equipment motion data in real time. The data processing module is connected to the laser scanning module via a high-speed data interface and includes an FPGA accelerator and an embedded processor. The FPGA is configured to execute a compressed sensing reconstruction algorithm, and the embedded processor is configured to run a lightweight PointNet++ architecture neural network model. The control calculation module is connected to the data processing module via the PCIe bus. It includes a fractional-order PID controller and a non-smooth correction unit. The fractional-order PID controller receives the coordinate and device position error and the predicted acceleration to generate an initial control command. The non-smooth correction unit adds a stability compensation term according to the error threshold. The actuator module is connected to the control calculation module via a PWM interface and includes a piezoelectric nozzle array and a high-voltage drive circuit. The triggering delay of the piezoelectric nozzle is less than 15 microseconds. The feedback closed-loop module is configured to transmit sprinkler coverage data back to the data processing module via a wireless communication module, triggering dynamic adjustments to the lidar scanning frequency and compressed sensing sampling rate to form closed-loop control.

Images