Garlic harvester automatic row control method and system based on machine vision

Abstract

The invention discloses a garlic harvester automatic row control method and system based on machine vision, and belongs to the field of agricultural machinery automatic control, and the method comprises the steps: employing a deep learning network to generate a plant density distribution diagram based on image data obtained through machine vision; based on the plant density distribution diagram, a density and speed coupling dynamic model is constructed, and the optimal advancing speed of the garlic harvester under the current plant density condition is solved through a nonlinear mapping function and an optimization algorithm; compensating visual perception based on the optimal advancing speed to obtain compensated visual perception information; based on the compensated visual perception information and the optimal advancing speed, constructing a multi-time scale prediction control framework to obtain a control strategy; and based on actual operation data, optimizing a density-speed coupling dynamic model and a control strategy, and realizing automatic row control of the garlic harvester. According to the invention, traditional multiple sets of speed control and path detection strategies are replaced, and the system adaptability is improved while the complexity of the control system is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic control of agricultural machinery, and in particular relates to a method and system for automatic row alignment control of a garlic harvester based on machine vision. Background Art

[0002] As modern agriculture evolves toward intelligent and automated processes, machine vision-based automatic control technologies for agricultural machinery have garnered widespread attention. For garlic harvesting, using machine vision to identify fallen garlic vines and the boundaries to be harvested, and developing automated control systems for garlic harvesting in rows and along edges, has become a key research direction. Related technologies also include the development of a flexible, active, straw-supporting feeding mechanism based on garlic growth patterns; research on methods for detecting and providing early warnings for weed and film blockage during the garlic harvest process; the development of a precise, flexible, positioning, and cutting mechanism for double-layer garlic cutting with an active weed removal system; the use of gravity sensors for automatic unloading devices; and the integrated development of automatic bag changing devices and yield estimation systems.

[0003] Existing mechanical garlic harvesting technology has problems with row alignment, missed harvests, high damage rates, and a lack of automatic unloading devices. Specifically, existing machine vision-based automatic row alignment control technology for garlic harvesters has the following technical issues:

[0004] Traditional garlic harvesters typically use a fixed-speed operation mode, which cannot adapt to changes in plant density in the field. In high-density areas, fixed high-speed operation will lead to reduced tracking accuracy and increased garlic damage rate. In low-density areas, fixed low-speed operation results in low operating efficiency, wasting energy and time. In existing technologies, machine vision systems and speed control systems are usually independent of each other and lack an effective coordination mechanism. The vision system cannot automatically adjust its parameter configuration according to the current travel speed, resulting in reduced visual recognition accuracy during high-speed operation. When the harvester speed changes, the blur level, exposure conditions, and field of view of the image acquisition will change accordingly. Existing technologies lack a compensation mechanism for the impact of speed changes on visual perception, making it difficult for the control system to cope with speed-varying scenarios. Traditional row control systems have fixed parameters and cannot continuously optimize control strategies based on actual operation data, making it difficult to adapt to complex and changing field environments. Existing technologies usually use a single control strategy, which cannot simultaneously take into account short-term precision control and long-term path planning, and the control effect is poor under conditions of complex terrain and uneven plant distribution.

[0005] Therefore, it is necessary to improve the intelligence level of garlic sowing and harvesting, innovate and develop unmanned production equipment for garlic, and carry out industrialization and demonstration promotion. Improving the level of unmanned garlic production is an important guarantee for promoting the sustainable and healthy development of the garlic industry. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention proposes a method and system for automatic row alignment control of a garlic harvester based on machine vision to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above objectives, the present invention provides a method for automatic row alignment control of a garlic harvester based on machine vision, comprising:

[0008] Based on the image data obtained by machine vision, a deep learning network is used to analyze the plant density characteristics of the current area and the area ahead in real time to generate a plant density distribution map;

[0009] A density-speed coupling dynamic model is constructed based on the plant density distribution map, and the optimal travel speed of the garlic harvester under the current plant density condition is solved by a nonlinear mapping function and an optimization algorithm;

[0010] Based on the optimal travel speed, adaptive image preprocessing, dynamic adjustment of the field of view, and dynamic adjustment of feature extraction parameters are used to compensate for visual perception, thereby obtaining compensated visual perception information;

[0011] constructing a multi-time-scale predictive control framework based on the compensated visual perception information and the optimal travel speed, and obtaining a control strategy based on the multi-time-scale predictive control framework;

[0012] Based on actual operation data, the density-speed coupling dynamic model and control strategy are continuously optimized to achieve automatic row control of the garlic harvester.

[0013] Optionally, based on image data acquired by machine vision, a deep learning network is used to analyze plant density characteristics of the current area and the area ahead in real time, and the process of generating a plant density distribution map includes:

[0014] Extract features from the collected images based on a multi-scale convolutional neural network to generate multi-scale features;

[0015] The multi-scale features are fused through the feature pyramid network, and the key area features are highlighted in combination with the attention mechanism to generate a preliminary density feature map;

[0016] The preliminary density feature map is processed by a density regression network to generate a plant density distribution map within the current field of view.

[0017] Optionally, the process of obtaining the optimal travel speed of the garlic harvester under the current plant density condition includes:

[0018] constructing a system adaptability index including an efficiency weight coefficient, a speed penalty coefficient, an accuracy weight coefficient, and an acceleration compensation coefficient based on the plant density distribution map, and determining a density-speed coupling dynamics model based on the system adaptability index;

[0019] A nonlinear mapping function was used to normalize plant density, and the interaction between plant density and travel speed was quantified using a density-speed interaction function.

[0020] Based on the interactive effect between plant density and travel speed, the gradient descent method is used to optimize and correct the parameters of the density-speed coupling dynamic model;

[0021] Based on the optimized parameters, the numerical optimization method is used to solve the travel speed that optimizes the system adaptability index, which is used as the optimal travel speed under the current plant density conditions.

[0022] Optionally, the expression of the system adaptability index is:

[0023] ;

[0024] Where, is the system adaptability index; is the plant density; is the travel speed; 、 、 、 They represent the efficiency weight coefficient, speed penalty coefficient, accuracy weight coefficient, and acceleration compensation coefficient respectively; is the normalized mapping function of plant density; is the density-velocity interaction function; is the acceleration, and · represents the dot product.

[0025] Alternatively, the expression for quantifying the interactive effect of plant density and travel speed through the density-speed interaction function is:

[0026] ;

[0027] Where, is the density-velocity interaction function; is the plant density; is the travel speed; 、 、 They represent the density linear influence coefficient, velocity linear influence coefficient, and density-velocity quadratic interaction coefficient, respectively. · represents the dot product.

