Radar voice interaction control method and system based on natural language processing

Abstract

The application discloses a radar voice interaction control method and system based on natural language processing, relates to the technical field of voice recognition, and solves the technical problems that it is difficult to combine radar echo signals with hyperspectral image data to detect target states, it is difficult to realize voice interaction control by using natural language processing, it is difficult to reasonably determine the priority of control instructions based on an environment comprehensive coefficient and an emergency degree value, it is difficult to establish a conflict detection mechanism to ensure the feasibility of instructions, and it is difficult to execute instructions under the conditions of multi-target game and dynamic priority change and monitor and evaluate deviation values to establish an error reporting mechanism. The application realizes target state detection by fusing radar echo signals and hyperspectral image data, realizes voice interaction control by using natural language processing, and dynamically generates hierarchical control signals in combination with an environment comprehensive coefficient and an emergency degree value.

Description

Technical Field

[0001] This invention belongs to the field of speech recognition technology, specifically a radar voice interaction control method and system based on natural language processing. Background Technology

[0002] In natural language processing, speech recognition technology uses pattern recognition and machine learning algorithms to extract acoustic features from speech signals and convert them into text; natural language understanding then performs semantic analysis on the text to grasp the user's control intentions. Radar technology detects objects by emitting radio waves and receiving reflected signals; in greenhouse scenarios, it can be used for non-contact crop detection. Combining the advantages of both technologies can meet the needs of intelligent control and remote monitoring in greenhouses, enabling voice-interactive control.

[0003] Existing radar voice interaction control methods and systems based on natural language processing have the following problems: it is difficult to combine radar echo signals with hyperspectral image data for target status detection; it is difficult to use natural language processing to achieve voice interaction control; it is difficult to reasonably determine the priority of control commands based on the comprehensive environmental coefficient and urgency value; it is difficult to establish a conflict detection mechanism to ensure the feasibility of commands; and it is difficult to execute commands in the case of multi-target game and dynamic priority changes, and to monitor and evaluate deviation values ​​to establish an error reporting mechanism. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a radar voice interaction control method and system based on natural language processing to solve the above-mentioned technical problems.

[0005] The first aspect of the present invention provides a radar voice interaction control method based on natural language processing, comprising the following steps:

[0006] S1: Inside the greenhouse, radar echo signals, voice signals, and hyperspectral image data are collected simultaneously, and preprocessed using a multi-threaded parallel preprocessing mechanism;

[0007] S2: Target state detection is performed by combining preprocessed radar echo signals with hyperspectral image data; the semantics of voice commands are parsed after preprocessed speech signals are processed by speech recognition to extract control intent and target parameters;

[0008] S3: Calculate the urgency level value by calculating the comprehensive environmental coefficient and the target status detection result; dynamically bind the target parameters in the voice command to the radar control logic, and generate radar control commands and hierarchical control signals through natural language processing; establish a conflict detection mechanism and verify the feasibility of the command;

[0009] S4: Dynamically execute commands based on multi-objective game theory and priorities; monitor and evaluate deviation values ​​after executing radar control commands, and establish an error reporting mechanism.

[0010] Preferably, step S2, which involves detecting the target state by combining the preprocessed radar echo signal with hyperspectral image data, includes the following steps:

[0011] Based on the radar beamwidth and crop planting density, an adaptive algorithm is used to divide the greenhouse into n rectangular grids;

[0012] Using a sliding window mechanism, the temporal echo intensity and phase difference data of the i-th rectangular grid are continuously acquired, where i∈{1,2,...,n}; vegetation index features are extracted from the hyperspectral image data; and the temporal echo intensity, phase difference data and vegetation index are fused to construct a joint feature vector.

[0013] A lightweight hybrid network model is constructed by combining convolutional neural networks and long short-term memory networks. A joint feature vector dataset is collected and the target state is labeled, including health status and pest / disease status. The labeled dataset is input into the lightweight hybrid network model for training. The trained lightweight hybrid network model outputs the target state detection result using real-time processed radar echo signals and hyperspectral image data.

[0014] Preferably, in step S2, the preprocessed speech signal is subjected to speech recognition to parse the semantics of the speech commands and extract the control intent and target parameters, including the following steps:

[0015] Collect pre-processed historical speech signals and input them into a deep learning-based speech recognition model for training. During the training process, the speech recognition model performs speech decoding and outputs corresponding text-based speech commands.

[0016] The pre-processed speech signal is input into the trained speech recognition model, and the recognition result is output; the speech command text is pre-processed and greenhouse rectangular grid area markers are added.

[0017] Collect a voice command text dataset containing various voice commands and their corresponding control intent labels, and preprocess it;

[0018] The preprocessed voice command text dataset is input into the BERT model for training; real-time acquired voice command text is input into the trained BERT model to output the corresponding control intent label; the token-level classification capability of the BERT model is used to identify the target parameters in the voice command text.

