Agricultural Information Big Data Service System

Abstract

The present application discloses an agricultural information big data service system, which relates to the field of big data service technology. It first obtains the voice input of the planting problem input by the farmer, then performs voice recognition and understanding on the planting problem voice input to obtain the semantic features of the planting problem, and then generates the planting problem answer text based on the planting problem semantic features, and finally converts the planting problem answer text into a voice signal and plays it through a speaker. The present application can realize the automatic answer to the planting problems of farmers based on agricultural information big data, thereby providing farmers with answers and services to planting problems in a more timely and effective manner, and promoting the improvement and development of agricultural production.

Description

Technical Field

[0001] The present application relates to the field of big data service technology, and more specifically, to an agricultural information big data service system. Background Art

[0002] Agriculture is an important pillar industry of the national economy. With the advancement of science and technology and the development of informatization, agricultural informatization has become an important direction for agricultural development. Agricultural informatization can improve the efficiency and quality of agricultural production, reduce resource waste, and promote sustainable agricultural development. As an important part of agricultural informatization, agricultural big data can provide scientific basis and support for agricultural production decision-making by collecting, storing, analyzing and applying agricultural data.

[0003] However, the traditional agricultural information big data service system has some shortcomings, such as serious agricultural information islands and low comprehensive data utilization rate. This is due to the fact that agricultural data involves many data types, inconsistent data structures and uneven data quality, resulting in a large workload for data analysis and collation, and low comprehensive data utilization rate. In addition, since agricultural products are greatly affected by the market, but farmers have limited access to market information, it is difficult to grasp the latest market information in a timely manner, resulting in information disconnection between agricultural producers and consumers. In addition, my country's agricultural production is relatively scattered, and it is difficult to collect and analyze agricultural-related data. It is difficult to be accurate, efficient and timely, resulting in inaccurate agricultural decision-making.

[0004] Therefore, an optimized agricultural information big data service system is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides an agricultural information big data service system. It can realize automatic answers to farmers' planting problems based on agricultural information big data, thereby providing farmers with answers and services to planting problems in a more timely and effective manner, and promoting the improvement and development of agricultural production.

[0006] According to one aspect of the present application, there is provided an agricultural information big data service system, which includes:

[0007] A voice input module is used to obtain voice input of planting problems input by farmers;

[0008] A planting question semantic understanding module, used for performing speech recognition and understanding on the planting question speech input to obtain the planting question semantic features;

[0009] A planting problem answering text generation module, used for generating a planting problem answering text based on the planting problem semantic features;

[0010] The answer voice playing module is used to convert the planting problem answer text into a voice signal and play it through a speaker.

[0011] The planting problem semantic understanding module includes:

[0012] A question speech recognition unit, used for performing speech recognition on the planting question speech input to obtain a text description of the planting question;

[0013] A sentence granularity division unit, used for dividing the text description of the planting problem into sentences to obtain a sequence of planting problem description sentences;

[0014] A sentence granularity vectorized encoding unit, used for vectorizing and encoding the sequence of the planting problem description sentences to obtain a sequence of planting problem description sentence vectors;

[0015] The planting problem semantic encoding unit is used to semantically encode the sequence of the planting problem description sentence vectors to obtain a planting problem semantic encoding feature vector as the planting problem semantic feature.

[0016] Specifically, the sentence granularity vectorized encoding unit is used to:

[0017] The sequence of the planting problem description sentences is converted into a sequence of the planting problem description sentence vectors through the FastText model.

[0018] Specifically, the planting problem semantic encoding unit is used to:

[0019] The sequence of the planting problem description sentence vectors is passed through a transformer-based context encoder to obtain the planting problem semantic encoding feature vector.

[0020] Specifically, the planting problem answering text generation module includes:

[0021] A feature semantic residual compensation unit, used to perform residual compensation on the feature semantics of the context-related coding by using the text semantics of the source text constituted by the sequence of the planting problem description sentence vectors to obtain an optimized planting problem semantic coding feature vector;

[0022] The planting question answering unit is used to pass the optimized planting question semantic encoding feature vector through the planting question answering model based on the AIGC model to obtain the planting question answering text.

