Facility agriculture voice interaction system based on large model

Abstract

The invention discloses a facility agriculture voice interaction system based on a large model, and relates to the technical field of voice interaction. Comprising a voice signal acquisition and enhancement module, a voice recognition and semantic standardization processing module, a feature extraction and emergency degree quantification module, an intelligent evaluation and emergency instruction judgment module and a priority scheduling and response execution module, voice signals are continuously monitored through a multi-microphone array of the wearable terminal, the direction of a voice source is enhanced by using a far-field beam forming technology, the signals are processed through a time domain / frequency domain noise reduction algorithm, and finally voice streams are cached in real time with millisecond-level sampling precision. Through the multi-microphone array and far-field beam forming technology, voice capture is optimized, noise is reduced, real-time voice recognition and semantic analysis of the edge server are combined, the priority of emergency instructions is quantified, it is ensured that resources are rapidly scheduled to process emergency tasks, the agricultural production efficiency is improved, and crop losses are reduced.

Description

Technical Field

[0001] The present invention relates to the field of voice interaction technology, and in particular to a facility agriculture voice interaction system based on a large model. Background Art

[0002] A large-scale model-based voice interaction system for facility agriculture refers to a wearable device developed by combining deep learning and large-scale language models (such as DeepSeek-R1) with speech recognition technology. This system enables agricultural workers to interact with the system in real time through voice to manage agricultural production. This system can access temperature and humidity data within greenhouses at any time. Agricultural workers simply use voice commands to ask, "What is the current temperature and humidity in greenhouse area XX?" and the system will provide instant feedback. Furthermore, agricultural workers can use voice commands to control greenhouse equipment, such as opening shade nets and adjusting fans, thereby achieving automated regulation. The system also proactively provides environmental warnings, such as temperature and humidity anomalies, through voice prompts, helping agricultural workers respond promptly and reducing manual intervention and reliance on mobile phones or computers. This technology significantly improves the convenience and efficiency of agricultural production, freeing agricultural workers to focus on more effective management and decision-making.

[0003] The existing technology has the following shortcomings: In the existing technology, agricultural workers may issue some emergency commands during the agricultural production process, such as "quickly adjust the temperature" or "urgently turn on the fan". These commands usually indicate that agricultural workers need the system to respond immediately and take measures to deal with sudden environmental changes, such as emergencies such as excessively high temperature or low humidity. However, when processing these emergency commands, existing voice interaction systems may fail to effectively identify the urgency behind the commands, resulting in delayed system responses or execution errors. Specifically, the system may treat these emergency commands in the same way as ordinary commands and fail to prioritize them according to the urgency of the commands. For example, when an agricultural worker issues "quickly adjust the temperature", the system may not prioritize this task, resulting in a delay in the temperature adjustment operation, which in turn affects the growth environment of the crops. Similarly, when an agricultural worker requests "urgently turn on the fan", the system may fail to start the fan immediately, causing the crops to face unsuitable environmental conditions, ultimately affecting the efficiency and quality of agricultural production.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a facility agriculture voice interaction system based on a large model, which optimizes voice capture and noise reduction through multi-microphone array and far-field beamforming technology, combines real-time voice recognition and semantic analysis of edge servers, quantifies the priority of emergency instructions, ensures the rapid scheduling of resources to handle emergency tasks, achieves immediate response, improves agricultural production efficiency and reduces crop losses, so as to solve the problems in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: a large-scale model-based facility agriculture voice interaction system, comprising a voice signal acquisition and enhancement module, a voice recognition and semantic standardization processing module, a feature extraction and emergency quantification module, an intelligent assessment and emergency command determination module, and a priority scheduling and response execution module:

[0007] The voice signal acquisition and enhancement module continuously monitors voice signals at agricultural workers' work sites through the multi-microphone array of the wearable terminal, uses far-field beamforming technology to enhance the direction of the voice source, and processes the signal through time-domain and frequency-domain noise reduction algorithms, ultimately caching the voice stream in real time with millisecond-level sampling accuracy.

[0008] The speech recognition and semantic standardization processing module performs speech recognition and semantic standardization processing on the collected raw speech data in the edge server, and writes the processed structured instructions into a high-quality data set;

[0009] The feature extraction and urgency quantification module performs multi-dimensional feature engineering on the data set to extract core indicators reflecting the urgency of the instruction, and conducts comprehensive analysis on the extracted core indicators to quantify the urgency of the instruction;

[0010] The intelligent assessment and emergency command determination module inputs the core indicators after quantitative analysis into a pre-trained classification-regression hybrid model. The model then intelligently evaluates the voice commands of agricultural workers to determine whether the commands are emergency commands.

[0011] The priority scheduling and response execution module, when it identifies that an instruction is an urgent instruction, immediately triggers a "priority transition", moves the current instruction into the privileged channel, bypasses the regular task queue, prioritizes the mobilization of resources and quickly activates the corresponding control module to achieve instant response and optimized decision-making.

[0012] Preferably, the specific steps for obtaining the voice command data of agricultural workers are as follows:

[0013] At the agricultural work site, agricultural workers first collect voice signals in the environment through the multi-microphone array on the smart terminal;

[0014] Next, far-field beamforming technology is applied, and through algorithm optimization, the directionality of the voice signal is enhanced, thereby focusing on the voice signals emitted by agricultural workers and reducing interference from background noise.

[0015] Then, the collected speech signal is processed for noise removal using a time domain / frequency domain noise reduction algorithm to improve the clarity and quality of the signal;

[0016] Finally, the processed voice signal is cached in real time with millisecond sampling accuracy, ensuring that the voice data can be quickly and accurately transmitted to the subsequent processing module for further analysis and processing.