[0019] Preferably, step S3, which calculates the environmental comprehensive coefficient and the urgency value based on the target state detection results, includes the following steps:

[0020] The formula for calculating the comprehensive environmental coefficient is:

[0021]

[0022] Where H is the environmental coefficient, H T and H D R represents ambient temperature and ambient humidity, respectively. Ti R Di and R Fi Let represent the soil temperature, humidity, and fertility of the i-th rectangular grid, respectively; and R represents the optimal ambient temperature and ambient humidity, respectively; Tmax R Dmax and R Fmax ε1 and ε2 represent the optimal soil temperature, humidity, and fertility, respectively; ε1 and ε2 are the influencing weights.

[0023] The formula for calculating urgency is:

[0024]

[0025] Among them, E i S is the urgency value within the i-th rectangular grid. i Let A be the planting area of ​​pest- and disease-prone crops within the i-th rectangular grid, and let A be the total area. ΔV i It is the vegetation index deviation value, V max It is the maximum value of the vegetation index deviation, B i σ1 is the pest and disease spread rate, H is the environmental comprehensive coefficient, σ2 is the confidence score of the BERT model output, σ2 is the confidence score of the lightweight hybrid network model output, and w1, w2 and w3 represent the weight coefficients, respectively.

[0026] Preferably, in step S3, the target parameters in the voice command are dynamically bound to the radar control logic, and radar control commands and hierarchical control signals are generated through natural language processing, including the following steps:

[0027] The target state detection results are bound to the target parameters in the voice commands to build a mapping table, and the mapping table is stored in the control logic database to generate radar control operation commands.

[0028] Based on the information in the mapping table, according to the target parameters and the bound control logic, the corresponding radar control command text is generated using natural language processing technology and converted into an encoded format;

[0029] Parse the control intent label and target parameters, and match the preset operation template from the control logic database; based on the urgency value, classify the control intent into three response priorities: high, medium, and low.

[0030] Based on the assigned priorities, hierarchical control signals are generated, including different priority identifiers and corresponding control operation commands; the generated control signals are then used to control the radar's operating status.

[0031] Preferably, step S3, which establishes a conflict detection mechanism and verifies the feasibility of the instruction, includes the following steps:

[0032] Create a three-dimensional collision detection space, including: radar beam coverage collision, command execution timing collision, and radar power allocation collision;

[0033] By constructing a conflict matrix, the matrix elements are: C_xy=α1*|m_x∩m_y|+α2*|t_x∩t_y|+α3*(P_x+P_y-P_max),

[0034] Where m_x∩m_y represents the overlapping area of ​​the radar beam coverage under the x-th and y-th commands; t_x∩t_y represents the overlapping duration of the command execution time under the x-th and y-th commands; P_x, P_y, and P_max are the radar power and maximum radar power under the x-th and y-th commands, respectively; α1, α2, and α3 are the weighting coefficients for radar beam coverage conflict, command execution timing conflict, and radar power allocation conflict, respectively.

[0035] By traversing all elements in the conflict matrix, if C_xy > 0.35, it is determined that there is a conflict between the x-th instruction and the y-th instruction;

[0036] The confidence interval of the resource utilization rate is obtained by calculating the available power margin of the radar in real time and predicting the resource utilization rate curve over a future time period through Monte Carlo simulation.

[0037] The verification instruction is not feasible when the resources required by the instruction exceed the confidence interval.

[0038] Preferably, step S4, which involves dynamically executing instructions based on multi-objective game theory and priority, includes the following steps: when multiple voice instructions are triggered simultaneously, a Nash equilibrium solution is calculated based on the urgency value, target response priority, and radar load. Then, time-division multiplexing technology is used to divide the radar beam into independent time slots at the microsecond level, and different instructions are executed separately.

[0039] Preferably, in step S4, after executing the radar control command, the deviation value is monitored and evaluated, and an error reporting mechanism is established, including the following steps:

[0040] By real-time monitoring of radar operation status, including radar transmission frequency, scanning angle, power and echo signal-to-noise ratio, the target response status, including target response time, is monitored and normalized.

[0041] By weighted evaluation of the deviation from the actual target, an error is reported to the user when the deviation exceeds a preset threshold.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This invention achieves high-precision detection of the health and pest status of greenhouse crops by fusing radar echo signals and hyperspectral image data, collecting time-series data using adaptive grid partitioning and sliding window mechanisms, constructing a joint feature vector by combining vegetation index features, and then processing it through a lightweight hybrid model composed of convolutional neural networks and long short-term memory networks, significantly improving environmental perception capabilities.

[0044] This invention combines a deep learning speech recognition model with the BERT model to first convert speech signals into text, and then accurately parse semantics to extract control intentions and parameters, enabling radar control to have a natural and fluent voice interaction function, greatly improving the ease and flexibility of operation, and reducing the complexity of manual operation.