[0023] Specifically, the feature semantic residual compensation unit includes:

[0024] A source text semantic cascade subunit, used for cascading the sequence of the planting problem description sentence vectors to obtain a planting problem source text semantic cascade encoding vector;

[0025] The planting problem dynamic memory enhancement subunit is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into the dynamic memory enhancement feature interaction module to obtain an optimized planting problem semantic coding feature vector.

[0026] Specifically, the planting problem dynamic memory enhancement subunit includes:

[0027] A planting problem feature collaborative enhancement secondary subunit is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into a feature collaborative enhancement network to obtain an implicit collaborative coding vector between planting problem features;

[0028] An adaptive memory processing secondary subunit is used to write the implicit collaborative coding vector between the planting problem features into the adaptive memory module to obtain a dynamic key vector of the planting problem;

[0029] The planting problem interactive feature coding secondary subunit is used to extract the planting problem dynamic key vector from the adaptive memory module, and input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into the interactive feature attention network to obtain the optimized planting problem source text semantic cascade coding vector and the optimized planting problem semantic coding feature vector;

[0030] The planting problem feature linear harmonization secondary subunit is used to input the optimized planting problem source text semantic cascade coding vector and the optimized planting problem semantic coding feature vector into the feature linear harmonization module to obtain the optimized planting problem semantic coding feature vector.

[0031] Specifically, the planting problem features synergistically enhance the secondary subunit, which is used to:

[0032] Adding the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem joint semantic representation vector;

[0033] Subtracting the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem semantic deviation representation vector;

[0034] Performing position point multiplication on the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector to obtain a planting problem interactive semantic representation vector;

[0035] The planting problem joint semantic representation vector, the planting problem semantic deviation representation vector and the planting problem interactive semantic representation vector are cascaded to obtain a planting problem multimodal semantic representation vector, and the planting problem multimodal semantic representation vector is point convolved and then input into a leakage corrected linear unit to obtain an implicit collaborative coding vector between planting problem features.

[0036] Specifically, the planting problem interactive feature encoding secondary subunit is used to:

[0037] Determine the planting problem source text semantic query embedding matrix, the planting problem source text semantic query bias vector, the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector;

[0038] Based on the planting problem source text semantic query embedding matrix and the planting problem source text semantic query bias vector, performing a first linear transformation on the planting problem source text semantic cascade encoding vector to obtain a planting problem source text semantic query vector;

[0039] Based on the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector, performing a second linear transformation on the planting problem source text semantic cascade encoding vector to obtain a planting problem source text semantic value vector;

[0040] The planting question dynamic key vector, the first linear transformation to obtain the planting question source text semantic query vector and the planting question source text semantic value vector are input into an interactive feature attention network based on the first converter structure to obtain an optimized planting question source text semantic cascade encoding vector.

[0041] This application has at least the following technical effects:

[0042] Compared with the prior art, the agricultural information big data service system provided by the present application first obtains the voice input of the planting problem input by the farmer, then performs voice recognition and understanding on the voice input of the planting problem to obtain the semantic features of the planting problem, and then generates a planting problem answer text based on the planting problem semantic features. Finally, the planting problem answer text is converted into a voice signal and played through a speaker. It can realize automatic answers to farmers' planting problems based on agricultural information big data, thereby providing farmers with answers and services to planting problems in a more timely and effective manner, and promoting the improvement and development of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present application.

[0044] Figure 1 It is a block diagram of an agricultural information big data service system according to an embodiment of the present application.

[0045] Figure 2 It is a block diagram of the planting problem semantic understanding module in the agricultural information big data service system according to the embodiment of the present application.

[0046] Figure 3 This is a block diagram of the planting question answer text generation module in the agricultural information big data service system according to an embodiment of the present application.

[0047] Figure 4 This is a flowchart of the agricultural information big data service method according to an embodiment of the present application.

[0048] Figure 5 This is a schematic diagram of the system architecture of the agricultural information big data service method according to an embodiment of the present application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0050] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0051] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0052] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0053] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0054] In response to the above problems, the technical concept of this application is to construct an automatic answer and service system for farmers' planting problems. By utilizing a semantic understanding model to perform semantic recognition and analysis on the planting problems input by farmers, a text of answers to the planting problems is generated. In this way, automatic answers to farmers' planting problems can be achieved based on agricultural information big data, thereby providing farmers with answers and services to planting problems in a more timely and effective manner, thereby promoting the improvement and development of agricultural production.