[0017] Preferably, multi-dimensional feature engineering is performed on the data set to extract core indicators reflecting the urgency of the instruction. The extracted indicators include the short-time energy amplitude change ratio in the voice signal and the number and distribution density of morphemes per unit time around the emergency keyword in the instruction. The short-time energy amplitude change ratio in the voice signal and the number and distribution density of morphemes per unit time around the emergency keyword in the instruction are comprehensively analyzed under the detection window to generate an energy surge reference value and an emergency word-adjacent morpheme density reference value, respectively. The urgency of the instruction is quantified by the energy surge reference value and the emergency word-adjacent morpheme density reference value.

[0018] Preferably, the specific steps of comprehensively analyzing the short-term energy amplitude change ratio in the speech signal within the detection window to generate the energy surge reference value are as follows:

[0019] First, the input speech signal is divided into multiple short-time frames. Then, the short-time energy of each frame signal is calculated. Then, the energy amplitude change ratio between two consecutive frames is calculated. The energy amplitude change ratio calculation expression is as follows:

[0020]

[0021] Where R i，i+1 is the energy amplitude change ratio, that is, the energy amplitude change ratio between the i-th frame and the i+1-th frame, E i and E i +1 represents the short-time energy value of the i-th frame and the i+1-th frame respectively;

[0022] By analyzing the energy amplitude change ratio between multiple frames and accumulating and weighting the energy amplitude change ratio of each frame, an energy surge reference value is further generated. The generation formula is as follows:

[0023]

[0024] Where ESRV is the energy surge reference value, ω i is the energy change weight coefficient of the i-th frame, and N is the total number of frames.

[0025] Preferably, the specific steps of comprehensively analyzing the number and distribution density of morphemes per unit time around the emergency keyword in the instruction within the detection window to generate an emergency word-adjacent morpheme density reference value are as follows:

[0026] Identify and extract the urgent keywords and their surrounding morphemes in the instruction within the detection window. Then, analyze the number of morphemes per unit time around the urgent keywords and their distribution density. The calculation expression is as follows:

[0027]

[0028] Where S density is the density of morphemes around the urgent word, n words is the number of effective morphemes per unit time, d avg is the average distance between the emergent word and its neighboring morphemes;

[0029] Based on the obtained morpheme distribution density S around the urgent word density , we further calculate the reference value of the emergency word-adjacent morpheme density to quantify the urgency of the instruction. The calculation expression is as follows:

[0030]

[0031] Where UWP-PDRV is the reference value of the density of urgent words and adjacent morphemes, is the weight of the jth urgent word, is the distribution density of surrounding morphemes of the jth emergency word, and m is the total number of emergency words.

[0032] Preferably, the energy surge reference value and the emergency word-neighboring morpheme density reference value after quantitative analysis are input into a pre-trained classification-regression hybrid model, and the instruction urgency coefficient is generated by the model. The voice instructions of agricultural workers are intelligently evaluated by the instruction urgency coefficient to determine whether the instructions are emergency instructions.

[0033] Preferably, the command urgency coefficient generated by intelligently evaluating the voice commands of agricultural workers through a pre-trained classification-regression hybrid model is compared with a pre-set command urgency coefficient reference threshold to determine whether the input command is an urgent command. The judgment logic is as follows:

[0034] If the instruction urgency coefficient is greater than the preset instruction urgency coefficient reference threshold, the current input instruction is judged to be an emergency instruction; if the instruction urgency coefficient is less than or equal to the preset instruction urgency coefficient reference threshold, the current input instruction is judged not to be an emergency instruction.

[0035] Preferably, when an instruction is identified as an urgent instruction, a "priority transition" is immediately triggered to move the current instruction into a privileged channel, bypassing the regular task queue, giving priority to mobilizing resources and quickly activating the corresponding control module. The specific steps are as follows:

[0036] When an instruction is identified as an urgent instruction, resources are dynamically scheduled based on the instruction urgency coefficient and available resource status to ensure that urgent tasks receive priority resource support. The resource scheduling optimization formula is as follows:

[0037] ResAlloc=argmax(α·CUC+(1-α)·R avail )

[0038] Where ResAlloc is the resource allocation, CUC is the instruction urgency coefficient, R avail is resource availability, α is the adjustment coefficient;

[0039] After resource scheduling is completed, the activation order of the control modules is determined based on the urgency of the instructions, the available resource status, and environmental conditions. The decision formula is as follows:

[0040] ModActSeq=sort_descend(β·CUC+γ·ResAlloc+δ·EnvDev)

[0041] Where ModActSeq is the activation order of the control modules, indicating the priority of each control module to be activated, EnvDev is the environmental deviation, β, γ, and δ are all weight coefficients, β is used to control the influence of the instruction urgency coefficient CUC in the overall decision, γ controls the influence of resource allocation ResAlloc on the overall decision, δ controls the influence of the environmental deviation EnvDev on the control module activation decision, and sort_descend is the sorting operation.

[0042] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0043] The present invention uses multi-microphone arrays and far-field beamforming technology to ensure clear capture of voice commands in noisy agricultural environments, and combines time-domain / frequency-domain noise reduction algorithms to optimize signal quality. By performing real-time speech recognition and semantic standardization on the edge server, the system can quickly convert commands into structured data, conduct in-depth feature analysis, and accurately quantify the urgency of the commands. The intelligent model ensures that urgent tasks are processed quickly by evaluating the priority of urgent commands, bypasses regular task queues, and prioritizes the scheduling of resources to key control modules for immediate response. This system optimizes the ability to respond to environmental changes in the agricultural production process, significantly reduces crop losses caused by environmental imbalances, and improves production efficiency and the stability of crop growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0045] Figure 1 This is a module schematic diagram of a large-scale model-based facility agriculture voice interaction system of the present invention. DETAILED DESCRIPTION

[0046] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0047] The present invention provides Figure 1 The large-scale model-based facility agriculture voice interaction system shown in the figure includes a voice signal acquisition and enhancement module, a voice recognition and semantic standardization processing module, a feature extraction and emergency quantification module, an intelligent evaluation and emergency command determination module, and a priority scheduling and response execution module:

[0048] The voice signal acquisition and enhancement module continuously monitors voice signals at agricultural workers' work sites through the multi-microphone array of the wearable terminal, uses far-field beamforming technology to enhance the direction of the voice source, and processes the signal through time-domain and frequency-domain noise reduction algorithms, ultimately caching the voice stream in real time with millisecond-level sampling accuracy.

