Agricultural machine accurate operation parameter real-time adjusting system based on vision measurement

Abstract

The invention provides an agricultural machine accurate operation parameter real-time adjustment system based on visual measurement, and relates to the technical field of agricultural machine operation, and the system comprises the steps: collecting crop growth state visual data and environment related visual information; extracting crop key visual features and environment associated visual features; calculating the real-time adjustment quantity of the agricultural machinery operation parameters; analyzing the operation parameter matching degree and the adjustment stability, and detecting abnormality; generating a corresponding processing strategy; and optimizing the regulation and control coefficient and the preset operation parameter reference model. According to the invention, through deep fusion of multi-source vision measurement and an adaptive regulation and control algorithm, quantitative extraction of crop growth states and environment information is realized, and measurement errors in a complex farmland environment are effectively reduced; by constructing a PID regulation and control mechanism fusing visual feature confidence and a dynamic reference model, real-time adaptive adjustment of operation parameters is realized, adjustment response delay is shortened, parameter matching precision is improved, and the method can adapt to dynamic changes of different crop growth periods, soil and illumination conditions.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery operation technology, and in particular to a real-time adjustment system for precise agricultural machinery operation parameters based on visual measurement. Background Technology

[0002] With the accelerated development of large-scale and intensive agriculture, agricultural machinery operations have become a core link in ensuring agricultural production efficiency and quality. Significant differences exist in soil texture (e.g., sandy soil, clay soil), crop type (e.g., food crops, cash crops), and growth stage across different regions, posing differentiated adaptation requirements for core parameters such as depth, speed, and row spacing in agricultural machinery operations. Simultaneously, the iterative upgrading of modern agricultural equipment technology, especially the integration of sensor technology, automatic control technology, and agricultural machinery, has provided the technological foundation for precise perception and dynamic adjustment of operational parameters, driving the transformation of agricultural machinery operations from traditional experience-based adjustment to technology-driven precision adjustment.

[0003] The general agricultural machinery operation parameter adjustment usually adopts a fixed parameter control mode or a simple manual experience setting method. This not only results in a slow adjustment response but also low parameter matching accuracy, making it difficult to adapt to the dynamic changes of different crop growth stages, soil types, and light conditions, and thus poor operational adaptability. Summary of the Invention

[0004] This invention provides a real-time adjustment system for precision operation parameters of agricultural machinery based on vision measurement, which solves the defects of existing technologies such as slow adjustment response, low parameter matching accuracy and poor operation adaptability.

[0005] On one hand, this invention provides a real-time adjustment system for precision agricultural machinery operation parameters based on visual measurement, including: a crop information acquisition module, a feature extraction module, an adaptive control module, an anomaly detection and feedback module, an anomaly handling module, and a parameter optimization module. The crop information acquisition module is used to collect visual data on crop growth status and environmentally related visual information. The feature extraction module is used to extract key crop visual features and environmentally related visual features from the crop growth status visual data and environmentally related visual information, and performs preprocessing to output a core crop visual feature set, an environmental visual feature set, and feature confidence parameters. The adaptive control module is used to calculate the real-time adjustment amount of agricultural machinery operation parameters based on the core crop visual feature set, environmental visual feature set, and feature confidence parameters, using a PID control algorithm based on a preset operation parameter benchmark model, and outputs operation parameter adjustment instructions and adjusted parameter values. The anomaly detection and feedback module is used to analyze the matching degree and adjustment stability of operation parameters, detect anomalies, and output anomaly detection results by combining the operation parameter adjustment instructions, adjusted parameter values, and combined crop and environmental visual feature data. The anomaly handling module is used to generate corresponding processing strategies for different anomaly types based on the anomaly detection results, and outputs anomaly handling results and correction parameters. The parameter optimization module is used to optimize the control coefficients and preset operation parameter benchmark models based on the anomaly handling results and correction parameters, according to historical control data and long-term crop visual characteristic time series data, using the gradient descent algorithm, and outputs the optimization results.

[0006] According to the real-time adjustment system for precision agricultural machinery operation parameters based on visual measurement provided by the present invention, the feature extraction module includes a preliminary feature extraction unit, a feature preprocessing unit, and a feature selection and fusion unit. The preliminary feature extraction unit is used to separate the crop region and the environment region based on visual data of crop growth status and associated visual information of the environment using an image segmentation algorithm, and extracts initial crop visual features and initial environment visual features using a feature point detection algorithm, outputting an initial feature dataset. The feature preprocessing unit is used to denoise and normalize the initial feature dataset, correcting deviations in the feature extraction process, and outputting a standardized feature dataset. The feature selection and fusion unit is used to remove redundant features from the standardized feature dataset through feature importance evaluation, fuse effective features to generate a core crop visual feature set and an environment visual feature set, and simultaneously calculate the extraction confidence parameters of each feature, outputting the core crop visual feature set, the environment visual feature set, and the feature confidence parameters.

[0007] The real-time adjustment system for precision agricultural machinery operation parameters based on visual measurement provided by this invention includes an adaptive control module comprising a parameter verification unit, an adjustment amount calculation unit, and a control command generation unit. The parameter verification unit verifies the completeness and validity of the crop core visual feature set, the environmental visual feature set, and feature confidence parameters, eliminating invalid data. The adjustment amount calculation unit calculates the real-time adjustment amount of the agricultural machinery operation parameters based on a preset operation parameter benchmark model and control coefficients using a PID control algorithm, and outputs the operation parameter adjustment amount. The control command generation unit generates corresponding operation parameter adjustment commands based on the operation parameter adjustment amount and calculates the adjusted parameter values.

[0008] According to the real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by the present invention, the step of presetting the benchmark model for operation parameters in the adjustment calculation unit includes: Visual data on crop growth status and environmental-related visual information at different growth stages and under different environmental conditions of the target crop are collected, along with historical data on corresponding agricultural machinery operation parameters. The historical data on agricultural machinery operation parameters includes the optimal operation parameters that have been verified in practice.

[0009] Feature annotation is performed on visual data of crop growth status and visual information related to the environment to establish a dataset that maps visual features to optimal operating parameters.

[0010] The mapping dataset is preprocessed, and the job parameter data is standardized to obtain a standardized mapping dataset.

[0011] Based on the mapping relationship between visual features and task parameters, a benchmark model structure with multiple inputs and multiple outputs is constructed.

[0012] The standardized mapping dataset is divided into a training set and a validation set. The benchmark model structure is trained using the training set, and the model output accuracy is evaluated using the validation set. If the accuracy does not reach the preset accuracy threshold, the model structure parameters are adjusted and the model is retrained to output the benchmark model with the correct job parameters.

[0013] According to the real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by the present invention, the step of the adjustment calculation unit outputting the adjustment amount of the operation parameter includes: By inputting the crop core visual feature set, the environmental visual feature set, and the feature confidence parameters into the benchmark model of the operating parameters, the benchmark operating parameters corresponding to the current operating conditions are obtained.

[0014] Obtain the current actual operating parameters of the agricultural machinery and calculate the deviation between the actual operating parameters and the benchmark operating parameters. Then, apply a weighted correction to the deviation values ​​using the feature confidence parameter to obtain the weighted deviation value.

[0015] Based on the weighted deviation value, the PID control algorithm is used to calculate the initial adjustment amount of the operating parameters.

[0016] Based on the physical parameter limitations and operational safety thresholds of agricultural machinery, the initial adjustment of operational parameters is constrained and verified. If the initial adjustment exceeds the allowable range, it is corrected to the maximum or minimum value within the allowable range.

[0017] The adjusted values ​​of the operation parameters after constraint verification are output as the final adjustment values ​​to the control command generation unit.