[0028] Optionally, based on the optimal travel speed, adaptive image preprocessing, dynamic adjustment of the field of view, and dynamic adjustment of feature extraction parameters are used to compensate for visual perception, and a process of obtaining compensated visual perception information includes:

[0029] establishing a speed image ambiguity mapping model based on the optimal traveling speed and image quality;

[0030] Adaptively preprocessing the collected image based on the current traveling speed and the speed image ambiguity mapping model to obtain an image quality assessment result;

[0031] The feature extraction parameters of the density estimation network are dynamically adjusted based on the current travel speed and the image quality evaluation result to obtain compensated visual perception information.

[0032] Optionally, the process of establishing the velocity image ambiguity mapping model includes:

[0033] Establish a mapping relationship between speed and blur kernel standard deviation;

[0034] Model the relationship between speed and image contrast;

[0035] Establish a model of the relationship between speed and texture degradation;

[0036] A velocity image fuzziness mapping model is established based on the mapping relationship between velocity and blur kernel standard deviation, the relationship model between velocity and image contrast, and the relationship model between velocity and texture degradation.

[0037] Optionally, the process of building a multi-timescale predictive control framework includes:

[0038] The control domain is divided into short-term domain, medium-term domain and long-term domain according to the time domain;

[0039] Constructing a hierarchical prediction model based on the short-term time domain, the medium-term time domain, and the long-term time domain;

[0040] Obtaining control decisions at different time scales based on the hierarchical prediction model;

[0041] The control decisions at different time scales are coordinated through a hierarchical decision structure and constraint hardening mechanism to obtain the final control instructions.

[0042] Optionally, the process of continuously optimizing the density-speed coupling dynamics model and control strategy based on actual operation data to achieve automatic row control of the garlic harvester includes:

[0043] Collect and store plant density distribution, optimal travel speed, visual compensation parameters, and control execution effect data during the operation in real time to build an annotated operation dataset;

[0044] An incremental learning method is used to update the parameters of the density-velocity coupled dynamics model, and new weights are assigned through a window weighting strategy.

[0045] Based on the reinforcement learning framework, the multi-timescale predictive control framework is optimized using line tracking deviation and operation efficiency as reward signals;

[0046] Based on the new weights and the optimized multi-time-scale predictive control framework, a knowledge base containing the characteristics of different operation scenarios is established. Based on the knowledge base containing the characteristics of different operation scenarios, adaptive migration of control parameters is achieved through similarity calculation, and optimization experience is shared during multi-machine collaborative operation.

[0047] The present invention also provides a method for automatically controlling rows of a garlic harvester based on machine vision, which is used to implement the method. The system includes:

[0048] The density perception module is used to analyze the plant density characteristics of the current area and the area ahead in real time based on image data acquired by machine vision using a deep learning network to generate a plant density distribution map;

[0049] A dynamic modeling module is used to construct a density-speed coupling dynamic model based on the plant density distribution map, and solve the optimal travel speed of the garlic harvester under the current plant density conditions through a nonlinear mapping function and an optimization algorithm;

[0050] a visual compensation module for compensating visual perception based on the optimal travel speed by using adaptive image preprocessing, dynamic adjustment of the visual field, and dynamic adjustment of feature extraction parameters to obtain compensated visual perception information;

[0051] a multi-scale control module, configured to construct a multi-time-scale predictive control framework based on the compensated visual perception information and the optimal travel speed, and obtain a control strategy based on the multi-time-scale predictive control framework;

[0052] The self-learning optimization module is used to continuously optimize the density-speed coupling dynamic model and control strategy based on actual operation data to achieve automatic row control of the garlic harvester.

[0053] Compared with the prior art, the present invention has the following advantages and technical effects:

[0054] The present invention provides a method for automatic row control of a garlic harvester based on machine vision. Through a deep learning network, plant density characteristics are analyzed in real time and a density distribution map is generated, achieving accurate perception of the field environment. The travel speed is dynamically adjusted based on a density-speed coupling dynamics model, effectively solving the problems of insufficient precision and low efficiency caused by the traditional fixed-speed operation mode. An adaptive visual compensation mechanism ensures the stability of the image recognition system at different speeds. A multi-time-scale predictive control framework is adopted to coordinate short-term precise control with long-term path planning, significantly improving the accuracy and adaptability of row control. Through a continuous learning optimization mechanism, the system can continuously accumulate experience from actual operations and improve itself, so that the control performance gradually improves over time. This method realizes intelligent control of garlic harvesting operations, effectively reduces the missed picking rate and damage rate, and improves operation efficiency and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0056] Figure 1 This is a flow chart of a method for automatic row alignment control of a garlic harvester based on machine vision according to an embodiment of the present invention;

[0057] Figure 2 A line graph comparing the recognition accuracy of the traditional density estimation method of an embodiment of the present invention and the deep learning density estimation network of this technical solution under different lighting conditions;

[0058] Figure 3 A bar chart comparing system adaptability indicators under different plant densities and travel speeds according to an embodiment of the present invention;

[0059] Figure 4 A dual-axis combination diagram showing changes in image quality at different travel speeds and improvement effects brought about by the compensation mechanism according to an embodiment of the present invention;

[0060] Figure 5 A boxplot showing the distribution of row tracking deviations of a garlic harvester before and after adopting a multi-time-scale predictive control framework according to an embodiment of the present invention;

[0061] Figure 6 This is an area chart showing the performance improvement trend of the self-learning optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0063] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0064] like Figure 1 As shown, this embodiment provides a method for automatic row control of a garlic harvester based on machine vision, comprising the following steps:

[0065] Step 1: Analyze machine vision image data using a deep learning network, estimate plant density characteristics for the current area and the area ahead in real time, and generate a plant density distribution map. This process uses a deep learning network to analyze machine vision image data, estimate plant density characteristics for the current area and the area ahead in real time, and generate a plant density distribution map. This process specifically includes the following sub-steps.

[0066] Step 1.1: Multi-scale feature extraction. A multi-scale convolutional neural network is applied to extract features from the captured images, generating feature maps at different scales. This network consists of four convolutional layers, each followed by batch normalization and a ReLU activation function, to extract local texture features, mid-scale shape features, and global spatial distribution features, respectively. Through multi-scale feature extraction, the system can simultaneously capture both the micromorphological and macroscopic distribution characteristics of the garlic plant.

[0067] The detailed structure of the multi-scale convolutional neural network is as follows: input layer: receives the original RGB image from the camera (size is 640×480 pixels); the first convolution unit: contains 32 3×3 convolution kernels, extracts local texture features, followed by a batch normalization layer and a ReLU activation function, and finally downsampled by a 2×2 maximum pooling layer; the second convolution unit: contains 64 3×3 convolution kernels, extracts shape features of medium complexity, also followed by a batch normalization layer and a ReLU activation function, and downsampled by a 2×2 maximum pooling layer; the third convolution unit: contains 128 3×3 convolution kernels, captures more abstract features, followed by a batch normalization layer and a ReLU activation function, and downsampled by a 2×2 maximum pooling layer; the fourth convolution unit: contains 256 3×3 convolution kernels, extracts high-level semantic features and global spatial relationships, and is followed by a batch normalization layer and a ReLU activation function.