[0049] At agricultural production sites, the smart terminals worn by agricultural workers (such as headphones, glasses, or wearable microphones) have built-in arrays of multiple microphones that can continuously capture surrounding sound signals. In order to obtain the speaker's voice more clearly, the system uses far-field beamforming technology, which can identify and enhance voice signals from a specific direction (i.e., the direction of the agricultural worker's speech) while suppressing interference noise from other directions. In addition, the system also processes the collected multi-channel audio data through time domain and frequency domain noise reduction algorithms to further remove background noise and improve voice clarity. The processed voice stream is sampled with millisecond-level time accuracy and cached in real time on the local device, providing high-quality voice input for subsequent voice recognition and command analysis. This method is particularly suitable for scenes with high environmental noise, such as farmland and greenhouses, and can greatly improve the accuracy and response efficiency of the voice interaction system.

[0050] The specific steps for obtaining agricultural workers' voice command data are as follows: At the agricultural workers' work site, the multi-microphone array on the smart terminal first collects voice signals from the environment. These microphone arrays are capable of capturing audio signals from different directions. Next, the system applies far-field beamforming technology and, through algorithm optimization, enhances the directionality of the voice signal, thereby focusing on the voice signal emitted by the agricultural workers and reducing the interference of background noise. Then, the collected voice signal is de-noised using a time domain / frequency domain noise reduction algorithm to improve signal clarity and quality, especially in noisy agricultural environments. Finally, the processed voice signal is cached in real time with millisecond sampling accuracy, ensuring that the voice data can be quickly and accurately transmitted to the subsequent processing module for further analysis and processing.

[0051] The speech recognition and semantic standardization processing module performs speech recognition (ASR) and semantic standardization on the collected raw speech data in the edge server, and writes the processed structured instructions into a high-quality data set;

[0052] After acquiring the original voice of the agricultural workers, the system transmits the voice data to the edge server (i.e., a device deployed near the site with computing power) for processing. First, the edge server converts the voice content into text information through speech recognition technology (ASR, Automatic Speech Recognition). Subsequently, the system performs "semantic standardization" on the converted text content, that is, normalizing, denoising, adjusting the word order, and unifying the expression to make it conform to the standard format for system command recognition (for example, converting "help me turn on the fan" to "turn on the fan"). After processing, these structured semantic instructions (i.e., control commands with clear content and clear fields) will be stored in a high-quality data set, which will be used for subsequent machine learning model training, instruction optimization analysis or execution decision-making to ensure that the system can accurately understand and quickly respond to similar voice commands in the future. This process helps to improve the accuracy, robustness and real-time performance of voice interaction systems in agricultural scenarios.

[0053] The feature extraction and urgency quantification module performs multi-dimensional feature engineering on the data set to extract core indicators reflecting the urgency of the instruction, and conducts comprehensive analysis on the extracted core indicators to quantify the urgency of the instruction;

[0054] Multi-dimensional feature engineering is performed on the data set to extract core indicators reflecting the urgency of the instructions. The extracted indicators include the short-time energy amplitude change ratio in the speech signal and the number and distribution density of morphemes per unit time around the urgent keywords in the instructions (such as "quick", "hurry up", and "immediately"). The short-time energy amplitude change ratio in the speech signal and the number and distribution density of morphemes per unit time around the urgent keywords in the instructions (such as "quick", "hurry up", and "immediately") are comprehensively analyzed under the detection window to generate energy surge reference values ​​and urgent word-neighboring morpheme density reference values, respectively. The energy surge reference values ​​and the urgent word-neighboring morpheme density reference values ​​are used to quantify the urgency of the instructions.

[0055] A sudden increase in the short-term energy amplitude change ratio in a speech signal can indicate that the current input command is urgent. Urgent commands are often accompanied by farmers' excitement or anxiety. In such situations, the short-term energy of the speech will fluctuate significantly and surge. Specifically, when farmers feel urgent, their tone of voice is often more rapid and emphatic, potentially increasing the intensity or explosiveness of their voices. This causes a sudden change in the energy amplitude of the speech signal over a short period of time. A sudden surge in speech energy is not just an increase in volume; it reflects the speaker's anxiety and urgency, especially in situations requiring a quick response, such as commands like "Adjust the temperature quickly" or "Turn on the fan urgently." When this energy change ratio reaches a certain level, it indicates that the speaker's emotions and needs are extremely urgent, requiring the system to respond with priority. By monitoring the short-term energy amplitude change ratio, the system can promptly detect these sudden spikes, determine the urgency of the command, and schedule tasks promptly and with high priority. This characteristic change is a key acoustic indicator of command urgency and effectively helps the system distinguish between routine commands and urgent ones.

[0056] The specific steps for comprehensively analyzing the short-term energy amplitude change ratio in the speech signal within the detection window to generate the energy surge reference value are as follows:

[0057] First, the input speech signal is divided into multiple short-time frames, with the duration of each frame ranging from 10 to 25 milliseconds. Then, the short-time energy of each frame signal is calculated. The short-time energy can be obtained by calculating the sum of the squares of the amplitudes of each frame signal. Then, the energy amplitude change ratio between two consecutive frames is calculated. The energy amplitude change ratio calculation expression is as follows:

[0058]

[0059] Where R i，i+1 is the energy amplitude change ratio, that is, the energy amplitude change ratio between the i-th frame and the i+1-th frame, E i and E i +1 represents the short-time energy value of the i-th frame and the i+1-th frame respectively;

[0060] By calculating the ratio of short-term energy amplitude changes between every two frames in the speech signal, it captures sudden energy spikes in the speech. This process effectively identifies rapid changes in speech and helps the system determine whether it is an urgent command.