[0018] The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by this invention includes an anomaly detection and feedback module comprising a data fusion unit, a threshold setting unit, and an anomaly analysis and judgment unit. The data fusion unit fuses the operation parameter adjustment instructions, the adjusted parameter values, and the combined visual feature data of the crop environment to output a fused dataset. The threshold setting unit sets an operation parameter matching threshold and an adjustment stability threshold based on the fused dataset. The anomaly analysis and judgment unit analyzes the matching degree between the adjusted parameter values ​​and the crop growth requirements, as well as the stability of the parameter adjustment process, based on the operation parameter matching threshold and the adjustment stability threshold, to determine whether any anomalies exist and outputs the anomaly detection result.

[0019] The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by the present invention includes an anomaly handling module comprising an anomaly type identification unit and a processing strategy generation unit. The anomaly type identification unit identifies the anomaly type based on anomaly detection results. The processing strategy generation unit matches preset processing schemes to different anomaly types, generating anomaly handling strategies and correction parameters.

[0020] According to the real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by the present invention, the step of the anomaly type identification unit identifying the anomaly type includes: Obtain core anomaly characterization data from the anomaly detection results, including the deviation between the adjusted parameter values ​​and crop growth requirements, the fluctuation range of parameter adjustment, the adjustment response delay duration, and the anomaly confidence of the combined visual feature data of crop environment.

[0021] Establish an anomaly type classification system, which includes parameter matching anomalies, regulation stability anomalies, feature data anomalies, and equipment association anomalies.

[0022] The core anomaly representation data is matched one by one with the anomaly type judgment conditions in the anomaly type classification system to determine the anomaly type corresponding to the core anomaly representation data.

[0023] If the core anomaly representation data matches multiple anomaly types simultaneously, the dominant anomaly type and associated anomaly types are determined by combining the anomaly occurrence sequence and data correlation in the anomaly detection results, and anomaly type identification results containing the dominant anomaly type, associated anomaly types, and anomaly representation details are generated.

[0024] The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided by the present invention includes the following steps in which the processing strategy generation unit generates anomaly handling strategies and correction parameters: Establish a mapping library between exception types and handling solutions, and preset corresponding handling priorities, core handling measures, and parameter correction rules for exception types.

[0025] Based on the dominant anomaly type and related anomaly types, the corresponding basic processing solutions are matched from the mapping relationship library. If related anomaly types exist, the processing measures corresponding to the related anomaly types are integrated to form an initial processing strategy.

[0026] The initial processing strategy is adaptively optimized by combining the parameters of the current operation scenario, eliminating processing measures that do not match the current scenario, and adding scenario adaptability adjustment items to obtain the optimized processing strategy.

[0027] Based on the optimization processing strategy and parameter correction rules, and combined with the quantitative indicators in the core anomaly characterization data, specific correction parameter values ​​are calculated. The correction parameters include adjustment algorithm parameter correction values, feature processing parameter correction values, and job parameter safety threshold correction values.

[0028] The validity of the corrected parameter values ​​is validated. If the validation passes, the optimized processing strategy and corrected parameters are output as the final result. If the validation fails, the processing strategy is re-optimized and the corrected parameter values ​​are adjusted.

[0029] The real-time adjustment system for precision agricultural machinery operation parameters based on visual measurement provided by this invention includes a parameter optimization module comprising an optimization data integration unit and a parameter optimization calculation unit. The optimization data integration unit retrieves historical control data and long-term crop visual characteristic time-series data stored in the system, organizes and integrates the data, and outputs an optimized dataset, based on anomaly handling strategies and correction parameters. The parameter optimization calculation unit iteratively optimizes the control coefficients and a preset operation parameter benchmark model using a gradient descent algorithm based on the optimized dataset, calculates the optimized control coefficients and operation parameter benchmark model parameters, and outputs the optimization results.

[0030] This invention provides a real-time adjustment system for precise agricultural machinery operation parameters based on visual measurement. Through the deep integration of multi-source visual measurement technology and adaptive control algorithms, it effectively solves the problems of traditional agricultural machinery operation parameter adjustment, such as reliance on human experience, slow response, and poor adaptability. It achieves precise quantitative extraction of crop growth status and environmental information, significantly improving feature extraction confidence and segmentation accuracy, providing reliable data support for operation parameter adjustment, and effectively reducing measurement errors in complex farmland environments. By constructing a PID control mechanism that integrates visual feature confidence and a dynamic benchmark model, real-time adaptive adjustment of operation parameters is achieved. Compared with traditional fixed parameter control, the adjustment response delay is shortened, parameter matching accuracy is improved, and it can adapt to dynamic changes in different crop growth stages, soil, and light conditions. By integrating visual data and operation parameter information through DS evidence theory, various anomalies such as parameter matching and adjustment stability can be quickly identified. Combined with scenario-adaptive processing strategies and parameter optimization mechanisms, the impact of abnormal operations on crop growth is reduced, thus lowering the operation failure rate. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the structure of a real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement provided in an embodiment of the present invention; Figure 2 This is a flowchart of the preset operation parameter benchmark model in the adjustment calculation unit provided in this embodiment of the invention; Figure 3 This is a flowchart of the process strategy generation unit generating an exception handling strategy and correction parameters provided in the embodiments of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0034] The following is combined Figures 1-3 This invention describes a vision-based measurement-based real-time adjustment system for precision agricultural machinery operation parameters.

[0035] like Figures 1-3 As shown in the figure, the real-time adjustment system for precision agricultural machinery operation parameters based on visual measurement provided in this embodiment of the invention includes: a crop information acquisition module, a feature extraction module, an adaptive control module, an anomaly detection and feedback module, an anomaly handling module, and a parameter optimization module. Visual measurement refers to the technology of acquiring visual data of a target scene through image acquisition equipment, and then analyzing and processing the data using computer vision algorithms to achieve quantitative measurement and state assessment of the characteristics of the target object. Its core advantages are non-contact measurement, strong real-time performance, and rich information dimensions, making it suitable for crop state monitoring in complex farmland environments.

[0036] The crop information acquisition module is used to collect visual data on crop growth status and related environmental visual information. It employs a multi-source image acquisition device combination scheme. The core equipment includes a high-definition RGB camera mounted on the top of the agricultural machinery cab, as well as auxiliary ambient light sensors and a GPS positioning module. The RGB camera collects intuitive growth status data such as crop morphology and color, the ambient light sensor simultaneously collects light intensity data, and the GPS module records the latitude and longitude information of the acquisition location to correlate with plot environmental differences. The acquisition sequence is one frame of image data every 0.5 seconds, simultaneously storing metadata such as image acquisition time, device parameters, and GPS coordinates. Data transmission uses a 5G industrial module to achieve real-time transmission of acquired data to the system's edge computing nodes, with transmission latency controlled within 100ms.

[0037] The feature extraction module is used to extract key crop visual features and environment-related visual features from crop growth status visual data and environmentally associated visual information, and performs preprocessing to output a core crop visual feature set, an environmental visual feature set, and feature confidence parameters. Key crop visual features represent core visual indicators of crop growth status, including crop height, canopy coverage, leaf texture complexity, and color mean. Environment-related visual features are visual indicators related to the crop growth environment, including soil bare area, weed coverage, and light reflection intensity. The feature confidence parameter is a quantitative indicator used to evaluate the reliability of the feature extraction results, with a value range of [0,1]. The closer the value is to 1, the more reliable the extraction result.