[0045] This invention determines command priority by calculating the comprehensive environmental coefficient and urgency value, dynamically binds voice command parameters with radar control logic to generate hierarchical control signals, establishes a conflict detection mechanism to verify command feasibility, and combines multi-objective game theory and priority with time-division multiplexing technology to allocate radar resources during execution, thereby enhancing the resource utilization efficiency of radar control. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of the method of the present invention;

[0048] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0049] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 As shown, a first aspect of the present invention provides a radar voice interaction control method based on natural language processing, comprising the following steps:

[0051] S1: Inside the greenhouse, radar echo signals, voice signals, and hyperspectral image data are collected simultaneously, and preprocessed using a multi-threaded parallel preprocessing mechanism;

[0052] S2: Target state detection is performed by combining preprocessed radar echo signals with hyperspectral image data; the semantics of voice commands are parsed after preprocessed speech signals are processed by speech recognition to extract control intent and target parameters;

[0053] S3: Calculate the urgency level value by calculating the comprehensive environmental coefficient and the target status detection result; dynamically bind the target parameters in the voice command to the radar control logic, and generate radar control commands and hierarchical control signals through natural language processing; establish a conflict detection mechanism and verify the feasibility of the command;

[0054] S4: Dynamically execute commands based on multi-objective game theory and priorities; monitor and evaluate deviation values ​​after executing radar control commands, and establish an error reporting mechanism.

[0055] Specifically, radar, microphone arrays, and hyperspectral cameras are installed inside the greenhouse to simultaneously acquire radar echo signals, voice signals, and hyperspectral image data. A multi-threaded parallel preprocessing mechanism is employed to perform noise filtering and signal normalization on the acquired radar echo signals to remove interference and enhance useful signals; voice signals undergo speech activity detection and endpoint recognition to determine the effective parts of the voice signals and remove silent segments; and hyperspectral image data undergoes radiometric correction and atmospheric correction to improve the accuracy and reliability of the image data. Wavelet filtering, voice endpoint detection, and radiometric correction techniques are used to optimize data quality. Combined with an adaptive grid partitioning strategy, the greenhouse is dynamically divided into rectangular monitoring units. A sliding window mechanism is used to extract time-series echo intensity, phase difference, and vegetation index features, constructing a lightweight hybrid model with joint feature vector input to achieve real-time detection of crop health and pest and disease status. A deep learning-based speech recognition model and a BERT semantic parsing model work together to complete the text conversion of voice commands, intent recognition, and parameter extraction such as spraying area and dosage, and achieve spatial positioning through the greenhouse grid coordinate system. By integrating environmental temperature and humidity, soil parameters, and crop status data, a comprehensive environmental coefficient and urgency level are calculated. Control logic is dynamically linked to generate tiered radar control commands, such as adjusting the scanning angle and activating spraying equipment. Priority is assigned based on a Nash equilibrium game theory model, and execution time slots are allocated in real-time according to radar load. Microsecond-level time-division multiplexing technology is used to achieve multi-command coordination. A three-dimensional conflict detection space is constructed, and the risks of beam coverage, temporal overlap, and power allocation are quantified through a conflict matrix. Monte Carlo simulation is used to predict resource occupancy, automatically avoiding conflict commands. During the execution phase, parameters such as radar transmission frequency, scanning angle, and echo signal-to-noise ratio are monitored in real-time. A weighted deviation value evaluation model is used to establish an error reporting mechanism.

[0056] In this embodiment, step S2, which involves detecting the target state by combining the preprocessed radar echo signal with hyperspectral image data, includes the following steps:

[0057] Based on the radar beamwidth and crop planting density, an adaptive algorithm is used to divide the greenhouse into n rectangular grids;

[0058] Using a sliding window mechanism, the temporal echo intensity and phase difference data of the i-th rectangular grid are continuously acquired, where i∈{1,2,...,n}; vegetation index features are extracted from the hyperspectral image data; and the temporal echo intensity, phase difference data and vegetation index are fused to construct a joint feature vector.

[0059] A lightweight hybrid network model is constructed by combining convolutional neural networks and long short-term memory networks. A joint feature vector dataset is collected and the target state is labeled, including health status and pest / disease status. The labeled dataset is input into the lightweight hybrid network model for training. The trained lightweight hybrid network model outputs the target state detection result using real-time processed radar echo signals and hyperspectral image data.

[0060] Specifically, first, the radar beamwidth and the planting density of crops inside the greenhouse are obtained. Based on these parameters, an adaptive algorithm is used to divide the monitoring area of ​​the greenhouse into n uniformly sized rectangular grids. The size and position of each grid are optimized and determined by the adaptive algorithm based on the radar beam coverage and crop distribution. Specifically, the K-means clustering algorithm can be used to dynamically divide the greenhouse into n rectangular grids, and the formula for calculating the grid side length is: Where L is the grid side length, θ is the radar beamwidth, and ρ is the planting density of crops in the greenhouse. For the i-th rectangular grid, a sliding window mechanism is used to continuously acquire temporal echo intensity and phase difference data within it. The size and step size of the sliding window are set according to the characteristics of the radar signal and the needs of crop dynamic changes to ensure that subtle changes in the target state can be captured. Simultaneously, vegetation index features corresponding to each rectangular grid are extracted from the hyperspectral image data, such as Normalized Difference Vegetation Index (NDVI) and Ratio Vegetation Index (RVI). These indices can effectively reflect the health and growth status of crops. The acquired temporal echo intensity, phase difference data, and extracted vegetation indices are fused to construct a joint feature vector. Specifically, after normalizing these different types of data, they are arranged and combined in a certain order into a vector, serving as a comprehensive feature describing the target state within each rectangular grid. A lightweight hybrid network model is constructed by combining a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network. CNNs are used to extract spatial features from data, while LSTMs are used to capture temporal features. Combining the two fully utilizes the spatiotemporal characteristics of radar echo signals and hyperspectral image data. A large dataset of joint feature vectors containing different target states, including health and pest / disease states, was collected and labeled. The labeled dataset was then fed into a lightweight hybrid network model for training. During training, backpropagation was used to adjust model parameters to minimize the error between predictions and actual labels, enabling the model to accurately learn the feature representations of target states. The trained lightweight hybrid network model can receive real-time processed radar echo signals and hyperspectral image data, outputting the detection results of target states such as crop health and pest / disease states within each rectangular grid, providing a basis for subsequent radar voice interaction control.