[0061] By analyzing the energy amplitude change ratio between multiple frames and accumulating and weighting the energy amplitude change ratio of each frame, an energy surge reference value is further generated. The generation formula is as follows:

[0062]

[0063] , where ESRV is the energy surge reference value, ω i is the energy change weight coefficient of the i-th frame, which is used to enhance the contribution of the key mutation frame to the energy surge feature. In one implementation, ω of all frames can be set to i =1, indicating equal weighted accumulation; in another way, ω i It can be dynamically adjusted based on the emotional characteristics of the frame, such as a sharp rise in tone, faster speaking speed, and increased volume, which is automatically learned and determined by the training model. N is the total number of frames, which represents the total number of frames after the entire speech signal is framed.

[0064] The Energy Surge Reference Value (ESRV) does not represent the energy ratio of a specific frame, nor does it represent a single energy peak. It is a scalar value that reflects the weighted statistical strength of energy mutation behavior in the entire speech segment. It is used to comprehensively assess whether there is a significant surge in the speaker's speech and serves as one of the input features of intelligent models.

[0065] Weighting factor ω i It can be a constant (equal weight) or dynamically assigned by the training model based on the tone of the frame, changes in speech speed, and the position of urgent keywords, thereby strengthening the contribution of key frames in statistics and suppressing the interference of stable frames.

[0066] The energy surge reference value, generated by comprehensively analyzing the short-term energy amplitude change ratio in the speech signal within the detection window, indicates that the current input command is urgent. Conversely, a lower energy surge reference value indicates that the current input command is not urgent. The energy surge reference value reflects the severity of energy changes in speech. In emergency situations, speech signals are often accompanied by emotional fluctuations. In particular, when farmers are in a state of urgency, their speech speed increases, and their tone becomes more rapid and forceful. This causes a sudden increase in the energy amplitude of the speech, resulting in a noticeable surge. This sudden energy change typically indicates a high level of urgency, requiring a rapid system response. On the other hand, a lower energy surge reference value indicates that the speech signal changes more smoothly, lacking a sense of urgency, and therefore the command may not be urgent.

[0067] In the instruction, the speech rate increases and the vocabulary becomes denser within the surrounding unit time of emergency keywords (such as "quickly", "hurry", "immediately"), which usually indicates that the current input instruction is an emergency instruction. First, emergency keywords like "quickly", "hurry", "immediately" are high-frequency emergency signals in speech, conveying the instruction intention that actions must be taken quickly and immediately. These words usually appear in urgent situations and are often accompanied by intense emotional changes, such as anxiety and eagerness, which prompt the speaker to increase the speech rate and strengthen the tone. Second, the increase in speech rate is a natural physiological and emotional reaction. When the speaker feels urgent or under pressure, they tend to speed up the rhythm of speech to convey urgency. In a tense state, the rhythm of the speech is more rapid, and the increase in speech rate is often an obvious manifestation of an emergency instruction. And vocabulary density means that the speaker conveys more information in a short time, usually through quick and short instructions to reduce waiting time and ensure efficient communication. This dense vocabulary distribution and increased speech rate can not only improve the efficiency of information transmission but also significantly enhance the sense of urgency of the instruction, helping the recipient quickly understand the priority of the task. Therefore, through the analysis of the characteristics of these language behaviors, the system can effectively identify the urgency in the voice instruction and thus make a quick response.

[0068] The specific steps for comprehensively analyzing the number and distribution density of morphemes within the surrounding unit time of emergency keywords (such as "quickly", "hurry", "immediately") in the instruction to generate an emergency word - adjacent morpheme density reference value are as follows:

[0069] Identify and extract the emergency keywords (such as "quickly", "hurry", "immediately") in the instruction and their surrounding morphemes (i.e., other words, phrases, or important markers in the instruction) under the detection window. Then, analyze the number of morphemes within the surrounding unit time of the emergency keyword and their distribution density. The number of morphemes refers to the number of effective words or phrases that can be clearly identified within a short time around the emergency word, usually considering a window of one to two seconds. The distribution density is measured by calculating the spatial distance between the emergency word and other morphemes. Morphemes with a shorter distance indicate an increase in speech rate or sentence compactness. The calculation formula is as follows:

[0070]

[0071] In the formula, S density is the distribution density of morphemes around the emergency word, representing the relationship between the number of morphemes and their distribution density within the surrounding unit time of the emergency keyword. n words is the number of effective morphemes within the unit time, referring to the number of effective morphemes that appear within the surrounding unit time of the emergency keyword (such as "quickly", "hurry", "immediately"). d avgIt is the average distance between an urgent word and its neighboring morphemes, and refers to the average time interval or spatial distance between an urgent word and its surrounding morphemes. In speech instructions, the distance between morphemes is usually measured by their position in the speech signal, reflecting the speed of speech. Morphemes with shorter distances usually mean faster speech, more compact sentences, and a stronger sense of urgency in instructions.

[0072] The purpose of this step is to capture the characteristics of faster speaking speed and dense vocabulary in urgent instructions, reflecting the urgency of the instructions.

[0073] Based on the obtained morpheme distribution density S around the urgent word density , we further calculate the emergency word-neighboring morpheme density reference value to quantify the urgency of the instruction. In this step, the density of the emergency word and the neighboring morphemes is combined to generate the final reference value. Specifically, the morpheme weight of the vocabulary-dense area and the strength of its emergency keywords are considered to adjust the judgment of the emergency instruction. The calculation expression is as follows:

[0074]

[0075] Where UWP-PDRV is the reference value of the density of urgent words and adjacent morphemes, is the weight of the jth urgent word, is the distribution density of surrounding morphemes of the jth emergency word, and m is the total number of emergency words.