[0055] It should be noted that the acquisition and processing of all information or data in this application are carried out in compliance with the relevant national data protection laws and policies and with the authorization given by the owner of the corresponding device.

[0056] Figure 1 FIG. 1 is a block diagram of an agricultural information big data service system according to an embodiment of the present application. Figure 1 As shown, according to the agricultural information big data service system 100 of the embodiment of the present application, it includes: a voice input module 110, which is used to obtain the voice input of the planting problem input by the farmer; a planting problem semantic understanding module 120, which is used to perform voice recognition and understanding on the planting problem voice input to obtain the semantic features of the planting problem; a planting problem answer text generation module 130, which is used to generate the planting problem answer text based on the planting problem semantic features; an answer voice playback module 140, which is used to convert the planting problem answer text into a voice signal and play it through a speaker.

[0057] Specifically, in the technical solution of the present application, first, voice input of planting questions input by farmers is obtained. Among them, since some farmers may not be familiar with keyboard input or text input, in order to provide farmers with a more convenient and quick way to input questions, the big data service system of agricultural information can directly obtain the voice input of planting questions input by farmers for analysis and processing, thereby solving the planting problems.

[0058] Next, in order to have a better semantic understanding of the planting problem and thus provide timely and accurate answers to the problem, in the technical solution of the present application, it is necessary to perform voice recognition on the voice input of the planting problem to obtain a text description of the planting problem, so as to facilitate subsequent semantic feature extraction, and to better understand the farmers' problems, answer the problems and provide related services.

[0059] Among them, since the planting problems input by the farmers may contain multiple aspects and details, and the planting problems usually involve multiple relevant factors and conditions. Therefore, in the technical solution of the present application, before performing semantic feature extraction, the text description of the planting problem needs to be segmented into sentences to obtain a sequence of planting problem description sentences. By segmenting the text description of the planting problem into multiple sentences, it can be helpful to better capture the specific details and key information of each planting problem in the future, and it is helpful to better analyze the contextual relationship of each sentence, thereby helping the system to understand the problem more comprehensively and provide more targeted answers.

[0060] Then, the sequence of the planting problem description sentences is converted into a sequence of planting problem description sentence vectors through the FastText model, so as to convert the text into a vector representation, so as to facilitate further semantic analysis and processing by the computer. Among them, by converting the sequence of the planting problem description sentences into a vector representation, the semantic information of the planting problem text can be converted into a numerical form, so as to facilitate computer processing and analysis. In addition, the vector representation can capture the semantic features and relationships of the sentence, which is helpful for subsequent semantic analysis and problem solving tasks. In particular, the FastText model is a word vector representation method based on the bag of words model, which can convert each word in the sentence into a vector, and obtain the vector representation of the entire sentence by averaging or weighted averaging. In this way, the important features of the sentence can be extracted, including word meaning, contextual relationship, etc., to provide useful information for subsequent semantic analysis.

[0061] Furthermore, the sequence of sentence vectors describing the planting problem is encoded through a converter-based context encoder to extract the global contextual semantic association feature information based on sentence granularity in the text description of the planting problem, that is, the global semantic understanding feature information of each sentence in the planting problem, thereby obtaining a semantic encoding feature vector of the planting problem.