[0038] The feature extraction module includes a preliminary feature extraction unit, a feature preprocessing unit, and a feature filtering and fusion unit. The preliminary feature extraction unit, based on visual data of crop growth status and associated visual information of the environment, uses an image segmentation algorithm to separate the crop region from the environment region, extracts initial crop visual features and initial environment visual features through a feature point detection algorithm, and outputs an initial feature dataset. Specific steps include: a. Perform size unification and color space conversion on the input RGB and multispectral images.

[0039] b. A semantic segmentation algorithm based on U-Net is adopted. The preprocessed image is taken as input, and the pixel-level segmentation mask is output. The loss function of the segmentation algorithm adopts the cross-entropy loss function L that fuses visual measurement accuracy, and the formula is expressed as:

[0040] Where N is the total number of pixels in the image, y i p is the true label of the i-th pixel, with a label value of 0 or 1. i The probability that the pixel is a crop region is predicted by the model. λ is the IoU weight, which takes a value of 0.3 to 0.5 to enhance the visual integrity of the region segmentation. IoU(Y,P) is the intersection-union ratio of the real segmentation mask Y and the predicted mask P, which is directly related to the accuracy of the visual measurement of the crop region.

[0041] c. Using the SIFT (Scale Invariant Feature Transform) algorithm, feature point detection is performed on the segmented crop and environment regions respectively. The extracted initial crop visual features include the position coordinates (x, y) of the feature points, the scale factor σ (ranging from 1.6 to 3.2, adapted to the visual scale of the crop canopy), and the orientation angle θ. The initial environment visual features include the gray values ​​of texture feature points in the soil region and the spacing of contour feature points in the weed region. The specific steps include: c.1. Construct Gaussian difference pyramids for the segmented crop and environmental regions, dividing them into 6 groups, each containing 5 difference layers. Initial Gaussian smoothing σ0 = 1.6, the sampling interval between adjacent groups is 2, and the σ increment between adjacent layers within a group is... Ultimately, it covers the crop canopy adaptation scale range of σ=1.6~3.2.

[0042] c.2. For each pixel in the pyramid, compare it with its 8 neighboring pixels in the same layer, and the corresponding 9 pixels in the upper and lower layers, for a total of 26 pixels. If the pixel is a local maximum or minimum value, it is determined to be an extreme point. All extreme points that meet the conditions are retained in the crop area, while only extreme points with significant grayscale changes (i.e., grayscale difference ≥ 20) are retained in the environment area.

[0043] c.3. Perform 3D quadratic function fitting on the extreme points to correct the position (x,y) and scale σ of the key points, and remove points with positioning error > 0.5 pixels and contrast < 0.03; use the Hessian matrix to determine whether the key points are located on the edge, that is, calculate the matrix eigenvalues. If the ratio of eigenvalues ​​is > 10, it is determined to be an edge point and removed; output the corrected key points of the crop area and the effective key points of the environmental area.

[0044] c.4. Taking the keypoint as the center, take a 16×16 neighborhood window on the image at its corresponding scale, and calculate the gradient magnitude of each pixel in the window; use a Gaussian weighting function with σ=1.5 to weight the gradient magnitude; calculate the gradient direction histogram in the range of 0°~360°, and the direction corresponding to the peak of the histogram is the main direction θ of the keypoint; if there is a secondary peak, add an auxiliary direction for the keypoint to improve feature robustness.

[0045] c.5. Centered on the key points, rotate the image to align the main direction with the x-axis, take a 16×16 neighborhood and divide it into 4×4 sub-regions. Calculate the gradient histograms of each sub-region in 8 directions to generate a 128-dimensional feature vector. For key points in the soil region, extract the gray mean and variance of their 16×16 neighborhood as supplementary texture features. For key points in the weed region, calculate the distance between them and their three adjacent key points as the contour distance feature. Normalize all feature vectors to finally output the initial crop visual features and the initial environment visual features.

[0046] d. Integrate the extracted initial features according to the structure of "crop features - environmental features - collected metadata" to form an initial feature dataset.

[0047] The feature preprocessing unit is used to denoise and normalize the initial feature dataset, correct biases in the feature extraction process, and output a standardized feature dataset. Specific steps include: To address the salt-and-pepper noise in the initial features, a median filtering algorithm is used. For feature point coordinate deviations, Gaussian smoothing is employed. The kernel function of the Gaussian filter is expressed as:

[0048] Where G(x,y) is the weight value of the Gaussian kernel function at coordinates (x,y), used to weight pixels when performing Gaussian smoothing on the feature point coordinate deviation during feature preprocessing. σ is the standard deviation of the Gaussian kernel, ranging from 0.8 to 1.2, adaptively adjusted according to the image noise intensity. (x,y) represents the horizontal and vertical coordinate offsets of the current pixel relative to the feature point within the neighborhood centered on the target feature point, characterizing the spatial positional relationship between the pixel and the feature point.

[0049] A linear correction model is used to correct systematic biases in the feature extraction process. For the eigenvalue f, the correction formula is expressed as:

[0050] Among them, f corr The corrected feature value is represented by k, which is the correction coefficient ranging from 0.95 to 1.05. b is the correction offset. k and b are determined through standard sample calibration experiments, i.e., feature extraction is performed using crop sample images with known true feature values, and the results are obtained by least squares fitting.

[0051] Use the min-max normalization method to map all eigenvalues to the interval [0, 1], and the formula is expressed as:

[0052] where f norm is the normalized eigenvalue, f min is the minimum value of the feature in the training sample set, and f max is the maximum value of the feature in the training sample set.

[0053] Output the standardized feature data set. The data set format is a two-dimensional matrix, where the rows represent the sample numbers and the columns represent the feature dimensions.

[0054] The feature screening and fusion unit is used to eliminate redundant features in the standardized feature data set through feature importance evaluation, fuse the effective features, generate the crop core visual feature set and the environmental visual feature set, and calculate the extraction confidence parameter of each feature, and output the crop core visual feature set, the environmental visual feature set and the feature confidence parameter.

[0055] Use the random forest algorithm to calculate the importance score of each feature. The number of decision trees in the random forest is set to 100. Among them, the calculation method of the feature importance score S is: by calculating the average value of the reduction in the Gini coefficient of the feature at all decision tree nodes, and introducing the correlation weight ω between the visual feature and the operation parameter (ω is set according to the influence degree of the feature on the core operation parameters such as seeding depth and fertilization amount, and the value range is 0.6 - 1.0). The final score S' = S × ω, S' ∈ [0, 1].

[0056] Set the feature importance threshold T = 0.2, which can be adaptively adjusted according to the operation scenario, and eliminate the features with S' < T.

[0057] Use the weighted average fusion algorithm to fuse the effective features of the same category (such as the mean value of the red channel and the mean value of the green channel in the crop color feature). The fusion formula is expressed as:

[0058] where m is the number of features of the same category, a is the feature index, and w a is the weight of the a-th feature, satisfying and w a is determined according to the feature importance score that is [[ID= forty-three]]to ensure that the fused features preferentially reflect the visual information strongly related to the operation parameters. is the normalized eigenvalue of the a-th feature.

[0059] Confidence C bDetermined by the end-to-end accuracy index of vision measurement, the calculation formula is expressed as:

[0060] Where b is the feature index, R b P represents the success rate of feature extraction, which is the proportion of samples in historical data where this feature was successfully extracted out of the total number of samples. b The accuracy of crop region segmentation refers to the segmentation accuracy of the region containing this feature, and is a core visual measurement indicator. b Image sharpness is scored, ranging from 0 to 1, and is obtained by calibrating the camera's focal length and exposure parameters. It directly affects the reliability of visual measurements. b This represents the noise level of the feature, i.e., the normalized noise variance.

[0061] Output the core visual feature set of crops (including 5-8 core features such as plant height and canopy coverage), the environmental visual feature set (including 3-5 core features such as soil bare area and light reflection intensity), and the corresponding feature confidence parameter set.