[0076] The extracted morpheme distribution density is combined with the weight of the urgent keyword to generate a quantitative reference value for the urgent word-neighboring morpheme density, providing a more accurate assessment of the urgency of the instruction. This reference value dynamically reflects the urgency of the voice instruction, helping the system determine whether the instruction is an urgent task and prioritize it.

[0077] The greater the emergency word - adjacent morpheme density reference value generated by comprehensively analyzing the number and distribution density of morphemes within a unit time around the emergency keywords (such as "quickly", "hurry", "immediately") in the instruction under the detection window, the more likely it indicates that the current input instruction is an emergency instruction. This is because emergency keywords (such as "quickly", "hurry", "immediately") often appear frequently in emergency instructions, and the high - concentration distribution of morphemes (such as other verbs, adjectives or adverbs) around these words means that the speaker conveys more information or requirements in an emergency situation. In this context, the speech rate is usually faster and the tone is more urgent, thus increasing the density of emergency keywords and surrounding morphemes. Specifically, when the number and distribution density of morphemes around the emergency word are high, it indicates that the speaker is trying to convey important information through concise and rapid language, reflecting a sense of urgency and high - priority requirements. On the contrary, if the density is low, it may indicate that the instruction is relatively ordinary and lacks urgency, because the speech rate is slow and the distribution of morphemes is loose, indicating that the emotional urgency of the instruction is weak.

[0078] The intelligent evaluation and emergency instruction determination module inputs the core indicators after quantitative analysis into a pre - trained classification - regression hybrid model, and through the model, it conducts intelligent evaluation on the voice instructions of agricultural workers to determine whether the instruction is an emergency instruction;

[0079] Input the energy surge reference value and emergency word - adjacent morpheme density reference value after quantitative analysis into a pre - trained classification - regression hybrid model, generate an instruction emergency degree coefficient through the model, and conduct intelligent evaluation on the voice instructions of agricultural workers through the instruction emergency degree coefficient to determine whether the instruction is an emergency instruction.

[0080] The "pre - trained classification - regression hybrid model" refers to a composite machine - learning architecture that combines the advantages of classification models and regression models. During the model training stage, it uses historical voice instruction data for joint learning to simultaneously achieve two types of task objectives: one is to determine whether the voice instruction is an emergency instruction (classification task), and the other is to quantify the emergency degree of the instruction (regression task). This model can not only output a clear classification result of "whether it is urgent", but also further give a continuous "instruction emergency degree coefficient", such as a floating - point value between 0 and 1, so as to achieve a fine - grained characterization of the degree of emergency. The typical architecture of this model can include a shared feature extraction layer (such as Transformer, Bi LSTM, convolutional layer, etc.), followed by two branches: the classification branch uses functions such as softmax or sigmoid to output labels, and the regression branch uses linear or ReLU activation functions to output continuous values. Through the multi - task learning framework to jointly optimize two objective functions, the model can more effectively capture the multi - dimensional information features expressing urgency in voice data during the training process, avoiding the ambiguity or misjudgment problems brought by single - label classification.

[0081] In agricultural voice interaction scenarios, this hybrid model is pre-trained on a large number of labeled samples, typically derived from historical data from real farmers using voice commands. Experts provide subjective assessments of urgency, or labels are automatically generated based on the consequences of equipment responses. Model training utilizes features such as "energy surge reference value" and "urgent word-neighboring morpheme density reference value," as well as parameters such as speech rate, pitch, sentence structure, and environmental context to form a comprehensive feature vector input. Through this combined classification-regression training model, the model can more discern non-explicit but semantically urgent commands (such as commands without explicit keywords but with highly urgent speech characteristics) and dynamically output an urgency coefficient. When a new voice command is input into the system, the model quickly assesses its urgency based on this historical learning experience. It not only determines its urgency but also prioritizes tasks and allocates resources based on the urgency coefficient, significantly improving the system's intelligent responsiveness and control efficiency in real agricultural scenarios.

[0082] The classification-regression hybrid model is not limited here. Any classification-regression hybrid model that can perform a comprehensive analysis of the energy surge reference value ESRV and the emergency word-neighboring morpheme density reference value UWP-PDRV to generate the instruction urgency coefficient CUC is acceptable. To implement the technical solution of the present invention, the present invention provides a specific implementation method.

[0083] The command urgency coefficient CUC is generated by the following formula: CUC = k1·ESRV+k2·UWP-PDRV, where k1 and k2 are the preset proportional coefficients of the energy surge reference value ESRV and the emergency word-neighboring morpheme density reference value UWP-PDRV, respectively, and both k1 and k2 are greater than 0.

[0084] Preset scaling factors refer to weight coefficients initialized using prior knowledge or historical data during model construction. They are used to adjust the influence of different features on the final model output. In the formula, k1 and k2 are two scaling factors, associated with the energy surge reference value (ESRV) and the emergency word-neighboring morpheme density reference value (UWP-PDRV), respectively. These coefficients weight the influence of the two input features based on the actual situation, so that the model can comprehensively consider the relative importance of these two features when generating the instruction urgency coefficient (CUC), where CUC∈{0,1}. In other words, by setting an appropriate ratio for k1 and k2, the model can reasonably adjust the contribution of the energy surge and emergency word density features in determining instruction urgency. These coefficients are typically determined through experimentation or data training to ensure model accuracy and generalization. It should be noted that the energy surge reference value (ESRV) and the emergency word-neighboring morpheme density reference value (UWP-PDRV) in the formula are normalized to dimensionless quantities before modeling, ensuring consistent dimensions and allowing for direct linear combination. k1 and k2 are trainable weighting coefficients, which indicate the importance the system attaches to speech spurt behavior and morpheme density behavior. They are usually obtained through model training or empirical setting and are used to adjust the comprehensive impact of feature input on the command urgency index CUC.