[0061] In this embodiment, step S2 involves performing speech recognition on the preprocessed speech signal, parsing the semantics of the speech commands, and extracting the control intent and target parameters. This includes the following steps:

[0062] Collect pre-processed historical speech signals and input them into a deep learning-based speech recognition model for training. During the training process, the speech recognition model performs speech decoding and outputs corresponding text-based speech commands.

[0063] The pre-processed speech signal is input into the trained speech recognition model, and the recognition result is output; the speech command text is pre-processed and greenhouse rectangular grid area markers are added.

[0064] Collect a voice command text dataset containing various voice commands and their corresponding control intent labels, and preprocess it;

[0065] The preprocessed voice command text dataset is input into the BERT model for training; real-time acquired voice command text is input into the trained BERT model to output the corresponding control intent label; the token-level classification capability of the BERT model is used to identify the target parameters in the voice command text.

[0066] Specifically, a large amount of preprocessed historical speech signals are collected, covering various possible commands and sentences, to construct a training dataset. The collected speech signals are then input into a deep learning-based speech recognition model for training. During training, the model uses speech decoding algorithms to convert the speech signals into corresponding text-based speech commands. Commonly used deep learning models include Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), or Transformers. After training on a large amount of data, the model learns the mapping relationship between speech signals and text. The real-time acquired and preprocessed speech signals are then input into the trained speech recognition model, which outputs the corresponding text-based speech commands. Preprocessing may include speech enhancement, background noise removal, and speech signal normalization to improve recognition accuracy. The output text-based speech commands are further preprocessed, including text cleaning (removing irrelevant characters and punctuation), word segmentation (dividing the text into words or phrases), and part-of-speech tagging to facilitate subsequent semantic analysis. Greenhouse rectangular grid area markers are added to associate the speech commands with specific greenhouse areas. For example, a user might mention "the crop in the 3rd row and 5th column" in a voice command, and the system can recognize and mark the corresponding rectangular grid area. A voice command text dataset containing various voice commands and their corresponding control intent labels is collected. Control intent labels can include specific operational instructions such as starting irrigation, stopping ventilation, and spraying pesticides. The collected voice command text dataset is preprocessed, including standardizing the text format, labeling intents, performing data augmentation such as synonym replacement, and sentence restructuring to enrich the training data. The preprocessed voice command text dataset is then input into the BERT model for training. BERT is a pre-trained language model with powerful semantic understanding capabilities. By pre-training on a large amount of text data, BERT can learn deep semantic information of language. Using the trained BERT model, real-time acquired voice command text is input into the model, and the model outputs the corresponding control intent labels. BERT's token-level classification capability is used to classify each word or phrase in the voice command text and identify the target parameters. For example, in the instruction "Spray pesticide on the crop in row 3, column 5 with a dose of 10 ml", the BERT model can identify the operation intention represented by "grid location, pesticide spraying" in "row 3, column 5" and the dose parameter represented by "10 ml".

[0067] In this embodiment, step S3, which calculates the environmental comprehensive coefficient and the urgency value based on the target state detection results, includes the following steps:

[0068] The formula for calculating the comprehensive environmental coefficient is:

[0069]

[0070] Where H is the environmental coefficient, H T and H D R represents ambient temperature and ambient humidity, respectively. Ti R Di and R Fi Let represent the soil temperature, humidity, and fertility of the i-th rectangular grid, respectively; and R represents the optimal ambient temperature and ambient humidity, respectively; Tmax R Dmax and R Fmax ε1 and ε2 represent the optimal soil temperature, humidity, and fertility, respectively; ε1 and ε2 are the influencing weights.

[0071] The formula for calculating urgency is:

[0072]

[0073] Among them, E i S is the urgency value within the i-th rectangular grid. i Let A be the planting area of ​​pest- and disease-prone crops within the i-th rectangular grid, and let A be the total area. ΔV i It is the vegetation index deviation value, V max It is the maximum value of the vegetation index deviation, B i σ1 is the pest and disease spread rate, H is the environmental comprehensive coefficient, σ2 is the confidence score of the BERT model output, σ2 is the confidence score of the lightweight hybrid network model output, and w1, w2 and w3 represent the weight coefficients, respectively.