[0068] By adjusting the receptive field size of features at different levels, the multi-scale convolutional neural network can simultaneously focus on the local detailed features of garlic plants (such as leaf texture and shape) and global distribution patterns (such as row spacing and plant arrangement), providing comprehensive feature expression for subsequent density estimation.

[0069] Step 1.2: Density Feature Aggregation and Analysis. Multi-scale features are fused using a feature pyramid network, combined with an attention mechanism to highlight key regional features, generating a preliminary density feature map. During the feature fusion process, an adaptive weight allocation strategy is employed to automatically adjust the weights of features at different scales based on the current scene conditions, improving the system's adaptability to various plant density patterns.

[0070] The specific implementation method of the feature pyramid network is as follows: top-down path: restore the high-level semantic features (output of the fourth convolution unit) to the same spatial resolution as the low-level features through upsampling; horizontal connection: connect the upsampled high-level features with the low-level features of the corresponding resolution (using element-level addition or channel-level connection); fusion feature processing: adjust the number of channels and fuse the features of the connected features through 1×1 convolution.

[0071] The attention mechanism adopts a spatial attention module, whose structure includes: convolution dimensionality reduction: using 1×1 convolution to reduce the number of feature map channels; attention map calculation: generating a spatial attention weight map through a series of convolution operations; feature weighting: applying the attention weight map to the original features to highlight key area information.

[0072] The adaptive weight allocation strategy dynamically adjusts the importance of features at each scale based on the complexity of the current scene, lighting conditions, and plant density distribution, thereby improving the system's adaptability to changing environments.

[0073] Step 1.3: Generate a density distribution map. The fused feature map is processed through a density regression network to generate a plant density distribution map within the current field of view. Density trends in the forward area are predicted based on the historical image sequence. The density regression network uses a fully convolutional architecture to output a density distribution map that matches the input image size. Each pixel value represents an estimated plant density value for the corresponding area.

[0074] The detailed structure of the density regression network is as follows: feature decoder: contains four transposed convolution layers, which gradually restore the spatial resolution of the feature map; the first transposed convolution layer: 128 3×3 convolution kernels with a step size of 2, upsampling the feature map; the second transposed convolution layer: 64 3×3 convolution kernels with a step size of 2, continuing upsampling; the third transposed convolution layer: 32 3×3 convolution kernels with a step size of 2, further upsampling; the fourth transposed convolution layer: 16 3×3 convolution kernels with a step size of 1, fine-tuning the features; output layer: 1×1 convolution layer, converting the feature map into a single-channel density map, and each pixel value represents the plant density estimate at the corresponding position; activation function: the ReLU activation function is used to ensure that the density estimation result is non-negative.

[0075] The density change trend prediction of the forward area adopts a recurrent neural network structure, specifically: input: a sequence of density distribution maps of 5 consecutive frames; feature extraction: using 3D convolution to extract spatiotemporal features; time series modeling: processing time series information through LSTM (long short-term memory network) units; prediction output: the fully connected layer generates a density change trend map within the next 3 seconds.

[0076] To verify the effectiveness of the plant density estimation network in this technical solution, Figure 2As shown in the figure, the accuracy comparison between the traditional density estimation method and the deep learning density estimation network of this technical solution under different lighting conditions is demonstrated. The density estimation accuracy of this technical solution is better than that of the traditional method under various lighting conditions, especially under low light conditions, where the accuracy improvement is more obvious, verifying the effectiveness of multi-scale feature extraction and attention mechanism.

[0077] Step 1.4: Calculate the rate of density change. Based on time series analysis of density distribution maps in consecutive frames, the spatial gradient and rate of change of plant density are calculated, providing a basis for subsequent speed adjustments. By analyzing density changes between consecutive frames using an optical flow algorithm, the system can predict density trends in the area ahead, enabling proactive control.

[0078] Step 2: Based on the plant density distribution map, a density-speed coupled dynamic model was constructed. A nonlinear mapping function and optimization algorithm were used to determine the optimal travel speed of the garlic harvester under the current plant density conditions. A mathematical relationship model was constructed between plant density, travel speed, and control accuracy, which quantitatively described the optimal travel speed under different plant density conditions.

[0079] Furthermore, the process of obtaining the optimal travel speed of a garlic harvester under current plant density conditions includes: constructing a system adaptability index based on the plant density distribution map, including an efficiency weight coefficient, a speed penalty coefficient, an accuracy weight coefficient, and an acceleration compensation coefficient, and determining a density-speed coupling dynamic model based on the system adaptability index; normalizing the plant density using a nonlinear mapping function, and quantifying the interactive effect of plant density and travel speed using a density-speed interaction function; optimizing and correcting the parameters of the density-speed coupling dynamic model based on the interactive effect of plant density and travel speed using a gradient descent method; and using the optimized parameters, using a numerical optimization method to solve the travel speed that optimizes the system adaptability index as the optimal travel speed under current plant density conditions. Specifically, the process includes the following sub-steps.

[0080] Step 2.1, system adaptability index definition. Define system adaptability index , used to comprehensively evaluate the specific plant density and travel speed The overall performance of the system under the following conditions:

[0081] ;

[0082] in is the system adaptability index; is the plant density; is the travel speed; 、 、 、 They represent the efficiency weight coefficient, speed penalty coefficient, accuracy weight coefficient and acceleration compensation coefficient respectively; is the normalized mapping function of plant density; is the density-velocity interaction function; The system adaptability index comprehensively considers the two dimensions of control accuracy and operating efficiency. A larger value indicates better system performance. · represents the dot product.

[0083] Step 2.2, nonlinear mapping function determination. Determine the nonlinear mapping function in the model and : Normalize the plant density and map the density values ​​in different ranges to Interval, to facilitate subsequent processing. The specific form is:

[0084] ;

[0085] in, is the normalized mapping function of plant density; is the plant density; is the base of natural logarithms; is the adjustment parameter for controlling the slope of the function; is the base density value.

[0086] The interactive effect between plant density and travel speed is described using a quadratic function:

[0087] ;

[0088] in, is the density-velocity interaction function; is the plant density; is the travel speed; 、 、 represent the density linear influence coefficient, velocity linear influence coefficient and density-velocity quadratic interaction coefficient respectively.

[0089] Step 2.3, parameter optimization and calibration. Through machine learning methods, the model parameters are optimized based on actual operation data. 、 、 、 and 、 、 、 Optimization and calibration are performed to enable the model to accurately reflect the performance characteristics of the specific garlic harvester. The optimization uses a gradient descent method to update the parameters by minimizing the mean square error between the predicted speed and the actual optimal speed.

[0090] To verify the validity of the density-velocity coupled dynamics model, Figure 3 As shown in the figure, the analysis results of system adaptability index under different plant density and travel speed conditions are shown. It shows an obvious regular distribution under different density-speed combinations, proving that the density-speed coupling model can accurately quantify the system performance under different conditions and provide a reliable theoretical basis for solving the optimal speed.