[0085] It can be seen from the instruction urgency coefficient that the larger the energy surge reference value generated after comprehensive analysis of the short-time energy amplitude change ratio in the voice signal under the detection window, the larger the emergency word-neighboring morpheme density reference value generated after comprehensive analysis of the number and distribution density of morphemes per unit time around the emergency keywords (such as "quick", "hurry up", "immediately") in the instruction under the detection window, the larger the instruction urgency coefficient generated when the agricultural worker's voice instructions are intelligently evaluated by the pre-trained classification-regression hybrid model, indicating that the instruction currently input by the agricultural worker is an urgent instruction, otherwise, it indicates that the instruction currently input by the agricultural worker is not an urgent instruction.

[0086] The command urgency coefficient generated by intelligently evaluating agricultural workers' voice commands using a pre-trained classification-regression hybrid model is compared with a pre-set reference threshold for the command urgency coefficient to determine whether the input command is an urgent command. The judgment logic is as follows:

[0087] If the instruction urgency coefficient is greater than the preset instruction urgency coefficient reference threshold, the current input instruction is judged to be an emergency instruction; if the instruction urgency coefficient is less than or equal to the preset instruction urgency coefficient reference threshold, the current input instruction is judged not to be an emergency instruction.

[0088] The priority scheduling and response execution module, when it identifies an urgent instruction, immediately triggers a "priority jump" and moves the current instruction into a privileged channel, bypassing the regular task queue, prioritizing resource mobilization and quickly activating the corresponding control module to achieve immediate response and optimized decision-making;

[0089] When a voice command is recognized as an urgent command, the "priority jump" operation is immediately triggered. Its core function is to ensure that urgent commands can receive the fastest and most effective response by dynamically adjusting the task scheduling strategy. Specifically, "priority jump" will raise the processing priority of the current command to the top level of the system, ensuring that the command bypasses the regular task queue, thereby eliminating the delay of waiting in line. This mechanism avoids the risk of queuing urgent commands together with regular tasks, allowing focus on the highest priority tasks in emergency situations, reducing response time and resource contention problems that may arise during processing.

[0090] By prioritizing resources and automatically allocating necessary hardware and computing resources, the system ensures that relevant control modules can be activated in the shortest possible time. For example, in an agricultural production environment, if the instruction involves adjusting the greenhouse temperature, the system will prioritize activating related facilities such as fans and temperature control equipment, rather than waiting for other non-critical tasks to be completed. Rapid activation of the corresponding control module means that the system can immediately control and adjust the relevant equipment after receiving emergency instructions, avoiding the impact of adverse environmental conditions on crops due to slow system response.

[0091] Another important role of this mechanism is to optimize decision-making. Through real-time feedback and adjustments, it not only ensures efficient response to emergency commands, but also adjusts command execution strategies based on current environmental changes to ensure optimal decision-making. In agricultural management, timely resource allocation and control responses are crucial to ensuring crop health and improving production efficiency, and this "priority transition" mechanism is the key to ensuring this rapid and efficient response.

[0092] When an urgent instruction is identified, a "priority transition" is immediately triggered, moving the current instruction into the privileged channel, bypassing the regular task queue, prioritizing resource allocation, and quickly activating the corresponding control module. The specific steps are as follows:

[0093] When an instruction is identified as an urgent instruction, resources are dynamically scheduled based on the instruction urgency coefficient and available resource status to ensure that urgent tasks receive priority resource support. The resource scheduling optimization formula is as follows:

[0094] ResAlloc=argmax(α·CUC+(1-α)·R avail )

[0095] , where ResAlloc is the resource allocation, which represents the optimal configuration set of emergency task resources and determines which devices and resources are allocated for the current emergency instruction, CUC is the instruction urgency coefficient, R avail is the resource availability, which represents the state set of available resources in the current system, and measures the health status and current load of available resources (such as fans, temperature control equipment, etc.). Its value range is [0, 1]. α is the adjustment coefficient, which ranges from [0, 1]. It controls CUC and R avail The weight of these two factors on resource scheduling decisions;

[0096] When the task is extremely urgent (such as the system detects abnormal temperature), α can be set to 0.8-0.9, and the system will give priority to responding to the task; when the resource load is serious, α can be set to 0.2-0.3, and the system will give priority to protecting resources that are currently in good operating status; if an adaptive mechanism is adopted, α can be dynamically generated through a historical data model to achieve self-learning resource scheduling.

[0097] In order to ensure that the parameters in the resource scheduling optimization model have a unified dimension and additivity, the present invention uses the instruction urgency coefficient CUC and resource availability R avail The CUC represents the urgency of the current instruction, and its value comes from the dimensionless score R output by the pre-trained classification-regression model. avail It indicates the operational health and availability of the resources that can be deployed in the current system. Its value comes from the weighted fusion and normalized output of each device status parameter (such as load rate, response delay, failure rate, etc.). avail Although they come from different technical paths, after unified standardization, they can participate in linear combination calculations as similar scoring factors.

[0098] The adjustment coefficient α∈[0,1] introduced in the resource scheduling optimization formula is essentially used to control CUC and R avail The proportion of the two types of indicators in the decision weight. Among them, α is used to adjust the impact intensity of the instruction urgency, and (1-α) represents the adjustment coefficient of the resource status. The two are complementary proportional factors. It should be noted that (1-α) and R avail The product of does not represent a physical or logical coupling between the two. Instead, it serves only as a mathematical weighting measure in the decision-making scoring mechanism, used to construct a flexible and adjustable scheduling function. This approach is consistent with the linear weighted evaluation model widely used in the field of multi-metric ranking and scheduling.

[0099] Furthermore, the aforementioned scheduling formula is used to select the resource set with the highest overall score from multiple candidate resources to execute the response action required by the emergency instruction. This formula does not model physical processes or derive state functions, but rather serves as a practical engineering heuristic scoring decision-making mechanism. It uses a linear combination of normalized parameters to perform ranking, ensuring the scheduling strategy has good responsiveness, adaptability, and system feasibility.