[0062] The adaptive control module calculates real-time adjustments to agricultural machinery operating parameters based on a preset operating parameter benchmark model, using a PID control algorithm, and employs the crop's core visual feature set, environmental visual feature set, and feature confidence parameters. It then outputs the operating parameter adjustment commands and the adjusted parameter values. Agricultural machinery operating parameters include sowing depth, fertilizer application rate, pesticide application rate, and operating speed. The operating parameter benchmark model is a mapping model between visual features and optimal operating parameters built from historical data, used to output the theoretically optimal operating parameters under different working conditions.

[0063] The adaptive control module includes a parameter verification unit, a control amount calculation unit, and a control command generation unit. The parameter verification unit verifies the completeness and validity of the crop core visual feature set, the environmental visual feature set, and the feature confidence parameters, eliminating invalid data. Specific steps include: Check whether the core feature set contains all the necessary feature dimensions (e.g., the core visual feature set of crops must include plant height and canopy coverage). If the number of missing dimensions is ≥2, it is considered invalid data; if the number of missing dimensions is 1, the mean of the feature from adjacent samples in the same plot is used to complete it.

[0064] The 3σ criterion is used to determine whether each eigenvalue is an outlier. For an eigenvalue f, if f < μ f -3σ f or f>μ f +3σ f (where μ) f σ is the mean of the features. f If the standard deviation of the feature is 0, it is considered an outlier; at the same time, if the feature confidence parameter C...b If the value is less than 0.5, the corresponding eigenvalue is considered invalid.

[0065] If the number of invalid feature values ​​in a single sample is ≥3, the sample is removed and the feature data of the valid sample in the previous frame is used as a temporary substitute; if the number of invalid feature values ​​is ≤2, the invalid values ​​are filled in by linear interpolation, and the validated valid feature dataset and confidence parameter set are output.

[0066] The adjustment calculation unit is used to calculate the real-time adjustment of agricultural machinery operation parameters based on a preset operation parameter benchmark model and control coefficient, using a PID control algorithm, and outputs the operation parameter adjustment.

[0067] The control instruction generation unit is used to generate corresponding operation parameter adjustment instructions based on the adjustment amount of operation parameters and calculate the adjusted parameter values.

[0068] In the adjustment calculation unit, the steps for presetting the baseline model of the operating parameters include: Visual data on crop growth status and environmental correlation visual information at different growth stages and under different environmental conditions were collected, along with corresponding historical data on agricultural machinery operation parameters. The historical data on agricultural machinery operation parameters included the optimal operation parameters verified in practice. Data collection scope: Covering the entire growth period of the target crop, with different environmental conditions including different soil types, soil moisture, light conditions, and pest and disease levels. Field trials were conducted, selecting three experimental fields with an area of ​​≥10 mu (approximately 1.65 acres). Each experimental field was divided into 10 plots, with different combinations of operation parameters set for each plot. The crop visual data and final yield and quality indicators for each plot were recorded. The operation parameters corresponding to the highest yield and best quality were determined as the optimal operation parameters for that condition.

[0069] Feature annotation is performed on visual data of crop growth status and visual information related to the environment to establish a dataset that maps visual features to optimal operating parameters.

[0070] A combination of manual and automatic annotation is used. The real values ​​of key features such as crop height and canopy coverage are manually annotated, while derived features such as the number of feature points and texture complexity are automatically annotated.

[0071] Using the visual feature set as input and the optimal task parameter set as output, a mapping relationship is constructed. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio to ensure consistent distribution of reproductive periods and environmental conditions across subsets. The visual feature set X = [x1, x2, ..., x...]. n ], where n is the feature dimension; the optimal job parameter set Y=[y1,y2,...,y m ], where m is the dimension of the operation parameters, such as sowing depth and fertilizer application rate.

[0072] The mapping dataset is preprocessed, and the job parameter data is standardized to obtain a standardized mapping dataset. The visual feature set is processed using denoising and normalization methods in the feature preprocessing unit. The job parameters are standardized using the Z-score standardization method, expressed by the formula:

[0073] in, Here are the standardized operation parameter values, where c is the sample number and y is the value of the sample. c The original job parameter value, μ y Let σ be the mean of the parameters for this task across the dataset. y The standard deviation is given. Output the standardized mapping dataset {X}. norm,i ,Y norm,i}

[0074] Based on the mapping relationship between visual features and operational parameters, a multi-input, multi-output baseline model structure is constructed. A BP neural network structure incorporating a visual attention mechanism is adopted, i.e., a multi-layer feedforward neural network based on the error backpropagation algorithm. The input layer has k0 nodes, including visual feature dimensions such as plant height, canopy coverage, and exposed soil area. A visual attention layer is added, with the core parameter being the attention weight matrix W, with dimensions k0×k0, used to strengthen the weights of visual features strongly correlated with operational parameters. Two hidden layers are set: the first hidden layer has 2k0+1 nodes, and the second hidden layer has k0+1 nodes. The output layer has k1 nodes, representing operational parameter dimensions such as sowing depth, fertilizer application rate, and operation speed. The activation functions used are the ReLU function (hidden layer) and the Linear function (output layer). The ReLU function formula is: f(x) = max(0,x). The loss function uses the mean squared error (MSE) incorporating visual measurement errors, expressed as:

[0075]

[0076] Among them, L MSE Where N is the mean squared error and N0 is the sample size. Let j be the standardized true value of the job parameter of the c-th sample. γ is the standardized value of the j-th task parameter of the c-th sample predicted by the model, and γ is the visual confidence error weight, which ranges from 0.2 to 0.4. Let c be the true value of the average feature confidence score for the c-th sample. The average feature confidence score for the c-th sample predicted by the model ensures that the model training simultaneously fits the visual feature-task parameter mapping relationship and the reliability of visual measurements.

[0077] The update formula for the attention weight matrix W is expressed as:

[0078] in, For the updated attention weight matrix, For the c-th visual feature x c With the j-th job parameter y j The covariance is used to strengthen the weights of visual features with high correlation. n1 is the total number of dimensions of the input visual features. For example, when there are three features, such as plant height, canopy coverage, and bare soil area, n1=3. k2 is the traversal index of the visual feature dimensions, used to calculate the sum of the normalization terms.

[0079] The standardized mapping dataset is divided into a training set and a validation set. The benchmark model structure is trained using the training set, and the model output accuracy is evaluated using the validation set. If the accuracy does not reach the preset accuracy threshold, the model structure parameters are adjusted and the model is retrained to output the benchmark model with the correct job parameters.

[0080] The optimizer used is the Adam optimizer, with an initial learning rate of 0.001 and a learning rate decay strategy of decreasing to 0.9 every 100 epochs; the number of training epochs is set to 500, and the batch size is set to 32.

[0081] Using the coefficient of determination R 2 The mean absolute error (MAE) is used to evaluate model accuracy, and the formula is expressed as:

[0082]

[0083] Among them, R 2 The coefficient of determination, with a value ranging from -∞ to 1, indicates that the closer the coefficient is to 1, the higher the accuracy of the model's prediction of the job parameters. A preset threshold R0 is also provided. 2 ≥0.9. Let c be the standardized true value of the job parameter for the c-th sample. The standardized value of the job parameter for the c-th sample predicted by the model. The population mean of standardized job parameters for all samples. MAE is the mean absolute error, with a value ≥ 0. The smaller the value, the smaller the average deviation between the job parameters predicted by the model and the actual values. The preset threshold is MAE ≤ 0.05.