[0091] Step 2.4, optimal speed solution algorithm. Based on the established density-speed coupling dynamic model, for any given plant density , solve the system adaptability index Optimal speed , as the optimal speed under the current density conditions. The solution process uses a numerical optimization method to find the optimal solution while satisfying the physical constraints of the garlic harvester.

[0092] Optimal speed The calculation process is:

[0093] Objective function setting: The system adaptability index defined in step 2.1 As the optimization goal, find the speed that maximizes its value .

[0094] Constraints: Set the physical constraints of the harvester, including the minimum and maximum allowable speed ranges and the maximum acceleration limit, to ensure that the solution meets the actual operation requirements.

[0095] Gradient solution: Use the nonlinear mapping function determined in step 2.2 and , for system adaptability indicators Find the partial derivative and get the velocity gradient information.

[0096] Iterative optimization: Use the gradient ascent method for iterative solution, starting from the initial speed value, and gradually adjust the speed value along the gradient direction until it converges to the optimal solution or reaches the maximum number of iterations.

[0097] Constraint processing: perform constraint checks on the speed values ​​obtained through iterative solution to ensure the final optimal speed within the permitted physical limits.

[0098] Through the above calculation process, the system can quickly calculate the corresponding optimal speed according to the real-time detected plant density. , achieving dynamic coordination between density and speed.

[0099] Step 3: Based on the output of the optimal speed and the density-speed coupling dynamics model, a compensation algorithm for the impact of speed changes on visual perception is developed. Adaptive image preprocessing and dynamic adjustment of feature extraction parameters are used to ensure stable recognition performance of the visual system. Based on the optimal speed command output from Step 2 and the current speed change trend, a compensation algorithm for the impact of speed changes on visual perception is developed. This algorithm compensates for visual perception and generates compensated visual perception information, ensuring that the visual system maintains stable recognition performance at different speeds.

[0100] Furthermore, the process of obtaining compensated visual perception information includes: establishing a speed image ambiguity mapping model based on the optimal travel speed and image quality; adaptively preprocessing the collected image based on the current travel speed and the speed image ambiguity mapping model to obtain an image quality assessment result; and dynamically adjusting the feature extraction parameters of the density estimation network based on the current travel speed and image quality assessment result to obtain compensated visual perception information. Specifically, the process includes the following sub-steps:

[0101] Step 3.1: Modeling the relationship between speed and image quality. Analyze the relationship between speed and image quality, establish a speed image ambiguity mapping model, and quantify the image change characteristics at different speeds.

[0102] The process of establishing the velocity image fuzziness mapping model is as follows: First, define the image fuzziness evaluation function :

[0103] ;

[0104] in, It is a comprehensive evaluation index of image blur; is the travel speed; is the velocity-related blur kernel standard deviation; is the image contrast index; is the texture degradation index; is the fuzziness weight coefficient; is the contrast weight coefficient; is the texture weight coefficient.

[0105] Secondly, establish the mapping relationship between speed and blur kernel standard deviation:

[0106] ;

[0107] in, is the velocity-related blur kernel standard deviation; is the travel speed; is the static fuzzy reference value; is the exposure time; is the focal length of the lens; is the angle between the motion direction and the imaging plane; is a sine function.

[0108] Model the relationship between speed and image contrast:

[0109] ;

[0110] in, is the image contrast index; is the travel speed; is the baseline contrast in static state; is the contrast attenuation coefficient; is the base of natural logarithms.

[0111] Model the relationship between speed and texture degradation:

[0112] ;

[0113] in, is the texture degradation index; is the travel speed; is the texture degradation parameter; is the texture degradation index parameter. Through controlled variable experiments, we collected image samples at different speeds, analyzed the changes in image clarity, contrast, texture characteristics and other parameters with speed, and constructed a prediction model.

[0114] Step 3.2: Adaptive image preprocessing. Based on the current speed and the speed-image relationship model established in Step 3.1, adaptive preprocessing is performed on the captured images. This includes dynamically adjusting the exposure time, applying a motion blur compensation filter, and adaptive contrast enhancement to improve image quality. The preprocessing parameters are automatically determined by the speed-image relationship model, enabling dynamic optimization of image quality.

[0115] Step 3.3, dynamic adjustment of the field of view. According to the speed command output in step 2, the field of view and focus of the visual system are dynamically adjusted to expand the front detection distance at high speed and enhance the close-range detail analysis at low speed. It is defined as high-speed travel mode, at this time the front detection distance is expanded to 8 to 12 meters, the field of view angle is adjusted to 45 to 60 degrees, and the focus is on long-distance path planning; when the travel speed is When the vehicle is moving at a medium speed, the front detection distance is set to 4 to 8 meters, the field of view angle is 30 to 45 degrees, and the distance information is balanced; when the vehicle is moving at a medium speed, the vehicle is in a medium speed mode. Defined as low-speed travel mode, this enhances close-range detail analysis, focusing on a detection range of 2 to 4 meters and a field of view of 15 to 30 degrees, improving local precision control capabilities. Dynamic adjustment of the field of view is achieved through adjustable-focus cameras or multi-camera arrays, ensuring optimal visual information at varying speeds.

[0116] Step 3.4: Adaptive adjustment of feature extraction parameters. Based on the current speed and the image quality assessment results from step 3.2, the density estimation network's feature extraction parameters, including convolution kernel size, pooling strategy, and attention mechanism weights, are dynamically adjusted to adapt to image characteristics at different speeds. A feedback mechanism is used to continuously evaluate and optimize parameter configurations to achieve stable feature extraction.

[0117] like Figure 4 As shown in the figure, the impact of speed changes on visual perception performance and the effect of the compensation mechanism are demonstrated. Without compensation, the image quality decreases significantly with increasing speed. After adopting the compensation mechanism of this technical solution, the image quality remains stable in a wider speed range, proving the effectiveness of compensation measures such as adaptive image preprocessing, dynamic adjustment of field of view, and adaptive adjustment of feature extraction parameters.

[0118] Step 4: Based on the output of the compensation algorithm (i.e., the compensated visual perception information and the output of the optimal travel speed), a multi-timescale predictive control framework is constructed. This framework also generates a control strategy to coordinate short-term precision control with long-term path planning. Based on the optimal speed sequence output from step 2 and the optimized visual perception information from step 3, the multi-timescale predictive control framework is introduced to coordinate short-term precision control with long-term path planning, enabling the garlic harvester to operate smoothly and efficiently in complex environments. This includes the following sub-steps.

[0119] Step 4.1: Control Domain Division. The control problem is divided into a millisecond-level short-term precision control domain and a second-level long-term planning control domain, corresponding to short-range row deviation correction and long-range path planning, respectively. The short-term control domain primarily handles immediate deviation correction, ensuring the harvester accurately tracks the garlic row at its current location. The long-term planning control domain focuses on the path ahead and changes in plant density, preparing for future speed and direction adjustments.