[0074] Specifically, temperature, humidity, and soil sensors are installed inside the greenhouse to collect environmental and soil data in real time. Using the target state detection results, the planting area of ​​crops affected by pests and diseases and the vegetation index deviation are determined. Confidence values ​​from the BERT model and the lightweight hybrid network model are collected. The optimal environmental temperature and humidity, as well as the optimal soil temperature, humidity, and fertility, are determined based on the type of crop planted. In this embodiment, the optimal environmental temperature is set to 25 degrees Celsius, the optimal environmental humidity to 65%, and the optimal soil temperature to 20 degrees Celsius, the optimal humidity to 70%, and the optimal fertility (EC value) to 2.5 mS / cm. ε1 and ε2 are the influence weights, 0.4 and 0.6, respectively. w1, w2, and w3 represent weight coefficients, 0.4, 0.4, and 0.2, respectively. Based on crop growth requirements, the influence weights and the weight coefficients of each parameter are determined and adjusted through experiments or expert experience. For each rectangular grid, the above parameters are substituted into the above formula to calculate the comprehensive environmental coefficient and the urgency level value.

[0075] In this embodiment, step S3 dynamically binds the target parameters in the voice command with the radar control logic, and generates radar control commands and hierarchical control signals through natural language processing, including the following steps:

[0076] The target state detection results are bound to the target parameters in the voice commands to build a mapping table, and the mapping table is stored in the control logic database to generate radar control operation commands.

[0077] Based on the information in the mapping table, according to the target parameters and the bound control logic, the corresponding radar control command text is generated using natural language processing technology and converted into an encoded format;

[0078] Parse the control intent label and target parameters, and match the preset operation template from the control logic database; based on the urgency value, classify the control intent into three response priorities: high, medium, and low.

[0079] Based on the assigned priorities, hierarchical control signals are generated, including different priority identifiers and corresponding control operation commands; the generated control signals are then used to control the radar's operating status.

[0080] Specifically, the target status detection results, such as the status of pests and diseases, and the crop health status, are bound to the target parameters in the voice commands.

[0081] For example, if the voice command is "Spray pesticide on the crop in row 3, column 5, with a dosage of 10 ml", then "row 3, column 5" and "spray pesticide dosage of 10 ml" will be bound to the target status detection result such as "the crop in row 3, column 5 is in a pest and disease state".

[0082] A mapping table is constructed, recording the status information, voice commands, and corresponding target parameters for each rectangular grid. This mapping table is stored in the control logic database for subsequent control command generation and execution. Based on the information in the mapping table, combined with the target parameters and the bound control logic, specific radar control operation commands are generated.

[0083] For example, if the target status detection result shows that "the crop in the 3rd row and 5th column is in a pest and disease state", and the voice command is "spray pesticide", then the generated operation command may be "spray 10 ml of pesticide on the crop in the 3rd row and 5th column".

[0084] Based on information from the mapping table, combined with target parameters and bound control logic, detailed radar control command text is generated using natural language processing techniques such as text generation models. For example, the generated command text might be: "Radar system, please spray pesticide on the crop in row 3, column 5, at a dosage of 10 ml, with the operation time being immediate." The generated radar control command text is then converted into an encoding format recognizable by the radar system. For example, the command text can be converted into binary code or other machine-readable formats for the radar system to execute. The control intent label and target parameters are parsed, and a preset operation template is matched from the control logic database. For example, if the control intent is "spray pesticide," the corresponding "spray pesticide" operation template is found in the database, which may contain specific spraying steps, dosage, time, and other information.

[0085] Based on urgency values, control intentions are categorized into three response priorities: high, medium, and low. When the urgency value is greater than a threshold of 0.8, the control intention is marked as high priority and requires immediate execution; when the urgency value is greater than or equal to 0.4 and less than or equal to the threshold of 0.8, it is marked as medium priority; and when the urgency value is less than the threshold of 0.4, it is marked as low priority and can be executed later. Based on the assigned priorities, hierarchical control signals are generated, including different priority identifiers and corresponding control operation instructions. The thresholds are adjusted according to actual conditions or historical data.

[0086] For example, a high-priority control signal may include a "high priority" label and a corresponding pesticide spraying instruction; a low-priority control signal may include a "low priority" label and a corresponding irrigation instruction.

[0087] The generated control signals are sent to the radar system to control the radar's operating status. The radar system then performs corresponding operations based on the received control signals, such as adjusting the radar beam direction or controlling the spraying equipment.

[0088] In this embodiment, step S3, which establishes a conflict detection mechanism and verifies the feasibility of the instruction, includes the following steps:

[0089] Create a three-dimensional collision detection space, including: radar beam coverage collision, command execution timing collision, and radar power allocation collision;

[0090] By constructing a conflict matrix, the matrix elements are: C_xy=α1*|m_x∩m_y|+α2*|t_x∩t_y|+α3*(P_x+P_y-P_max),

[0091] Where m_x∩m_y represents the overlapping area of ​​the radar beam coverage under the x-th and y-th commands; t_x∩t_y represents the overlapping duration of the command execution time under the x-th and y-th commands; P_x, P_y, and P_max are the radar power and maximum radar power under the x-th and y-th commands, respectively; α1, α2, and α3 are the weighting coefficients for radar beam coverage conflict, command execution timing conflict, and radar power allocation conflict, respectively.