[0084] The steps for the adjustment calculation unit to output the adjustment amount of the operation parameter include: The crop core visual feature set, environmental visual feature set, and feature confidence parameters are input into the benchmark model for operational parameters to obtain the benchmark operational parameters corresponding to the current working conditions. Considering the influence of the feature confidence parameters, the input features are weighted, and the weighting formula is expressed as follows:

[0085] in, The weighted visual feature value is b. The influence of reliable visual features is strengthened by confidence weighting, while the interference of low-confidence features is weakened. This is the b-th feature value after standardization.

[0086] The weighted feature set is input into the benchmark model to obtain standardized benchmark operation parameters. Convert standardized baseline operating parameters into actual baseline operating parameters. The formula is expressed as .

[0087] The current actual operating parameters of the agricultural machinery are obtained by reading them through the agricultural machinery CAN bus. The reading frequency is consistent with the feature acquisition frequency. The parameter types include analog quantities (such as fertilizer application rate) and digital quantities (such as operating speed gear).

[0088] The deviation between the actual operating parameters and the benchmark operating parameters is calculated using the following formula:

[0089] Where e0 is the original deviation between the baseline operating parameters and the current actual operating parameters. A positive deviation indicates that the current parameters are lower than the optimal value, and a negative deviation indicates that the current parameters are higher than the optimal value. These are the current actual operating parameter values.

[0090] The bias value is weighted and corrected by combining the feature confidence parameter to obtain the weighted bias value, which is expressed by the formula:

[0091] Where e is the weighted correction value, which weakens the weight of the deviation value corresponding to low-confidence visual features and improves the reliability of deviation calculation.

[0092] Based on the weighted deviation value, the PID control algorithm is used to calculate the initial adjustment amount of the operating parameters, expressed by the formula:

[0093]

[0094] in, This refers to the initial operational parameter adjustments for the k3rd sampling period, such as the increase / decrease in sowing depth and the adjustment range for fertilizer application. K p This is a proportionality coefficient, set based on the crop's visual sensitivity, determining the strength of the deviation's response to the adjustment amount. For example, when canopy cover dominates, K... p Take a value of 1.2 to 1.8. K represents the weighted bias value for the k3th sampling period.i t4 is the integral coefficient used to eliminate the steady-state error of the system. t4 is the traversal index of the sampling period, and T1 is the sampling period, which is fixed at 0.5s. The integral weight is dynamically adjusted to represent the average feature confidence level of the t4th sampling period. d K is the differential coefficient used to predict the trend of deviation changes, improve the rapid response of regulation, and suppress overshoot. p K i K d The parameters are determined using a heuristic (Ziegler-Nichols) tuning method that incorporates visual measurement accuracy for the PID controller. The steps are as follows: Set Ki=0 and Kd=0, and select a farmland scene with stable visual characteristics, such as uniform crop growth and sufficient sunlight. Gradually increase Kp until the system exhibits constant-amplitude oscillations, and record the critical proportionality coefficient Kc and the critical oscillation period Tc at this point.

[0095] PID coefficients are calculated based on critical parameters and visual feature types: if the core visual features are morphological characteristics such as plant height and canopy coverage, K... p =0.6K c K i =2K p / T c K d =K p ·T c / 8; If it refers to physiological characteristics such as chlorophyll content, K p =0.5K c K i =1.5K p / T c K d =K p ·T c / 10.

[0096] To avoid saturation of the integral, for the integral term Set the amplitude limit range [-A,A]. A is determined based on the maximum allowable adjustment of the operation parameters and the fluctuation range of visual characteristics. For example, if the fertilizer application rate A = 5 kg / h, A is taken as 3 kg / h when the canopy coverage fluctuation is greater than 20%.

[0097] Based on the physical parameter limitations and operational safety thresholds of agricultural machinery, the initial adjustment amounts of operational parameters are constrained and verified. If the initial adjustment amounts exceed the allowable range, they are corrected to the maximum or minimum value within the allowable range. Physical parameter limitations are provided by the agricultural machinery manufacturer; for example, the adjustment range for sowing depth is 2-5 cm, and the adjustment range for fertilizer application rate is 10-50 kg / h. Operational safety thresholds are determined based on crop growth requirements; for example, the nitrogen application rate during the wheat tillering stage should not exceed 30 kg / mu.

[0098] The method for constraining and verifying the initial operation parameter adjustment is as follows: calculate the adjusted parameter value P. temp =P current +ΔP(k3), if P temp Within the constraints, ΔP(k3) is the effective adjustment; if P temp If the upper limit is reached, then the final adjustment amount of the operating parameter ΔP = upper limit - P current If P temp If the value is less than the lower limit, then ΔP = lower limit - P current .

[0099] The adjusted values ​​of the operation parameters after constraint verification are output as the final adjustment values ​​to the control command generation unit.

[0100] The anomaly detection and feedback module combines operational parameter adjustment instructions, adjusted parameter values, and crop-environment combined visual feature data to analyze the matching degree and adjustment stability of operational parameters, detect anomalies, and output anomaly detection results. Operational parameter matching degree refers to the degree of fit between the adjusted operational parameters and the crop's growth requirements, quantified by a matching degree score. Adjustment stability refers to the degree of fluctuation during the operational parameter adjustment process, expressed as the variance of the adjustment amount. Crop-environment combined visual feature data is a comprehensive feature set integrating core crop visual features and environmental visual features, containing information about the relationships between features.

[0101] The anomaly detection and feedback module includes a data fusion unit, a threshold setting unit, and an anomaly analysis and judgment unit. The data fusion unit is used to fuse operational parameter adjustment commands, adjusted parameter values, and crop environment combined visual feature data, outputting a fused dataset. Based on the timestamp, it merges adjustment commands and adjusted parameter values ​​P from the same acquisition time. new Joint visual feature set X combined Alignment is performed to ensure temporal consistency of the data. Data fusion is performed using the DS evidence theory, which prioritizes visual measurements. The adjusted parameter values ​​and the joint visual feature set are treated as two independent sources of evidence. The basic probability allocation function for each source of evidence is calculated, and the fusion formula is expressed as follows:

[0102]

[0103]

[0104] Where m(A) is the basic probability assignment of proposition A after fusion, with a value range of [0,1], representing the degree of trust in proposition A. K is the conflict coefficient, used to measure the degree of conflict between the adjusted parameters and visual features. The smaller K is, the smaller the conflict and the more reliable the fusion result. β is the weight of the visual feature evidence source, fixed at 0.6, highlighting the core role of visual measurement data in anomaly detection. This represents the basic probability assignment of proposition A to the joint visual feature set of crops and the environment. This represents the basic probability assignment of the adjusted parameter values ​​to proposition A. A1 is an element in the set of propositions with adjusted parameters, and A2 is an element in the set of propositions with visual features. A1∩A2=A indicates that the intersection of the two propositions is the target proposition A, such as "parameter matching is normal" and "parameter matching is abnormal".

[0105] The threshold setting unit is used to set the job parameter matching threshold and adjust the stability threshold based on the fused dataset. The job parameter matching threshold is set by using a statistical method to calculate the mean matching score μ of the "normal parameter matching" samples in the historical fused data. match and standard deviation σ match Set the matching threshold T match =μ match -2σ match Current matching score <T match When this occurs, it is determined to be an abnormal parameter matching. The method for setting the regulation stability threshold is: calculate the variance σ of the historical regulation data. ΔP Set a stability threshold T stab =2σ ΔP When the variance of the current adjustment is greater than T stab When this occurs, it is determined to be an abnormality in regulatory stability.

[0106] μ is recalculated every 100 sets of new data collected. match σ match σ ΔP Update the threshold to ensure it adapts to changes in the working environment.