[0120] The control domain is divided into three areas according to the spatial domain. The specific implementation includes: near-field area (0 to 2 meters): responsible for immediate and precise control, with a response time of less than 50 milliseconds; mid-field area (2 to 5 meters): responsible for medium-term path planning, with a response time of 50 to 500 milliseconds; far-field area (5 to 10 meters): responsible for long-term strategy planning, with a response time of 0.5 to 3 seconds.

[0121] The control domain is divided into time domains. The specific implementations based on the control delay and prediction window include: short-term time domain: 0 to 200 milliseconds, corresponding to current state feedback control; medium-term time domain: 200 milliseconds to 2 seconds, corresponding to forward-looking path planning; long-term time domain: 2 seconds to 10 seconds, corresponding to long-term speed planning.

[0122] Control weight allocation: Control instructions in different time and space domains are dynamically weighted according to the current state. For example, the weight of far-field control is increased in high-speed mode, and the weight of near-field control is increased in high-precision mode.

[0123] Step 4.2: Hierarchical prediction model construction. Prediction models at different time scales are constructed, including: a short-term model: Based on the current state and control input, it predicts system state changes within 50 to 200 milliseconds for immediate and precise control; a medium-term model: Based on the current trajectory and plant density distribution, it predicts the optimal path and speed adjustment within 0.5 to 2 seconds; and a long-term model: Using the global density distribution map, it plans the optimal travel strategy within 3 to 10 seconds.

[0124] The prediction model adopts the model predictive control framework and continuously updates the control strategy through rolling optimization.

[0125] The specific structure of the hierarchical prediction model is as follows:

[0126] (1) Short-term prediction model. Dynamic model: A simplified vehicle kinematic model is used, including state variables such as position, heading angle, velocity, and steering angle; sampling frequency: 50 Hz (updated every 20 milliseconds); prediction window: 10 frames (predicting the next 200 milliseconds).

[0127] Objective function:

[0128] ;

[0129] in, is the short-term prediction objective function; is the number of steps in the short-term prediction window; represents the summation operator; is the lateral deviation weight coefficient; For the lateral deviation of the step; is the heading deviation weight coefficient; For the Step heading deviation; is the steering smoothing weight coefficient; For the The steering angle change per step.

[0130] Constraints:

[0131] ;

[0132] ;

[0133] ;

[0134] in, is the maximum steering angle limit; The maximum steering angle change limit; The maximum travel speed limit; 、 、 Represent the absolute values ​​of steering angle, steering angle change and travel speed respectively;

[0135] (2) Medium-term prediction model. Dynamic model: Combines vehicle kinematics and a simplified dynamic model, taking into account sideslip and acceleration limits; Sampling frequency: 10 Hz (updated every 100 milliseconds); Prediction window: 20 frames (predicting the next 2 seconds).

[0136] Objective function:

[0137] ;

[0138] in, is the medium-term forecast objective function; is the number of steps in the medium-term prediction window; represents the summation operator; is the trajectory accuracy weight coefficient; For the Step trajectory tracking error; is the path smoothing weight coefficient; For the the curvature of the step path; is the comfort weight coefficient; For the lateral acceleration of the step;

[0139] Constraints:

[0140] ;

[0141] ;

[0142] ;

[0143] in, is the maximum lateral acceleration limit; is the maximum path curvature limit; It is the minimum speed limit; The maximum travel speed limit; 、 、 represent the absolute values ​​of lateral acceleration, path curvature, and travel speed, respectively;

[0144] (3) Long-term prediction model. Dynamic model: a simplified point mass model focusing on velocity planning; sampling frequency: 1 Hz (update once per second); prediction window: 10 frames (predicting the next 10 seconds).

[0145] Objective function:

[0146] ;

[0147] in, is the long-term prediction objective function; is the number of steps for the long-term prediction window; represents the summation operator; For the System adaptability index of the step; For the Plant density of the step; For the The speed of the steps; is the speed smoothing weight coefficient; For the The speed of the step.

[0148] Constraints:

[0149] ;

[0150] ;

[0151] ;

[0152] in, It is the minimum speed limit; For the The speed of the steps; The maximum travel speed limit; For the Step and The absolute value of the step's travel speed; For the The speed of the steps; is the maximum allowable speed change; represents the summation operator; is the number of steps for the long-term prediction window; For the The time interval between steps; The maximum allowed operating time.

[0153] The rolling optimization process of the prediction model includes: predicting the future state → solving the optimization problem → executing the first step in the optimal control sequence → obtaining the new state → re-predicting and optimizing, repeating the cycle to achieve closed-loop control.

[0154] Step 4.3: Multi-scale control strategy coordination. A multi-scale control strategy coordination mechanism is designed to ensure that control decisions at different time scales are coordinated and consistent. A hierarchical decision-making structure is employed. Long-term planning provides target trajectories and speed range constraints for medium-term control. Medium-term control provides reference paths and speed trends for short-term control. Short-term control then executes specific control instructions, achieving seamless hierarchical control.

[0155] Detailed implementation of the control strategy coordination mechanism:

[0156] Hierarchical information transmission mechanism: adopts a two-way information flow design, including top-down transmission, medium-term to short-term transmission, and bottom-up feedback. Among them, the top-down transmission process includes long-term planning to set the speed target range and the general path direction to the medium-term controller; the medium-term to short-term transmission process includes the medium-term controller generating a reference path and providing it to the short-term controller for tracking; the bottom-up feedback process includes the actual execution effect of the short-term controller being fed back to the upper-level controller for adjustment and optimization.

[0157] Constraint hardening mechanism: As the control level decreases, the constraints gradually harden. From soft constraints in long-term planning, soft constraints are converted into more specific hard constraints in medium-term control, and finally precise control is performed in the short-term control layer. Among them, the target speed and path direction in the soft constraints in long-term planning can be adjusted appropriately according to the situation.

[0158] Buffer management: The system sets up control buffers between controllers at each layer, including a speed buffer that stores the speed trajectory generated by long-term planning for reference by the medium-term controller, and a path buffer that stores the path points generated by medium-term planning for tracking by the short-term controller, to ensure smooth transition of control signals.

[0159] Weight allocation algorithm: Dynamically assigns weights to controllers at each level based on the current operation status, including plant density weighting, speed weighting, and environmental weighting. Plant density weighting includes increasing the weight of short-term controllers in high-density areas to ensure accurate row tracking, and increasing the weight of long-term controllers in low-density areas to improve operation efficiency. Speed ​​weighting includes increasing the weight of medium- and long-term controllers to avoid sudden changes in direction during high-speed operations, and increasing the weight of short-term controllers to improve tracking accuracy during low-speed operations. Environmental weighting includes automatically increasing the weight of short-term controllers to ensure operation safety in complex terrain or in adverse weather conditions.