[0062] Accordingly, if Figure 2As shown, the planting problem semantic understanding module 120 includes: a problem speech recognition unit 121, which is used to perform speech recognition on the planting problem speech input to obtain a planting problem text description; a sentence granularity division unit 122, which is used to segment the planting problem text description in units of sentences to obtain a sequence of planting problem description sentences; a sentence granularity vectorization encoding unit 123, which is used to vectorize the sequence of planting problem description sentences to obtain a sequence of planting problem description sentence vectors; and a planting problem semantic encoding unit 124, which is used to semantically encode the sequence of planting problem description sentence vectors to obtain a planting problem semantic encoding feature vector as the planting problem semantic feature. Among them, the problem speech recognition unit 121 is used to perform speech recognition on the speech input of the planting problem and convert the speech into a text description. It uses speech recognition technology to convert the speech input of the planting problem into a processable text form for subsequent processing and analysis. The sentence granularity division unit 122 segments the text description of the planting problem in units of sentences and divides the text description into a sequence of multiple sentences. The purpose of this is to split the description of the planting problem into smaller semantic units to facilitate subsequent processing and understanding. The sentence-granularity vectorized coding unit 123 performs vectorized coding on the sequence of descriptive sentences of the planting problems, and converts each sentence into a vector representation. The vectorized coding can convert the text into a numerical representation that can be processed by a computer for subsequent calculations and analysis. The planting problem semantic coding unit 124 performs semantic coding on the sequence of planting problem description sentence vectors, and further converts the sentence vectors into semantic coding feature vectors. The semantic coding can capture the semantic associations and semantic information between sentences, thereby extracting the semantic features of the planting problems for subsequent question understanding and answering. These four units together constitute the planting problem semantic understanding module, which realizes the conversion process from the speech input of the planting problem to the semantic features, and provides a basis for subsequent question answering and processing.

[0063] More specifically, the sentence granularity vectorization encoding unit 123 is used to: convert the sequence of the planting problem description sentences into a sequence of the planting problem description sentence vectors through the FastText model. Among them, FastText is a text classification and word vector representation method based on the bag-of-words model. : FastText represents each word as a character-level n-gram (a combination of n consecutive characters), so that the sub-word information inside the word can be captured. FastText uses technologies such as hierarchical Softmax and negative sampling to make the model training and reasoning very fast. It can process large-scale text data and generate high-quality word vector representations in a shorter time. FastText can be used for text classification tasks. It represents text based on the average or weighted average of word vectors, and then classifies it through a softmax classifier. In addition, FastText can also be used to calculate the similarity between texts, such as query expansion and relevance ranking in search engines. In the sentence-granularity vectorized encoding unit, the FastText model can be used to convert the sequence of planting question description sentences into a vector sequence. It learns the vector representation of words in each sentence and combines technologies such as the bag-of-words model and hierarchical Softmax to represent the sentence in vector form. This can convert the text description into a numerical representation that can be processed by a computer, providing a basis for subsequent question understanding and answering.

[0064] More specifically, the planting problem semantic encoding unit 124 is used to: pass the sequence of the planting problem description sentence vectors through a converter-based context encoder to obtain the planting problem semantic encoding feature vector. It is worth mentioning that in the planting problem semantic encoding unit, the converter-based context encoder mentioned refers to a model architecture, and the commonly used representative is the Transformer model. The Transformer model is an encoder-decoder structure based on a self-attention mechanism, which is widely used in natural language processing tasks. The main function of the converter model is to capture the context information and semantic associations in the sentence, thereby generating a richer semantic encoding feature vector. It is composed of a multi-layer self-attention mechanism and a feedforward neural network, and can perform parallel calculation and encoding on the input sentence sequence. Specifically, the converter model can pay attention to and assign weights to different positions in the sentence through the self-attention mechanism, thereby capturing the semantic associations within the sentence. It can simultaneously consider the information of all positions in the sentence without being restricted by the sequential calculation of the traditional recurrent neural network, so it has better parallel computing capabilities and longer-distance dependency modeling capabilities. Through the converter-based context encoder, the sequence of planting problem description sentence vectors can be encoded into a more semantically expressive semantic encoding feature vector. These feature vectors can better capture the semantic information of planting questions, including the relationship, importance and semantic similarity between sentences, and provide richer semantic representation for subsequent question understanding, answering and reasoning.

[0065] Furthermore, the semantic coding feature vector of the planting problem is passed through the planting problem question-answering model based on the AIGC model to obtain the planting problem answer text. Among them, the semantic coding feature vector of the planting problem contains the semantic information and important features of the problem. By inputting the semantic coding feature vector of the planting problem into the question-answering model, the model can use its learned knowledge and semantic reasoning ability to convert the semantic information into a specific answer text. In this way, the abstract semantic information can be converted into practical answers that farmers can understand and apply. Furthermore, in order to facilitate farmers' understanding, the planting problem answer text is further converted into a voice signal and played through a speaker.