[0107] The anomaly analysis and judgment unit is used to analyze the matching degree between the adjusted parameter values ​​and the crop growth requirements and the stability of the parameter adjustment process based on the matching threshold and adjustment stability threshold of the operation parameters, to determine whether there are any abnormalities, and to output the anomaly detection results.

[0108] Based on the basic probability distribution in the fused data, the matching score S is calculated. match =100×m, where m is the basic probability of the proposition "parameter matching is normal" after fusion. Calculate the variance σ of the adjustment amount over the most recent 5 sampling periods. current =var(ΔP(k3-4),ΔP(k3-3),...,ΔP(k3)). If S match <Tmatch And σ curren t≤T stab If S match ≥T match And σ current >T stab If S..., then it is determined to be an abnormality in regulatory stability. match <T match And σ current >T stab If S match ≥T match And σ current ≤T stab If so, it is considered normal.

[0109] Anomaly detection results include anomaly type, anomaly occurrence time, matching score, stability variance, and anomaly confidence.

[0110] The anomaly handling module is used to generate corresponding handling strategies for different anomaly types based on the anomaly detection results, and output the anomaly handling results and correction parameters.

[0111] The anomaly handling module includes an anomaly type identification unit and a processing strategy generation unit. The anomaly type identification unit identifies the anomaly type based on the anomaly detection results. The processing strategy generation unit matches preset processing schemes to different anomaly types and generates anomaly handling strategies and correction parameters.

[0112] The steps for the exception type identification unit to identify exception types include: The core anomaly representation data in the anomaly detection results are obtained, including the matching deviation between the adjusted parameter values ​​and crop growth requirements, the fluctuation range of parameter adjustment, the adjustment response delay duration, and the anomaly confidence level of the combined visual feature data of the crop environment. The matching deviation value and fluctuation range are calculated from the anomaly detection results, the adjustment response delay duration is extracted from the system log, and the anomaly confidence level of the combined visual feature data is output by the feature extraction module.

[0113] The matching deviation value is calculated as follows:

[0114] Where ΔS is the matching deviation value, and ΔS>0 indicates that there is a matching deviation.

[0115] The calculation method for the fluctuation range of parameter adjustment is as follows:

[0116] Among them, A ΔP The value of A represents the fluctuation range of the parameter adjustment; a larger value indicates a more unstable adjustment process. ΔP ≥2×Tstab T stab To adjust the stability threshold and the response delay time t delay If the time is ≤0.2s, it is determined to be an abnormal adjustment stability. max(ΔP(k3-4),...,ΔP(k3)) represents the maximum value of the operation parameter adjustment in the most recent 5 sampling periods. min(ΔP(k3-4),...,ΔP(k3)) represents the minimum value of the operation parameter adjustment in the most recent 5 sampling periods.

[0117] Adjusting the response delay time t delay The time difference between the completion of feature acquisition and the execution of adjustment commands by the agricultural machinery is calculated using system timestamps, and the calculation method is as follows:

[0118] Among them, t exec t represents the instruction execution time. collect This refers to the feature acquisition time.

[0119] The anomaly confidence of the crop environment joint visual feature data is the minimum value of the confidence parameters of each feature in the crop environment joint visual feature data, expressed as: ,in, C1, C2, ..., Cb represent the anomaly confidence scores of the combined visual feature data of crop environment, while C1, C2, ..., Cb represent the confidence scores of individual visual features.

[0120] An anomaly classification system is established, including parameter matching anomalies, regulation stability anomalies, characteristic data anomalies, and equipment correlation anomalies. The criterion for determining parameter matching anomalies is a matching deviation value ΔS ≥ 10, and C... abn The value ≥0.6 is primarily due to a significant discrepancy between the baseline operating parameters output by the benchmark model and the actual requirements. The criterion for determining abnormal stability is the fluctuation amplitude A. ΔP ≥2×T stab , and t delay ≤0.2s, the core reason is an unreasonable PID control coefficient or external interference with operating parameters (such as uneven ground). The criterion for judging abnormal characteristic data is C. abn If the value is less than 0.5 and ΔS < 10, the core reason is that the feature acquisition equipment malfunctions or harsh field conditions lead to unreliable feature extraction. The criterion for judging abnormal equipment association is t. delay The failure rate is >0.2s and the failure rate of adjustment command execution is ≥30%. The core reason is the failure of the agricultural machinery actuator or the communication module.

[0121] The core anomaly representation data is matched one by one with the anomaly type judgment conditions in the anomaly type classification system to determine the anomaly type corresponding to the core anomaly representation data.

[0122] If the core anomaly representation data matches multiple anomaly types simultaneously, the dominant anomaly type and associated anomaly types are determined by combining the anomaly occurrence sequence and data correlation in the anomaly detection results. This generates an anomaly type identification result containing the dominant anomaly type, associated anomaly types, and anomaly representation details. The determination rule is: the anomaly type that occurs first is determined as the dominant anomaly type, and the one that occurs later is the associated anomaly type. The correlation between each anomaly type and the anomaly representation data is calculated, and the anomaly type with the larger absolute correlation value is determined as the dominant anomaly type. The steps are as follows: Let variable X be a quantitative representation vector of a certain anomaly type, taking the value 1 when the anomaly type is met and 0 when it is not. Variable Y is a quantitative vector of core anomaly representation data: the matching deviation value ΔS is selected, and the fluctuation amplitude A is adjusted by parameter. ΔP Adjusting the response delay time t delay Confidence level C of joint visual feature anomalies abn The corresponding quantitative indicators in the data retain their values ​​at the original scale.

[0123] Select the X and Y data from the most recent U consecutive sampling periods (U≥5, consistent with the number of periods for stability calculation) to construct pairs of data samples {(X1,Y1),(X2,Y2),...,(X U ,Y U )}.

[0124] Calculate the mean of vector X respectively. The mean of the Y vector ,in For the first Within each sampling period, the quantized representation value of a certain anomaly type, For the first The quantized values ​​of the core anomaly characterization data within each sampling period. The formula for calculating the covariance is expressed as:

[0125] Here, cov(X,Y) is the covariance, a statistic that measures the degree of linear correlation between two variable vectors X and Y, reflecting the consistency of their changing trends. If cov(X,Y)>0, X and Y are positively correlated; if cov(X,Y)=0, X and Y are not linearly correlated; if cov(X,Y)<0, X and Y are negatively correlated.

[0126] Calculate the standard deviation of vector X respectively Standard deviation of the Y vector .

[0127] Finally, the Pearson correlation coefficient is calculated, expressed by the formula:

[0128] Where r is the Pearson correlation coefficient, with a value range of [-1, 1]. The closer the absolute value is to 1, the higher the correlation between the anomaly type and the anomaly representation data.

[0129] Example: If the core data matches both parameter matching anomalies and regulation stability anomalies, and the parameter matching anomaly occurs first with a correlation of 0.8, and the regulation stability anomaly occurs later with a correlation of 0.5, then the dominant anomaly type is parameter matching anomaly, and the associated anomaly type is regulation stability anomaly.

[0130] The steps for the processing strategy generation unit to generate exception handling strategies and correction parameters include: Establish a mapping library between exception types and handling solutions, pre-setting corresponding handling priorities, core handling measures, and parameter correction rules for each exception type. The mapping library is defined as follows: The priority for parameter mismatch is 1, which is the highest. The core handling measure is to suspend the current adjustment strategy, temporarily use manually calibrated operating parameters, re-collect sample data of the current working conditions to supplement the baseline model training set, and retrain the model. The parameter correction rule is to correct the output weights of the baseline model, with a correction coefficient α = 1 - ΔS / 100.