[0160] Conflict Resolution Strategy: When controllers at different levels generate conflicting control commands, a hierarchical arbitration mechanism is employed, including safety priority, precision priority, efficiency balance, and smooth transition. Safety priority prioritizes safety constraints, with any potentially dangerous control commands immediately rejected. Precision priority prioritizes tracking accuracy while ensuring safety, prioritizing precise short-term controller commands over coarse long-term commands. Efficiency balance selects the control strategy that improves overall operational efficiency while ensuring accuracy requirements are met. Smooth transition involves adopting a gradual transition approach to avoid sudden changes in control commands when switching control strategies.

[0161] Adaptive coordination mechanism: The system automatically adjusts coordination strategies based on actual operational results, continuously optimizing system performance through performance monitoring, parameter tuning, and strategy learning. Performance monitoring involves real-time monitoring of the execution performance of controllers at each layer, including metrics such as tracking accuracy, response speed, and stability. Parameter tuning involves automatically adjusting weight distribution parameters and conflict resolution thresholds based on performance monitoring results. Strategy learning involves using machine learning methods to learn the optimal coordination strategy from historical operational data and continuously update coordination algorithm parameters.

[0162] Step 4.4: Control Smooth Transition and Constraint Handling. Implement a control smooth transition and constraint handling mechanism to ensure the continuity and physical feasibility of control instructions. Control smooth transition involves designing a smoothing function to handle sudden changes in the control variable, preventing the harvester from accelerating or turning suddenly. Constraint handling considers the physical limitations of the garlic harvester (such as maximum steering angle and maximum acceleration) to ensure that the generated control instructions are within the physically feasible range.

[0163] The specific implementation of controlling smooth transition and constraint handling includes:

[0164] Smooth function design: Speed ​​smoothing uses an S-shaped speed curve and achieves smooth acceleration and deceleration through jerk limitation. The speed change function is defined as:

[0165] ;

[0166] in, For the moment The instantaneous speed of is the initial velocity; is the target speed; is a natural constant; is the smoothing coefficient, which controls the steepness of the transition and is usually set to 3-5; is the current time variable; The midpoint of the transition time.

[0167] The jerk constraint is:

[0168] ;

[0169] in, is displacement About time The third derivative of is the acceleration; is the absolute value operator; The maximum jerk limit is usually set to 0.5m / s³.

[0170] Steering smoothing uses an exponential moving average filter to process steering instructions and reduce steering jitter. The filter function is:

[0171] ;

[0172] in, For the Smooth steering angle of the step; is the discrete time step index; is the smoothing factor, which controls the filtering strength; For the The original steering command of the step; smoothing factor The value range is 0.1-0.3, the higher the density The smaller.

[0173] Initial conditions:

[0174] ;

[0175] in, is the smooth steering angle at the initial moment; is the original steering instruction at the initial moment;

[0176] Physical constraint processing includes: (1) speed constraint: considering the maximum / minimum speed limit of the garlic harvester; (2) acceleration constraint: dynamically adjusting the maximum acceleration according to vehicle performance and ground conditions; (3) steering constraint: considering the maximum steering angle and steering rate to ensure that the steering instruction is within the feasible range; (4) obstacle avoidance constraint: ensuring that the control instruction will not cause the machine to collide with the plant or leave the garlic row; (5) constraint softening technology: for multiple constraints that are difficult to meet simultaneously, the relaxation variable and penalty function method are used to find the best compromise solution during the optimization process; (6) safety check module: all control instructions are verified by the safety check module before execution to ensure that the control instructions will not lead to dangerous situations.

[0177] like Figure 5The figure shows the distribution comparison of the garlic harvester row tracking deviation before and after the multi-time scale predictive control framework is adopted. After adopting this technical solution, the row tracking deviation is reduced and the distribution is more concentrated. The median deviation is reduced from ±5.2 cm of the traditional method to ±1.8 cm, which proves the effectiveness of the multi-time scale predictive control framework in improving control accuracy.

[0178] Step 5: Based on the density distribution map, the output of the compensation algorithm, the optimal speed output, and the multi-timescale predictive control framework, a self-learning optimization system is established to optimize the density-speed mapping relationship by continuously collecting operational data. Based on the output data of the complete control process from Steps 1 to 4, a self-learning optimization system is established. By continuously collecting actual operational data, the density-speed mapping relationship is continuously optimized, gradually improving system performance over time. Continuously optimizing the density-speed coupling dynamic model and control strategy based on actual operational data to achieve automatic row control of the garlic harvester involves the following sub-steps.

[0179] Step 5.1: Operation Data Collection and Labeling. During harvester operation, key operation data is automatically collected and stored, including the plant density distribution generated in Step 1, the optimal speed calculated in Step 2, the visual compensation effect in Step 3, the execution of control instructions in Step 4, as well as information such as row alignment accuracy and environmental conditions, forming a complete operation data set. The system automatically labels row alignment performance indicators for different density and speed combinations, providing training data for subsequent model optimization.

[0180] The operation data acquisition system consists of a multi-layer architecture, including a data acquisition layer, a data preprocessing layer, an automatic labeling system, and a data management system. The data acquisition layer integrates multi-source data, including visual system images, GPS positioning, IMU attitude, steering angle sensors, and speed sensors. The visual data sampling frequency is 10Hz, and other sensor data is 50Hz. Data integrity is ensured through a combination of distributed caching and persistent storage. The data preprocessing layer unifies timestamps from different data sources, identifies and tags abnormal data points, and performs lossy compression for large-capacity data such as images. The automatic labeling system automatically calculates performance indicators such as row deviation mean, standard deviation, speed stability, and energy consumption, classifies data according to environmental conditions such as lighting, weather, and terrain, and assigns a quality score to each set of data that influences subsequent learning weights. The data management system supports incremental additions to the database, implements a hierarchical storage strategy for long-term preservation of core data and regular cleanup of auxiliary data, and provides an interface that supports efficient time series and semantic queries.

[0181] Step 5.2: Model Adaptive Update. Based on collected actual operational data, the parameters of the density-velocity coupled dynamics model are regularly updated to more accurately reflect the optimal operating state of the harvester in a specific field. This model update utilizes an incremental learning approach, assigning greater weight to new data to ensure the model can adapt promptly to environmental changes.