[0066] Accordingly, if Figure 3 As shown, the planting question answer text generation module 130 includes: a feature semantic residual compensation unit 131, which is used to perform residual compensation on the feature semantics of the context-related encoding by using the text semantics of the source text composed of the sequence of the planting question description sentence vectors to obtain an optimized planting question semantic encoding feature vector; and a planting question answering unit 132, which is used to pass the optimized planting question semantic encoding feature vector through a planting question answering model based on the AIGC model to obtain the planting question answer text.

[0067] In the technical solution of the present application, when the sequence of the planting problem description sentence vectors is passed through a converter-based context encoder to obtain a planting problem semantic coding feature vector, the converter-based context encoder will perform context-association coding on the sequence of the planting problem description sentence vectors, and, considering the correspondence between the feature semantics of the context-association coding and the text semantics of the source text, preferably, in the technical solution of the present application, the text semantics of the source text constituted by the sequence of the planting problem description sentence vectors is used to perform residual compensation on the feature semantics of the context-association coding to improve semantic richness and generation robustness. Specifically, the feature semantic residual compensation unit includes: a source text semantic cascade subunit, which is used to cascade the sequence of the planting problem description sentence vectors to obtain a planting problem source text semantic cascade coding vector; and a planting problem dynamic memory enhancement subunit, which is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into a dynamic memory enhancement feature interaction module to obtain an optimized planting problem semantic coding feature vector. Among them, the dynamic memory enhanced feature interaction module is a new feature interaction fusion architecture, which aims to achieve efficient fusion and enhancement between feature vectors from different sources through multi-stage feature processing and optimization. Specifically, the dynamic memory enhanced feature interaction module integrates the feature collaborative enhancement network, the adaptive memory module, the interactive feature attention network, and the feature linear reconciliation module to generate a highly optimized and highly expressive semantic encoding feature vector of the planting problem.

[0068] Specifically, the planting problem dynamic memory enhancement subunit includes: a planting problem feature collaborative enhancement secondary subunit, which is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into a feature collaborative enhancement network to obtain an implicit collaborative coding vector between planting problem features; an adaptive memory processing secondary subunit, which is used to write the implicit collaborative coding vector between planting problem features into an adaptive memory module to obtain a planting problem dynamic key vector; a planting problem interactive feature coding secondary subunit, which is used to extract the planting problem dynamic key vector from the adaptive memory module, and input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into an interactive feature attention network respectively to obtain an optimized planting problem source text semantic cascade coding vector and an optimized planting problem semantic coding feature vector; a planting problem feature linear harmonization secondary subunit, which is used to input the optimized planting problem source text semantic cascade coding vector and the optimized planting problem semantic coding feature vector into a feature linear harmonization module to obtain the optimized planting problem semantic coding feature vector.

[0069] Specifically, the planting problem feature collaborative enhancement secondary sub-unit is used to: add the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem joint semantic representation vector; subtract the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem semantic deviation representation vector; perform positional point multiplication on the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector to obtain a planting problem interactive semantic representation vector; cascade the planting problem joint semantic representation vector, the planting problem semantic deviation representation vector and the planting problem interactive semantic representation vector to obtain a planting problem multimodal semantic representation vector, and perform point convolution on the planting problem multimodal semantic representation vector and then input it into a leakage corrected linear unit to obtain an implicit collaborative coding vector between planting problem features.

[0070] More specifically, the process of the above planting problem characteristics synergistically enhancing the secondary sub-units can be expressed as follows:

[0071]

[0072] in, represents the semantic cascade encoding vector of the planting problem source text, represents the semantic encoding feature vector of the planting problem, Indicates adding by position point. It means subtracting by position. It means point multiplication by position. represents a cascade function, represents point convolution, represents the leakage corrected linear unit, Represents the implicit co-encoding vector between planting problem features.