[0092] By traversing all elements in the conflict matrix, if C_xy > 0.35, it is determined that there is a conflict between the x-th instruction and the y-th instruction;

[0093] The confidence interval of the resource utilization rate is obtained by calculating the available power margin of the radar in real time and predicting the resource utilization rate curve over a future time period through Monte Carlo simulation.

[0094] The verification instruction is not feasible when the resources required by the instruction exceed the confidence interval.

[0095] Specifically, a three-dimensional conflict detection model is constructed, corresponding to radar beam coverage conflict, command execution timing conflict, and radar power allocation conflict, respectively. Each dimension defines a corresponding conflict judgment criterion to quantify the potential conflict degree between commands at different levels. A conflict matrix is ​​constructed, where each element corresponds to the conflict degree between two commands. α1, α2, and α3 are the weight coefficients for radar beam coverage conflict, command execution timing conflict, and radar power allocation conflict, specifically 0.5, 0.3, and 0.2, respectively. Their values ​​are determined based on the relative importance of different conflict types in the actual application scenario, and the sum of α1, α2, and α3 is 1. All elements in the conflict matrix are traversed; when the matrix element value > 0.35, a conflict is determined between the x-th command and the y-th command. The threshold of 0.35 can be dynamically adjusted according to the command response priority. This means that there is an irreconcilable contradiction between the two commands in the use of radar resources, requiring further processing to avoid resource conflicts. The available power margin of the radar is calculated in real time, which involves monitoring the current power consumption and total power capacity of the radar. Using Monte Carlo simulation, based on current instruction execution and historical data, the system predicts the resource utilization curve for a future time period and calculates the confidence interval for resource utilization. When the resources required by an instruction exceed the predicted confidence interval, the system determines that the instruction is infeasible. This means that under the current resource allocation, executing the instruction may cause system resource overload or affect the execution of other critical instructions, therefore, the instruction needs to be rejected or adjusted.

[0096] In this embodiment, step S4, which involves dynamically executing instructions based on multi-objective game theory and priority, includes the following steps: when multiple voice instructions are triggered simultaneously, a Nash equilibrium solution is calculated based on the urgency value, target response priority, and radar load. Then, time-division multiplexing technology is used to divide the radar beam into independent time slots at the microsecond level, and different instructions are executed separately.

[0097] Specifically, relevant information for all simultaneously triggered voice commands is collected, including urgency values, target response priorities, and radar load. The collected data is preprocessed, such as through standardization. A game theory model is constructed, treating multiple voice commands as multiple players, each aiming to acquire radar resources as quickly as possible to execute its task. A strategy set is defined for each command, including different combinations of execution time and resource requirements. A payoff function is established, considering factors such as urgency values, target response priorities, and radar load, quantifying the payoff of each command under different strategy combinations. Game theory algorithms, such as iterative elimination of inferior strategies, optimal response dynamics, or simulated annealing, are used to find Nash equilibrium solutions. A Nash equilibrium solution corresponds to a strategy combination in which no single command can increase its payoff by unilaterally changing its strategy. Based on the execution order and time allocation determined by the Nash equilibrium solution, time-division multiplexing technology is used to divide the radar beam into multiple independent time slots at the microsecond level. A corresponding time slot is allocated to each voice command, ensuring that the radar operation corresponding to that command is executed independently within the allocated time. The radar control operations corresponding to different commands are executed sequentially according to the allocated time slot order. In another embodiment, optimization algorithms such as the Hungarian algorithm can be used to solve the conflict matrix, with the goal of minimizing the total conflict amount, i.e., minimizing sumC_xy, where the summation iterates through all conflicting command pairs. The Hungarian algorithm obtains the optimal solution by finding the minimum weighted matching in the conflict matrix, i.e., adjusting parameters such as command execution order, beam pointing, or power allocation to minimize the total conflict amount. Based on the solution results of the Hungarian algorithm, a set of conflict-free or conflict-minimized command execution schemes can be obtained as the final execution sequence of radar control commands.

[0098] In this embodiment, step S4, after executing the radar control command, monitors and evaluates the deviation value and establishes an error reporting mechanism, including the following steps:

[0099] By real-time monitoring of radar operation status, including radar transmission frequency, scanning angle, power and echo signal-to-noise ratio, the target response status, including target response time, is monitored and normalized.

[0100] By weighted evaluation of the deviation from the actual target, an error is reported to the user when the deviation exceeds a preset threshold.