[0182] The model adaptive update system utilizes a comprehensive technical framework comprised of three core components: an incremental learning framework, a parameter update strategy, and an adaptive adjustment mechanism. The incremental learning framework automatically adjusts the update frequency based on job duration, employing a learning rate adjustment mechanism that gradually decreases as system stability improves. L2 regularization is used to prevent overfitting and preserve model generalization. The parameter update strategy utilizes a windowed weighting approach, whereby recent data is given a higher weight, while the weight of historical data decays exponentially over time. This strategy incorporates batch update technology and a validation mechanism to ensure update quality. The adaptive adjustment mechanism dynamically optimizes the system through environmental awareness and a performance monitoring system. The automatic adjustment of the update frequency is typically triggered every 1-2 hours of job completion or when the accumulated new data exceeds a threshold. The batch update technology groups parameters and employs an alternating update strategy to ensure system stability. The validation mechanism verifies the updated model against a historical "golden dataset," rolling back updates if performance degrades. Environmental awareness detects environmental changes, such as weather and terrain, and triggers updates to the corresponding parameter groups. The performance monitoring system continuously monitors system performance and triggers emergency updates when performance indicators drop below a threshold.

[0183] Step 5.3: Control strategy evolution. Using reinforcement learning techniques, the control strategy is continuously optimized and evolved. The system learns from successful control decisions and avoids repeating failed strategies. Real-time feedback obtained during the harvest process (such as row tracking deviation, speed changes, and plant damage rate) is used as a reward signal for reinforcement learning to guide the optimization of the control strategy.

[0184] The control strategy evolution framework adopts a comprehensive technical architecture, which includes four key components that work together with each other, namely the policy representation system, the reinforcement learning algorithm system, the exploration and utilization balance mechanism, and the safety constraint learning framework. Specifically, the policy representation system expresses the control strategy through parameterized functions (such as deep neural networks), defines the state space containing key variables such as plant density, speed, row deviation and path curvature, and the action space including target speed adjustment and steering angle adjustment; the reinforcement learning algorithm system combines the TRPO (Trust Region Policy Optimization) algorithm to ensure the stability of policy updates, adopts the experience replay mechanism to improve sample utilization efficiency, and converts long-term goals (operation efficiency and accuracy) into immediately measurable reward signals through reward shaping technology; the exploration and utilization balance mechanism realizes local exploration through Gaussian noise, conducts targeted exploration based on historical data, and gradually reduces the exploration ratio as the system performance improves; the safety constraint learning framework clearly defines the safe operation boundaries of the control strategy, maximizes the reward while meeting safety constraints, and is equipped with a fast fallback strategy in the event of exploration failure to ensure that the system always remains safe and controllable during the optimization process.

[0185] Step 5.4: Knowledge base accumulation and sharing. Build an operational knowledge base to record the optimal control strategies and parameter configurations for different fields, different garlic varieties, and different weather conditions. This knowledge base supports cross-field and cross-season knowledge transfer, enabling the system to quickly adapt to new operating environments without having to start from scratch. In scenarios where multiple harvesters are working together, experience sharing between different machines is supported, accelerating the learning process for the entire fleet.

[0186] The knowledge base system utilizes a comprehensive technical architecture comprised of four interrelated core components: a knowledge representation structure, a knowledge transfer mechanism, a distributed knowledge sharing architecture, and a knowledge base maintenance mechanism. The knowledge representation structure constructs a complete knowledge system using scenario descriptors, policy parameter sets, performance metrics, and credibility scores. The knowledge transfer mechanism enables cross-scenario knowledge application through similarity calculation, parameter adaptation, and migration verification. The distributed knowledge sharing architecture, comprised of a central knowledge base, local knowledge caches, asynchronous updates, and conflict resolution, ensures efficient knowledge sharing. The knowledge base maintenance mechanism ensures high-quality operation through knowledge compression, knowledge elimination, and continuous verification. Among them, the scene descriptor includes environmental factors such as terrain characteristics, plant density distribution, and variety characteristics; the strategy parameter group includes the optimal control parameters corresponding to a specific scene; the performance index includes recording the actual performance of the parameter group in a specific scene; the credibility score includes evaluating the reliability of the parameter group based on data volume and consistency; the similarity calculation process includes calculating the environmental similarity based on the scene descriptor; parameter adaptation includes automatically adjusting parameters according to environmental differences; migration verification includes verifying the validity of the migrated knowledge in the new environment and recording adaptive adjustments; the central knowledge base includes storing the comprehensive experience of all harvesters; the local knowledge cache includes each harvester maintaining its own common scene knowledge; asynchronous update includes regular synchronization of local knowledge with the central knowledge base; conflict resolution includes selecting the optimal solution based on performance indicators and credibility scores; knowledge compression includes merging similar parameter settings to reduce redundancy; knowledge elimination includes removing knowledge items that have not been used for a long time and have poor performance; continuous verification includes regular verification of the validity of knowledge items and updating credibility scores.

[0187] like Figure 6 The results show the performance improvement trend of the self-learning optimization system. System performance continues to improve with extended usage, with rapid performance improvement in the initial operation, followed by stabilization and a slow upward trend. After 30 days of continuous use, the system's overall performance improved by approximately 18% compared to its initial state, verifying the effectiveness and continuous improvement capabilities of the self-learning optimization system and demonstrating the long-term benefits of the knowledge base accumulation and sharing mechanism.

[0188] A garlic harvester automatic row alignment control system based on machine vision, used to implement the above-mentioned garlic harvester automatic row alignment control method based on machine vision, comprising:

[0189] The density perception module is used to analyze image data through a deep learning network, estimate plant density characteristics in real time, and generate a density distribution map.

[0190] The density perception module is used to analyze the plant density characteristics of the current area and the front area in real time based on the image data obtained by machine vision using a deep learning network to generate a plant density distribution map.

[0191] The dynamic modeling module is used to construct a density-speed coupling dynamic model based on the plant density distribution map, and solve the optimal travel speed of the garlic harvester under the current plant density conditions through nonlinear mapping functions and optimization algorithms.

[0192] The visual compensation module is used to compensate for visual perception based on the optimal travel speed by using adaptive image preprocessing, dynamic adjustment of the field of view and dynamic adjustment of feature extraction parameters to obtain compensated visual perception information.

[0193] The multi-scale control module is used to construct a multi-time-scale predictive control framework based on the compensated visual perception information and the optimal travel speed, and obtain a control strategy based on the multi-time-scale predictive control framework.

[0194] The self-learning optimization module is used to continuously optimize the density-speed coupling dynamic model and control strategy based on actual operation data to achieve automatic row control of the garlic harvester.

[0195] Through the density-speed coupling dynamic model, the present invention enables the system to adjust the optimal travel speed in real time according to the plant density, reduce the speed to improve accuracy in high-density areas, and increase the speed to improve efficiency in low-density areas; through the compensation mechanism of speed change for visual perception, the present invention enables the system to maintain stable recognition performance under different lighting conditions, different plant densities and different speeds; through the multi-time scale predictive control framework, the present invention can simultaneously take into account short-term precision control and long-term path planning, thereby improving the accuracy and stability of row control; through the self-learning optimization system, the present invention enables the harvester to continuously learn and optimize from actual operations, and the system performance continues to improve with the use time; through the unified density-speed coordination framework, the present invention replaces the traditional multiple speed control and path detection strategies, reduces the complexity of the control system while improving the system adaptability, and reduces the difficulty of system maintenance and configuration.