[0131] The priority for handling stability anomalies is 2, and the core solution is to adjust the PID control coefficient, typically K. p K i Reduce by 20%, K d Increase by 10% to enable anti-interference filtering. The parameter adjustment rule is K. p =0.8×K p K i =0.8×K i K d =1.1×K d .

[0132] The priority for abnormal feature data is 3. The core handling measures are to restart the feature acquisition device, clear the camera lens, extend the feature acquisition interval to 1 second, and temporarily use historical valid feature data. The parameter correction rules are to lower the feature confidence threshold to 0.4 and use the neighborhood mean method when completing missing features.

[0133] The priority of the equipment association anomaly is 4. The core handling measures are to issue an equipment fault alarm, suspend agricultural machinery operation, record the fault code, and upload it to the remote monitoring platform. The parameter correction rules are to adjust the communication baud rate to 19200bps, add a command retransmission mechanism, and retransmit 3 times.

[0134] Based on the dominant anomaly type and related anomaly types, corresponding basic processing solutions are matched from the mapping database. If related anomaly types exist, the processing measures corresponding to the related anomaly types are merged to form an initial processing strategy. The fusion rule is as follows: all processing measures for the dominant anomaly type are retained, processing measures from the related anomaly types that do not conflict with the dominant anomaly type are selected, and conflicting measures are eliminated.

[0135] The initial treatment strategy is adaptively optimized based on the parameters of the current operating scenario. Treatment measures that are incompatible with the current scenario are eliminated, and scenario adaptability adjustment items are added to obtain the optimized treatment strategy. Scenario parameters include crop growth stage, soil type, and weather conditions.

[0136] Example: If the current scenario is during the wheat grain-filling stage, which is a sensitive growth period, it is not advisable to frequently adjust the operation parameters. The dominant anomaly type is adjustment stability anomaly. The initial handling strategy is to adjust the PID coefficient + anti-interference filtering. After optimization, the PID coefficient adjustment measures are removed, only the anti-interference filtering is retained, and the adaptation item "reducing the operation speed by 5%" is added. If the current scenario is a sunny day with sufficient light, the initial handling strategy for feature data anomalies is to restart the equipment + clean the lens. After optimization, the adaptation item "adjusting the camera exposure parameters (shortening the exposure time by 20%)" is added.

[0137] Based on the optimization processing strategy and parameter correction rules, and combined with the quantitative indicators in the core anomaly characterization data, specific correction parameter values ​​are calculated. The correction parameters include adjustment algorithm parameter correction values, feature processing parameter correction values, and job parameter safety threshold correction values.

[0138] Example: If the dominant anomaly type is a regulatory stability anomaly, the fluctuation amplitude A ΔP =3×T stab The correction factor is K. p =0.8×K p K i =0.7×K i The greater the fluctuation range, the more the integral coefficient is reduced, K d =1.2×K d If the feature data is abnormal, C abn =0.3, the corrected confidence threshold T'=0.4-(0.5-0.3)×0.5=0.3; if the parameter matching is abnormal, ΔS=15, the corrected upper limit of the safety threshold=original upper limit×(1-ΔS / 200)=original upper limit×0.925.

[0139] The corrected parameter values ​​are validated to ensure they meet the physical limitations of agricultural machinery and the requirements for crop growth operations. If the validation passes, the optimized processing strategy and corrected parameters are output as the final result. If the validation fails, the processing strategy is re-optimized and the corrected parameter values ​​are adjusted. The corrected PID coefficients, communication baud rate, and other parameters must be within the allowable range of the agricultural machinery; the safety thresholds of the corrected operating parameters must meet the agronomic requirements of each growth stage of the target crop; if the corrected parameters exceed equipment limits, the correction range is reduced; if they do not meet agronomic requirements, a new processing measure for the associated anomaly type is selected, and the corrected parameters are recalculated.

[0140] The parameter optimization module is used to optimize the control coefficients and preset operation parameter benchmark models based on the anomaly handling results and correction parameters, according to historical control data and long-term crop visual characteristic time series data, using the gradient descent algorithm, and outputs the optimization results.

[0141] The parameter optimization module includes an optimization data integration unit and a parameter optimization calculation unit. The optimization data integration unit is used to retrieve historical control data and long-term crop visual characteristic time-series data stored in the system based on anomaly handling strategies and correction parameters, organize and integrate the data, and output an optimized dataset.

[0142] The parameter optimization calculation unit is used to iteratively optimize the control coefficients and the preset operational parameter baseline model using the gradient descent algorithm based on the optimization dataset. It calculates the optimized control coefficients and operational parameter baseline model parameters, and outputs the optimization results. Specific steps include: The objective function J(θ) is the negative reciprocal of the assignment performance score, expressed as:

[0143] Where θ represents the parameters to be optimized, i.e., the control coefficient and the baseline model parameters. M is the number of samples in the optimization dataset, and S... d (θ) represents the performance score of the d-th sample under parameter θ.

[0144] The gradient of the objective function with respect to each parameter to be optimized is calculated using the backpropagation algorithm, and is expressed as:

[0145] Where ∇J(θ) is the gradient vector of the objective function, that is, the vector formed by the partial derivatives of the objective function J(θ) with respect to all parameters θ to be optimized. It is the partial derivative of the objective function with respect to a single parameter, and is the basic building block of the gradient vector ∇J(θ).

[0146] The parameters are updated using the stochastic gradient descent (SGD) algorithm, and the update formula is as follows:

[0147] Where, θ h+1 Let θ be the parameter value for the (h+1)th iteration. h Let η be the parameter value for the h-th iteration, and η be the learning rate. The initial value is 0.001, and an adaptive learning rate strategy is adopted, which gradually decreases during the iteration process. The iteration terminates when the change in the objective function value is ≤1e-5.

[0148] In summary, the real-time adjustment system for precise agricultural machinery operation parameters based on visual measurement provided by this invention effectively solves the problems of traditional agricultural machinery operation parameter adjustment, such as reliance on human experience, slow response, and poor adaptability, through the deep integration of multi-source visual measurement technology and adaptive control algorithms. It achieves precise quantitative extraction of crop growth status and environmental information, significantly improving feature extraction confidence and segmentation accuracy, providing reliable data support for operation parameter adjustment, and effectively reducing measurement errors in complex farmland environments. The system constructs a PID control mechanism and dynamic benchmark model that integrates visual feature confidence, enabling real-time adaptive adjustment of operation parameters. Compared to traditional fixed parameter control, the adjustment response delay is shortened, parameter matching accuracy is improved, and it can adapt to dynamic changes in different crop growth stages, soil conditions, and light conditions. By integrating visual data and operation parameter information through DS evidence theory, various anomalies such as parameter matching and adjustment stability can be quickly identified. Combined with scenario-adaptive processing strategies and parameter optimization mechanisms, the impact of abnormal operations on crop growth is reduced, thus lowering the operation failure rate.

[0149] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time adjustment system for precision operation parameters of agricultural machinery based on vision measurement, characterized in that, include: The crop information acquisition module is used to collect visual data on crop growth status and environmental visual information. The feature extraction module is used to extract key visual features of crops and environmental visual features from the visual data of crop growth status and the visual information associated with the environment, and to perform preprocessing to output the core visual feature set of crops, the visual feature set of the environment and feature confidence parameters. The adaptive control module is used to calculate the real-time adjustment amount of agricultural machinery operation parameters based on the crop core visual feature set, environmental visual feature set and feature confidence parameters, and a preset operation parameter benchmark model, and output the operation parameter adjustment command and the adjusted parameter value. The anomaly detection and feedback module is used to combine the operation parameter adjustment instructions and the adjusted parameter values ​​with the combined visual feature data of crop environment to analyze the matching degree and adjustment stability of operation parameters, detect anomalies, and output the anomaly detection results. The anomaly handling module is used to generate corresponding handling strategies for different anomaly types based on the anomaly detection results, and output the anomaly handling results and correction parameters. The parameter optimization module is used to optimize the control coefficient and the preset operation parameter benchmark model based on the anomaly processing results and correction parameters, according to historical control data and long-term crop visual characteristic time series data, using the gradient descent algorithm, and output the optimization results.

2. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 1, characterized in that, The feature extraction module includes a preliminary feature extraction unit, a feature preprocessing unit, and a feature filtering and fusion unit. The preliminary feature extraction unit is used to separate the crop region and the environment region based on the crop growth status visual data and the environment-related visual information, and to extract the initial crop visual features and the initial environment visual features through the feature point detection algorithm, and output the initial feature dataset. The feature preprocessing unit is used to denoise and normalize the initial feature dataset, correct the deviation in the feature extraction process, and output a standardized feature dataset. The feature selection and fusion unit is used to remove redundant features from the standardized feature dataset by evaluating feature importance, fuse effective features to generate a crop core visual feature set and an environment visual feature set, and calculate the extraction confidence parameters of each feature, and output the crop core visual feature set, the environment visual feature set and the feature confidence parameters.

3. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 1, characterized in that, The adaptive control module includes a parameter verification unit, a control amount calculation unit, and a control instruction generation unit; the parameter verification unit is used to verify the completeness and validity of the crop core visual feature set, the environmental visual feature set, and the feature confidence parameters, and to remove invalid data; The adjustment calculation unit is used to calculate the real-time adjustment of agricultural machinery operation parameters based on a preset operation parameter benchmark model and control coefficient using a PID control algorithm, and output the operation parameter adjustment. The control instruction generation unit is used to generate corresponding operation parameter adjustment instructions based on the operation parameter adjustment amount, and to calculate the adjusted parameter value.

4. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 3, characterized in that, The steps in the adjustment calculation unit for presetting the benchmark model of the operating parameters include: Collect visual data of crop growth status and environmental-related visual information of the target crop at different growth stages and under different environmental conditions, as well as historical data of corresponding agricultural machinery operation parameters, wherein the historical data of agricultural machinery operation parameters includes the optimal operation parameters that have been verified in practice. Feature annotation is performed on the visual data of crop growth status and the visual information related to the environment to establish a dataset that maps visual features to the optimal operating parameters; The mapping dataset is preprocessed, and the job parameter data is standardized to obtain a standardized mapping dataset. Based on the mapping relationship between visual features and task parameters, a benchmark model structure with multiple inputs and multiple outputs is constructed. The standardized mapping dataset is divided into a training set and a validation set. The benchmark model structure is trained using the training set, and the model output accuracy is evaluated using the validation set. If the accuracy does not reach the preset accuracy threshold, the model structure parameters are adjusted and the model is retrained to output the benchmark model of the job parameters.

5. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 4, characterized in that, The step of the adjustment calculation unit outputting the operation parameter adjustment amount includes: The crop core visual feature set, environmental visual feature set, and feature confidence parameters are input into the operation parameter benchmark model to obtain the benchmark operation parameters corresponding to the current working condition. Obtain the current actual operating parameters of the agricultural machinery, calculate the deviation between the actual operating parameters and the benchmark operating parameters; and perform a weighted correction on the deviation value by combining the feature confidence parameter to obtain a weighted deviation value. Based on the weighted deviation value, a PID control algorithm is used to calculate the initial adjustment amount of the operating parameters; Based on the physical parameter limitations and operational safety thresholds of agricultural machinery, the initial operational parameter adjustment amount is constrained and verified. If the initial adjustment amount exceeds the allowable range, it is corrected to the maximum or minimum value within the allowable range. The adjusted values ​​of the operation parameters after constraint verification are output as the final adjustment values ​​to the control command generation unit.

6. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 1, characterized in that, The anomaly detection and feedback module includes a data fusion unit, a threshold setting unit, and an anomaly analysis and judgment unit. The data fusion unit is used to fuse the operation parameter adjustment instructions, the adjusted parameter values, and the crop environment combined visual feature data to output a fused dataset. The threshold setting unit is used to set an operation parameter matching threshold and an adjustment stability threshold based on the fused dataset. The anomaly analysis and judgment unit is used to analyze the matching degree between the adjusted parameter values ​​and the crop growth requirements and the stability of the parameter adjustment process based on the operation parameter matching threshold and the adjustment stability threshold, to determine whether there are any anomalies, and to output the anomaly detection result.

7. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 1, characterized in that, The anomaly handling module includes an anomaly type identification unit and a processing strategy generation unit; the anomaly type identification unit is used to identify the anomaly type based on the anomaly detection result; the processing strategy generation unit is used to match a preset processing scheme for different anomaly types and generate anomaly handling strategies and correction parameters.

8. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 7, characterized in that, The steps for the anomaly type identification unit to identify the anomaly type include: The core anomaly characterization data in the anomaly detection results are obtained, including the matching deviation between the adjusted parameter values ​​and crop growth requirements, the fluctuation range of parameter adjustment, the adjustment response delay duration, and the anomaly confidence of the crop environment combined visual feature data. Establish an anomaly type classification system, which includes parameter matching anomalies, adjustment stability anomalies, feature data anomalies, and equipment association anomalies. The core anomaly representation data is matched one by one with each anomaly type determination condition in the anomaly type classification system to determine the anomaly type corresponding to the core anomaly representation data. If the core anomaly representation data matches multiple anomaly types simultaneously, then by combining the anomaly occurrence sequence and data correlation in the anomaly detection results, the dominant anomaly type and associated anomaly types are determined, and an anomaly type identification result containing the dominant anomaly type, associated anomaly types, and anomaly representation details is generated.

9. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 8, characterized in that, The steps of the processing strategy generation unit in generating the exception handling strategy and correction parameters include: Establish a mapping relationship library between anomaly types and handling schemes, and preset corresponding handling priorities, core handling measures and parameter correction rules for the anomaly types; Based on the dominant anomaly type and associated anomaly type, the corresponding basic processing scheme is matched from the mapping relationship library. If associated anomaly type exists, the processing measures corresponding to the associated anomaly type are integrated to form an initial processing strategy. The initial processing strategy is adaptively optimized by combining the parameters of the current operation scenario, eliminating processing measures that do not match the current scenario, and adding scenario adaptability adjustment items to obtain an optimized processing strategy; Based on the optimization processing strategy and the parameter correction rules, and combined with the quantitative indicators in the core anomaly characterization data, specific correction parameter values ​​are calculated. The correction parameters include adjustment algorithm parameter correction values, feature processing parameter correction values, and job parameter safety threshold correction values. The validity of the modified parameter value is verified. If the verification passes, the optimized processing strategy and the modified parameter are output as the final result. If the verification fails, the processing strategy is re-optimized and the modified parameter value is adjusted.

10. The real-time adjustment system for precision agricultural machinery operation parameters based on vision measurement according to claim 7, characterized in that, The parameter optimization module includes an optimization data integration unit and a parameter optimization calculation unit. The optimization data integration unit is used to retrieve historical control data and long-term crop visual characteristic time series data stored in the system based on the anomaly handling strategy and correction parameters, organize and integrate the data, and output an optimized dataset. The parameter optimization calculation unit is used to iteratively optimize the control coefficient and the preset operation parameter benchmark model using the gradient descent algorithm based on the optimization dataset, calculate the optimized control coefficient and operation parameter benchmark model parameters, and output the optimization result.

Images