[0101] Specifically, real-time acquisition of key operating parameters such as radar transmission frequency, scanning angle, power, and echo signal-to-noise ratio (SNR) can reflect the radar's operating status and performance. For example, the radar transmission frequency is measured using a frequency counter, the scanning angle information is obtained using an angle encoder, the radar transmission power is detected using a power sensor, and the echo SNR is calculated using an SNR estimation algorithm. Simultaneously, the target's response to radar control commands is monitored by installing response monitoring sensors on the target equipment or system to record the time delay from receiving the radar command to executing the corresponding operation. The acquired radar operating status parameters and target response times are normalized, mapping them to the same numerical range, such as 0 to 1. Appropriate weighting coefficients are set according to the degree of influence of each monitored parameter on the radar control effect. The weighting coefficients for radar transmission frequency, scanning angle, power, echo SNR, and target response time are 0.25, 0.2, 0.1, 0.25, and 0.2, respectively. The determination of these weighting coefficients can be based on expert experience, historical data statistical analysis, or optimization using machine learning algorithms. Using preset ideal target values, such as ideal transmission frequency, scanning angle, power, echo signal-to-noise ratio, and target response time, as a benchmark, the deviation between the actual monitored value and the ideal target value is calculated. For each parameter, the deviation value can be expressed as the difference or relative error between the actual value and the ideal value. Then, the deviation values ​​of each parameter are weighted and summed according to weighting coefficients to obtain the comprehensive deviation value. A deviation threshold of 0.75 is preset, which can be determined and adjusted according to factors such as the radar system's design requirements, mission type, and historical operating data. During actual operation, if the calculated comprehensive deviation value exceeds this preset threshold, it indicates that the radar's actual operating state deviates significantly from the expected target, which may affect the normal execution of commands, and an error message is immediately sent to the user.

[0102] Please see Figure 2 As shown, this invention is a radar voice interaction control system based on natural language processing, comprising the following modules:

[0103] Data acquisition and preprocessing module: In the greenhouse, radar echo signals, voice signals and hyperspectral image data are acquired simultaneously, and preprocessed using a multi-threaded parallel preprocessing mechanism;

[0104] State detection and semantic parsing module: Target state detection is performed by combining preprocessed radar echo signals with hyperspectral image data; the semantics of voice commands are parsed after preprocessed speech signals are processed by speech recognition to extract control intent and target parameters;

[0105] Dynamic control logic management module: It calculates the urgency value by calculating the comprehensive environmental coefficient and the target status detection result; it dynamically binds the target parameters in the voice command to the radar control logic, generates radar control commands and hierarchical control signals through natural language processing; it establishes a conflict detection mechanism and verifies whether the command is feasible;

[0106] Control execution and status monitoring module: dynamically executes commands based on multi-objective game theory and priorities; monitors and evaluates deviation values ​​after executing radar control commands and establishes an error reporting mechanism.

[0107] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A radar voice interaction control method based on natural language processing, characterized in that, Includes the following steps: S1: Inside the greenhouse, radar echo signals, voice signals, and hyperspectral image data are collected simultaneously, and preprocessed using a multi-threaded parallel preprocessing mechanism; S2: Target state detection is performed by combining preprocessed radar echo signals with hyperspectral image data; the semantics of voice commands are parsed after preprocessed speech signals are processed by speech recognition to extract control intent and target parameters; S3: Calculate the urgency level value by calculating the comprehensive environmental coefficient and the target status detection result; dynamically bind the target parameters in the voice command to the radar control logic, and generate radar control commands and hierarchical control signals through natural language processing; establish a conflict detection mechanism and verify the feasibility of the command; S4: Dynamically execute commands based on multi-objective game theory and priorities; monitor and evaluate deviation values ​​after executing radar control commands, and establish an error reporting mechanism; Step S3 involves calculating the comprehensive environmental coefficient and the urgency level value based on the target state detection results, including the following steps: The formula for calculating the comprehensive environmental coefficient is: in, It is the comprehensive environmental coefficient. and These represent ambient temperature and ambient humidity, respectively. , and Let represent the soil temperature, humidity, and fertility of the i-th rectangular grid, respectively; and These represent the optimal ambient temperature and ambient humidity, respectively. , and These represent the optimal soil temperature, humidity, and fertility, respectively. and It affects the weight; The formula for calculating urgency is: in, It is the urgency value within the i-th rectangular grid. It represents the planting area of ​​pest- and disease-prone crops within the i-th rectangular grid. It is the total area. This is the vegetation index deviation value. It is the maximum value of the vegetation index deviation. It is the rate of spread of pests and diseases. It is the comprehensive environmental coefficient. It is the confidence score output by the BERT model. The confidence level of the output of the lightweight hybrid network model. , and These represent the weighting coefficients; Step S4 involves monitoring and evaluating deviation values ​​after executing radar control commands and establishing an error reporting mechanism, including the following steps: By real-time monitoring of radar operation status, including radar transmission frequency, scanning angle, power and echo signal-to-noise ratio, the target response status, including target response time, is monitored and normalized. By weighted evaluation of the deviation from the actual target, an error is reported to the user when the deviation exceeds a preset threshold; In actual operation, if the calculated comprehensive deviation value exceeds this preset threshold, it indicates that there is a large deviation between the actual operating state of the radar and the expected target, which may affect the normal execution of commands. An error message will be immediately sent to the user.