[0196] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for automatic row control of a garlic harvester based on machine vision, characterized in that: The following steps are involved: Based on the image data obtained by machine vision, a deep learning network is used to analyze the plant density characteristics of the current area and the area ahead in real time to generate a plant density distribution map; A density-speed coupling dynamic model is constructed based on the plant density distribution map, and the optimal travel speed of the garlic harvester under the current plant density condition is solved by a nonlinear mapping function and an optimization algorithm; Based on the optimal travel speed, adaptive image preprocessing, dynamic adjustment of the field of view, and dynamic adjustment of feature extraction parameters are used to compensate for visual perception, thereby obtaining compensated visual perception information; constructing a multi-time-scale predictive control framework based on the compensated visual perception information and the optimal travel speed, and obtaining a control strategy based on the multi-time-scale predictive control framework; Based on actual operation data, the density-speed coupling dynamic model and control strategy are continuously optimized to achieve automatic row control of the garlic harvester.

2. The automatic row control method for garlic harvester based on machine vision according to claim 1, wherein Based on image data acquired by machine vision, a deep learning network is used to analyze plant density characteristics in the current area and the area ahead in real time. The process of generating a plant density distribution map includes the following: Extract features from the collected images based on a multi-scale convolutional neural network to generate multi-scale features; The multi-scale features are fused through the feature pyramid network, and the key area features are highlighted in combination with the attention mechanism to generate a preliminary density feature map; The preliminary density feature map is processed by a density regression network to generate a plant density distribution map within the current field of view.

3. The automatic row control method for garlic harvester based on machine vision according to claim 1, wherein The process of obtaining the optimal travel speed of the garlic harvester under the current plant density conditions includes: constructing a system adaptability index including an efficiency weight coefficient, a speed penalty coefficient, an accuracy weight coefficient, and an acceleration compensation coefficient based on the plant density distribution map, and determining a density-speed coupling dynamics model based on the system adaptability index; A nonlinear mapping function was used to normalize plant density, and the interaction between plant density and travel speed was quantified using a density-speed interaction function. Based on the interactive effect between plant density and travel speed, the gradient descent method is used to optimize and correct the parameters of the density-speed coupling dynamic model; Based on the optimized parameters, the numerical optimization method is used to solve the travel speed that optimizes the system adaptability index, which is used as the optimal travel speed under the current plant density conditions.

4. The automatic row control method for garlic harvester based on machine vision according to claim 3, wherein The expression of the system adaptability index is: ； Where, is the system adaptability index; is the plant density; is the travel speed; 、 、 、 They represent the efficiency weight coefficient, speed penalty coefficient, accuracy weight coefficient, and acceleration compensation coefficient respectively; is the normalized mapping function of plant density; is the density-velocity interaction function; is the acceleration, and · represents the dot product.

5. The automatic row control method for garlic harvester based on machine vision according to claim 4 is characterized in that, The expression for quantifying the interactive effect of plant density and travel speed through the density-speed interaction function is: ； Where, is the density-velocity interaction function; is the plant density; is the travel speed; 、 、 They represent the density linear influence coefficient, velocity linear influence coefficient, and density-velocity quadratic interaction coefficient, respectively. · represents the dot product.

6. The automatic row control method for garlic harvester based on machine vision according to claim 1, characterized in that: Based on the optimal travel speed, adaptive image preprocessing, dynamic adjustment of the field of view, and dynamic adjustment of feature extraction parameters are used to compensate for visual perception, and the process of obtaining compensated visual perception information includes: establishing a speed image ambiguity mapping model based on the optimal traveling speed and image quality; Adaptively preprocessing the collected image based on the current traveling speed and the speed image ambiguity mapping model to obtain an image quality assessment result; The feature extraction parameters of the density estimation network are dynamically adjusted based on the current travel speed and the image quality evaluation result to obtain compensated visual perception information.

7. The method for automatic row control of a garlic harvester based on machine vision according to claim 6, wherein: The process of establishing the velocity image ambiguity mapping model includes: Establish a mapping relationship between speed and blur kernel standard deviation; Model the relationship between speed and image contrast; Establish a model of the relationship between speed and texture degradation; A velocity image fuzziness mapping model is established based on the mapping relationship between velocity and blur kernel standard deviation, the relationship model between velocity and image contrast, and the relationship model between velocity and texture degradation.

8. The method for automatically controlling rows of a garlic harvester based on machine vision according to claim 7, wherein: The process of building a multi-timescale predictive control framework includes: The control domain is divided into short-term domain, medium-term domain and long-term domain according to the time domain; Constructing a hierarchical prediction model based on the short-term time domain, the medium-term time domain, and the long-term time domain; Obtaining control decisions at different time scales based on the hierarchical prediction model; The control decisions at different time scales are coordinated through a hierarchical decision structure and constraint hardening mechanism to obtain the final control instructions.

9. The automatic row control method for garlic harvester based on machine vision according to claim 8, characterized in that: The process of continuously optimizing the density-speed coupling dynamics model and control strategy based on actual operation data to achieve automatic row control of the garlic harvester includes: Collect and store plant density distribution, optimal travel speed, visual compensation parameters, and control execution effect data during the operation in real time to build an annotated operation dataset; An incremental learning method is used to update the parameters of the density-velocity coupled dynamics model, and new weights are assigned through a window weighting strategy. Based on the reinforcement learning framework, the multi-timescale predictive control framework is optimized using line tracking deviation and operation efficiency as reward signals; Based on the new weights and the optimized multi-time-scale predictive control framework, a knowledge base containing the characteristics of different operation scenarios is established. Based on the knowledge base containing the characteristics of different operation scenarios, adaptive migration of control parameters is achieved through similarity calculation, and optimization experience is shared during multi-machine collaborative operation.

10. A garlic harvester automatic row control system based on machine vision, characterized in that: For implementing the method according to claim 1, the system comprises: The density perception module is used to analyze the plant density characteristics of the current area and the area ahead in real time based on image data acquired by machine vision using a deep learning network to generate a plant density distribution map; A dynamic modeling module is used to construct a density-speed coupling dynamic model based on the plant density distribution map, and solve the optimal travel speed of the garlic harvester under the current plant density conditions through a nonlinear mapping function and an optimization algorithm; a visual compensation module for compensating visual perception based on the optimal travel speed by using adaptive image preprocessing, dynamic adjustment of the visual field, and dynamic adjustment of feature extraction parameters to obtain compensated visual perception information; a multi-scale control module, configured to construct a multi-time-scale predictive control framework based on the compensated visual perception information and the optimal travel speed, and obtain a control strategy based on the multi-time-scale predictive control framework; The self-learning optimization module is used to continuously optimize the density-speed coupling dynamic model and control strategy based on actual operation data to achieve automatic row control of the garlic harvester.

Images