[0100] Through this step, the system can dynamically evaluate the urgency of the instruction CUC and the availability of current resources R avail , and flexibly schedules tasks based on a predefined weight coefficient α. This mechanism ensures that resources are quickly allocated and responded to when urgent tasks are high priority, preventing resource conflicts or overloads from impacting task execution. Ultimately, this enables efficient processing of urgent tasks and maximizes resource utilization, improving overall response efficiency and system reliability.

[0101] After resource scheduling is completed, the activation order of the control modules is determined based on the urgency of the instructions, the available resource status, and environmental conditions. The decision formula is as follows:

[0102] ModActSeq=sort_descend(β·CUC+γ·ResAlloc+δ·EnvDev)

[0103] In the formula, ModActSeq is the activation order of the control modules, indicating the priority of each control module (such as fans, sunshade nets, temperature control systems, etc.) to be activated. By analyzing the emergency instructions, the system will automatically determine the activation order of each control module based on factors such as the urgency of the instruction, resource allocation, and environmental changes. nvDev is the environmental deviation, which indicates the degree of deviation between the current environment and the ideal target environment and reflects the urgency of the control response. Its value range is 0-1. β, γ, and δ are all weight coefficients. β is used to control the influence of the instruction urgency coefficient CUC on the overall decision, γ controls the influence of resource allocation ResAlloc on the overall decision, and δ controls the influence of the environmental deviation EnvDev on the control module activation decision. sort_descend is a sorting operation. The activation order of the control modules is sorted according to the weighted values ​​of β·CUC, γ·ResAlloc, and δ·EnvDev. Through sorting, the system can ensure that tasks with high urgency, urgent resource allocation needs, or large deviations in environmental conditions are prioritized for activation of the corresponding control modules to achieve the fastest response and optimal decision execution.

[0104] To ensure a clear evaluation basis and feasibility for the activation order of control modules, this paper normalizes the three ranking factors involved: the command urgency coefficient (CUC), the resource allocation priority score (ResAlloc), and the environmental deviation (EnvDev). These three factors are dimensionless scoring indicators with values ​​limited to the interval 0 and 1. CUC represents the urgency of the current voice command, ResAlloc represents the priority of each control module in the resource scheduling results, and EnvDev reflects the degree of deviation between the current actual environmental state and the target environmental conditions.

[0105] Before the system activates a control module, it extracts the three scoring factors corresponding to each candidate control module at the current moment. Based on a preset weighting strategy or the system's adaptive adjustment mechanism, a linear weighted calculation is performed on the weight coefficients β, γ, and δ to form a comprehensive ranking score. β is the adjustment weight for CUC, γ is the adjustment weight for ResAlloc, and δ is the adjustment weight for EnvDev. These three factors can be fixed empirical values ​​or dynamically adjusted by the system based on historical data, self-learning parameters, or operational feedback.

[0106] β, γ, and δ are used to adjust the influence of the three scoring factors mentioned above on the module ranking score. They can be preset values ​​(for example, β = 0.4, γ = 0.3, δ = 0.3) or dynamically updated through online tuning or system learning mechanisms to cope with changes in control decision objectives in different scenarios.

[0107] Finally, all candidate control modules are ranked by score using a sorting operation. "sort_descend" sorts the modules' comprehensive scores in descending order, resulting in a priority list for module activation. Based on this priority list, the system selects the control modules with the highest comprehensive scores to prioritize their response to the current emergency command, achieving high-priority responsiveness and optimized control accuracy driven by multiple factors.

[0108] The present invention uses multi-microphone arrays and far-field beamforming technology to ensure clear capture of voice commands in noisy agricultural environments, and combines time-domain / frequency-domain noise reduction algorithms to optimize signal quality. By performing real-time speech recognition and semantic standardization on the edge server, the system can quickly convert commands into structured data, conduct in-depth feature analysis, and accurately quantify the urgency of the commands. The intelligent model ensures that urgent tasks are processed quickly by evaluating the priority of urgent commands, bypasses regular task queues, and prioritizes the scheduling of resources to key control modules for immediate response. This system optimizes the ability to respond to environmental changes in the agricultural production process, significantly reduces crop losses caused by environmental imbalances, and improves production efficiency and the stability of crop growth.

[0109] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0110] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0111] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0112] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0113] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0116] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0117] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0118] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A facility agriculture voice interaction system based on a large model, characterized by: It includes voice signal acquisition and enhancement module, voice recognition and semantic standardization processing module, feature extraction and emergency quantification module, intelligent evaluation and emergency command judgment module, and priority scheduling and response execution module: The voice signal acquisition and enhancement module continuously monitors voice signals at agricultural workers' work sites through the multi-microphone array of the wearable terminal, uses far-field beamforming technology to enhance the direction of the voice source, and processes the signal through time-domain and frequency-domain noise reduction algorithms, ultimately caching the voice stream in real time with millisecond-level sampling accuracy. The speech recognition and semantic standardization processing module performs speech recognition and semantic standardization processing on the collected raw speech data in the edge server, and writes the processed structured instructions into a high-quality data set; The feature extraction and urgency quantification module performs multi-dimensional feature engineering on the data set to extract core indicators reflecting the urgency of the instruction, and conducts comprehensive analysis on the extracted core indicators to quantify the urgency of the instruction; The intelligent assessment and emergency command determination module inputs the core indicators after quantitative analysis into a pre-trained classification-regression hybrid model. The model then intelligently evaluates the voice commands of agricultural workers to determine whether the commands are emergency commands. The priority scheduling and response execution module, when it identifies an emergency instruction, immediately triggers a "priority transition", moving the current instruction into a privileged channel, bypassing the regular task queue, prioritizing resource mobilization and quickly activating the corresponding control module to achieve immediate response and optimized decision-making.