2. The radar voice interaction control method based on natural language processing according to claim 1, characterized in that, Step S2 involves detecting the target state by combining the preprocessed radar echo signal with hyperspectral image data, including the following steps: Based on the radar beamwidth and crop planting density, an adaptive algorithm is used to divide the greenhouse into n rectangular grids; Using a sliding window mechanism, the temporal echo intensity and phase difference data of the i-th rectangular grid are continuously acquired, where, Extract vegetation index features from hyperspectral image data; fuse time-series echo intensity, phase difference data, and vegetation indices to construct a joint feature vector; A lightweight hybrid network model is constructed by combining convolutional neural networks and long short-term memory networks. A joint feature vector dataset is collected and the target state is labeled, including health status and pest / disease status. The labeled dataset is input into the lightweight hybrid network model for training. The trained lightweight hybrid network model outputs the target state detection result using real-time processed radar echo signals and hyperspectral image data.

3. The radar voice interaction control method based on natural language processing according to claim 2, characterized in that, In step S2, the preprocessed speech signal is subjected to speech recognition, and the semantics of the speech commands are parsed to extract the control intent and target parameters. This includes the following steps: Collect pre-processed historical speech signals and input them into a deep learning-based speech recognition model for training. During the training process, the speech recognition model performs speech decoding and outputs corresponding text-based speech commands. The pre-processed speech signal is input into the trained speech recognition model, and the recognition result is output; the speech command text is pre-processed and greenhouse rectangular grid area markers are added. Collect a voice command text dataset containing various voice commands and their corresponding control intent labels, and preprocess it; The preprocessed voice command text dataset is input into the BERT model for training; real-time acquired voice command text is input into the trained BERT model to output the corresponding control intent label; the token-level classification capability of the BERT model is used to identify the target parameters in the voice command text.

4. The radar voice interaction control method based on natural language processing according to claim 1, characterized in that, In step S3, the target parameters in the voice command are dynamically bound to the radar control logic, and radar control commands and hierarchical control signals are generated through natural language processing, including the following steps: The target state detection results are bound to the target parameters in the voice commands to build a mapping table, and the mapping table is stored in the control logic database to generate radar control operation commands. Based on the information in the mapping table, according to the target parameters and the bound control logic, the corresponding radar control command text is generated using natural language processing technology and converted into an encoded format; Parse the control intent label and target parameters, and match the preset operation template from the control logic database; based on the urgency value, classify the control intent into three response priorities: high, medium, and low. Based on the assigned priorities, hierarchical control signals are generated, including different priority identifiers and corresponding control operation commands; the generated control signals are then used to control the radar's operating status.

5. The radar voice interaction control method based on natural language processing according to claim 4, characterized in that, Step S3 establishes a conflict detection mechanism and verifies the feasibility of the instructions, including the following steps: Create a three-dimensional collision detection space, including: radar beam coverage collision, command execution timing collision, and radar power allocation collision; By constructing a conflict matrix, the matrix elements are: , in, This represents the overlapping area of ​​the radar beam coverage under the x-th and y-th instructions; This indicates the overlap in execution time between instructions x and y. , and These are the radar power and maximum radar power under the xth and yth instructions, respectively; , and These are the weighting coefficients for radar beam coverage conflict, command execution timing conflict, and radar power allocation conflict, respectively. By iterating through all elements in the conflict matrix, when If the value is 0.35, then it is determined that there is a conflict between instruction x and instruction y; The confidence interval of the resource utilization rate is obtained by calculating the available power margin of the radar in real time and predicting the resource utilization rate curve over a future time period through Monte Carlo simulation. The verification instruction is not feasible when the resources required by the instruction exceed the confidence interval.

6. The radar voice interaction control method based on natural language processing according to claim 1, characterized in that, The step S4, which involves dynamically executing instructions based on multi-objective game theory and priority, includes the following steps: when multiple voice instructions are triggered simultaneously, a Nash equilibrium solution is calculated based on the urgency level, target response priority, and radar load. Time-division multiplexing technology is then used to divide the radar beam into independent time slots at the microsecond level, and different instructions are executed separately.

7. A radar voice interaction control system based on natural language processing, using the radar voice interaction control method based on natural language processing as described in any one of claims 1-6, characterized in that, Includes the following modules: Data acquisition and preprocessing module: In the greenhouse, radar echo signals, voice signals and hyperspectral image data are acquired simultaneously, and preprocessed using a multi-threaded parallel preprocessing mechanism; State detection and semantic parsing module: Target state detection is performed by combining preprocessed radar echo signals with hyperspectral image data; the semantics of voice commands are parsed after preprocessed speech signals are processed by speech recognition to extract control intent and target parameters; Dynamic control logic management module: It calculates the urgency value by calculating the comprehensive environmental coefficient and the target status detection result; it dynamically binds the target parameters in the voice command to the radar control logic, generates radar control commands and hierarchical control signals through natural language processing; it establishes a conflict detection mechanism and verifies whether the command is feasible; Control execution and status monitoring module: dynamically executes commands based on multi-objective game theory and priorities; monitors and evaluates deviation values ​​after executing radar control commands and establishes an error reporting mechanism.

Images