2. A large-scale model-based facility agriculture voice interaction system according to claim 1, characterized in that: The specific steps for obtaining voice command data from agricultural workers are as follows: At the agricultural work site, agricultural workers first collect voice signals in the environment through the multi-microphone array on the smart terminal; Next, far-field beamforming technology is applied, and through algorithm optimization, the directionality of the voice signal is enhanced, thereby focusing on the voice signals emitted by agricultural workers and reducing interference from background noise. Then, the collected speech signal is processed for noise removal using a time domain / frequency domain noise reduction algorithm to improve the clarity and quality of the signal; Finally, the processed voice signal is cached in real time with millisecond sampling accuracy, ensuring that the voice data can be quickly and accurately transmitted to the subsequent processing module for further analysis and processing.

3. A large-scale model-based facility agriculture voice interaction system according to claim 1, characterized in that: Multi-dimensional feature engineering is performed on the data set to extract core indicators reflecting the urgency of the instructions. The extracted indicators include the short-time energy amplitude change ratio in the speech signal and the number and distribution density of morphemes per unit time around the emergency keywords in the instructions. The short-time energy amplitude change ratio in the speech signal and the number and distribution density of morphemes per unit time around the emergency keywords in the instructions are comprehensively analyzed under the detection window to generate energy surge reference values ​​and emergency word-neighboring morpheme density reference values, respectively. The energy surge reference values ​​and emergency word-neighboring morpheme density reference values ​​are used to quantify the urgency of the instructions.

4. A large-scale model-based facility agriculture voice interaction system according to claim 3, characterized in that: The specific steps for comprehensively analyzing the short-term energy amplitude change ratio in the speech signal within the detection window to generate the energy surge reference value are as follows: First, the input speech signal is divided into multiple short-time frames. Then, the short-time energy of each frame signal is calculated. Then, the energy amplitude change ratio between two consecutive frames is calculated. The energy amplitude change ratio calculation expression is as follows: Where R i，i+1 is the energy amplitude change ratio, that is, the energy amplitude change ratio between the i-th frame and the i+1-th frame, E i and E i +1 represents the short-time energy value of the i-th frame and the i+1-th frame respectively; By analyzing the energy amplitude change ratio between multiple frames and accumulating and weighting the energy amplitude change ratio of each frame, an energy surge reference value is further generated. The generation formula is as follows: Where ESRV is the energy surge reference value, ω i is the energy change weight coefficient of the i-th frame, and N is the total number of frames.

5. The large-scale model-based facility agriculture voice interaction system according to claim 3, characterized in that: The specific steps for comprehensively analyzing the number and distribution density of morphemes per unit time around the emergency keyword in the instruction within the detection window to generate the emergency word-adjacent morpheme density reference value are as follows: Identify and extract the urgent keywords and their surrounding morphemes in the instruction within the detection window. Then, analyze the number of morphemes per unit time around the urgent keywords and their distribution density. The calculation expression is as follows: Where S density is the density of morphemes around the urgent word, n words is the number of effective morphemes per unit time, d avg is the average distance between the emergent word and its neighboring morphemes; Based on the obtained morpheme distribution density S around the urgent word density , we further calculate the reference value of the emergency word-adjacent morpheme density to quantify the urgency of the instruction. The calculation expression is as follows: Where UWP-PDRV is the reference value of the density of urgent words and adjacent morphemes, is the weight of the jth urgent word, is the distribution density of surrounding morphemes of the jth emergency word, and m is the total number of emergency words.

6. A large-scale model-based facility agriculture voice interaction system according to claim 3, characterized in that: The energy surge reference value and the emergency word-neighboring morpheme density reference value after quantitative analysis are input into a pre-trained classification-regression hybrid model. The instruction urgency coefficient is generated by the model, and the voice instructions of agricultural workers are intelligently evaluated based on the instruction urgency coefficient to determine whether the instructions are urgent instructions.

7. A large-scale model-based facility agriculture voice interaction system according to claim 6, characterized in that: The command urgency coefficient generated by intelligently evaluating agricultural workers' voice commands using a pre-trained classification-regression hybrid model is compared with a pre-set reference threshold for the command urgency coefficient to determine whether the input command is an urgent command. The judgment logic is as follows: If the instruction urgency coefficient is greater than the preset instruction urgency coefficient reference threshold, the current input instruction is judged to be an emergency instruction; if the instruction urgency coefficient is less than or equal to the preset instruction urgency coefficient reference threshold, the current input instruction is judged not to be an emergency instruction.

8. A large-scale model-based facility agriculture voice interaction system according to claim 7, characterized in that: When an urgent instruction is identified, a "priority transition" is immediately triggered, moving the current instruction into the privileged channel, bypassing the regular task queue, prioritizing resource allocation, and quickly activating the corresponding control module. The specific steps are as follows: When an instruction is identified as an urgent instruction, resources are dynamically scheduled based on the instruction urgency coefficient and available resource status to ensure that urgent tasks receive priority resource support. The resource scheduling optimization formula is as follows: ResAlloc=argmax(α·CUC+(1-α)·R avail ) Where ResAlloc is the resource allocation, CUC is the instruction urgency coefficient, R avail is resource availability, α is the adjustment coefficient; After resource scheduling is completed, the activation order of the control modules is determined based on the urgency of the instructions, the available resource status, and environmental conditions. The decision formula is as follows: ModActSeq=sort_descend(β·CUC+γ·ResAlloc+δ·EnvDev) Where ModActSeq is the activation order of the control modules, indicating the priority of each control module to be activated, ENvDev is the environmental deviation, β, γ, and δ are all weight coefficients. β is used to control the influence of the instruction urgency coefficient CUC in the overall decision, γ controls the influence of resource allocation ResAlloc on the overall decision, δ controls the influence of the environmental deviation EnvDev on the control module activation decision, and sort_descend is the sorting operation.

Images