[0073] Among them, the process of inputting the semantic cascade coding vector of the planting problem source text and the semantic coding feature vector of the planting problem into the feature synergy enhancement network to obtain the implicit synergy coding vector between the planting problem features is a step of deep association mining and feature expression enhancement. This process goes beyond simple linear combinations and deeply explores the complex nonlinear relationship between the semantic cascade coding vector of the planting problem source text and the semantic coding feature vector of the planting problem, including direct correlations and those indirect or implicit relationships. For example, certain specific word combinations may have potential effects on crop growth conditions, and these subtleties can be captured by this network. The feature synergy enhancement network uses point convolution layers and activation functions (such as Leaky ReLU) to extract higher-level abstract features from the original features. This not only represents more complex patterns or concepts, but also helps the model better understand and predict planting-related problems. The network identifies and strengthens features that are critical to understanding planting problems, while suppressing irrelevant information, so that the generated implicit synergy coding vector is more focused on the core aspects of the problem, improving the effectiveness and accuracy of subsequent processing steps.

[0074] More specifically, the process of the above-mentioned adaptive memory processing secondary sub-unit can be expressed as follows:

[0075]

[0076] in, represents the implicit collaborative coding vector between the features of the planting problem, represents a 1×1 convolution operation, Represents the dynamic key vector of the planting problem.

[0077] Among them, the process of writing the implicit co-coding vector between the features of the planting problem into the adaptive memory module to obtain the dynamic key vector of the planting problem is a crucial step in the system. This process not only enhances the depth of understanding of the planting problem, but also provides more accurate and context-related input for subsequent processing. Specifically: when the implicit co-coding vector generated by the feature co-enhancement network is written into the adaptive memory module, a memory structure that can be dynamically adjusted according to the current input state is actually constructed. The adaptive memory module is similar to the memory unit in the recurrent neural network (RNN), but it has greater flexibility and adaptability. This means that it can adjust its own memory content based on the new feature vector input to better adapt to the specific needs of the current scene. The dynamic key vectors of the planting problem obtained after processing by the adaptive memory module represent the optimized memory content, which reflects the key features and semantic information of the current planting problem. These dynamic key vectors act as a bridge, connecting the original planting problem description with the subsequent tasks (such as problem solving), and help to achieve more accurate information retrieval and matching in the subsequent steps.

[0078] Among them, the planting problem interactive feature encoding secondary sub-unit is used to: determine the planting problem source text semantic query embedding matrix, the planting problem source text semantic query bias vector, the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector; based on the planting problem source text semantic query embedding matrix and the planting problem source text semantic query bias vector, perform a first linear transformation on the planting problem source text semantic cascade coding vector to obtain the planting problem source text semantic query vector; based on the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector, perform a second linear transformation on the planting problem source text semantic cascade coding vector to obtain the planting problem source text semantic value vector; input the planting problem dynamic key vector, the first linear transformation to obtain the planting problem source text semantic query vector and the planting problem source text semantic value vector into the interactive feature attention network based on the first converter structure to obtain the optimized planting problem source text semantic cascade coding vector.

[0079] More specifically, the process of interactive feature encoding of the secondary sub-units of the above planting problem can be expressed as follows:

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] in, represents the dynamic key vector of the planting problem, is the semantic query embedding matrix of the planting question source text, is the semantic query bias vector of the planting question source text, is the semantic query vector of the planting question source text, is the embedding matrix of the semantic value of the source text of the planting question, is the semantic value bias vector of the planting question source text, is the semantic value vector of the planting question source text, is the transposed vector of the dynamic key vector of the planting problem, is the length of the dynamic key vector of the planting problem, represents the normalized exponential function, represents matrix multiplication, represents the semantic concatenated encoding vector of the source text of the problem after optimization. is the planting question semantic query embedding matrix, is the planting question semantic query bias vector, is the planting question semantic query vector, is the planting question semantic value embedding matrix, is the planting problem semantic value bias vector, is the planting problem semantic value vector, Represents the semantic encoding feature vector of the optimized planting problem.

[0087] Among them, the process of extracting the dynamic key vector of the planting problem from the adaptive memory module and inputting the semantic cascade encoding vector of the planting problem source text and the planting problem semantic encoding feature vector into the interactive feature attention network for processing aims to further optimize and align these feature vectors, so as to provide higher quality input for the subsequent planting problem answering. This process first generates a dynamic key vector rich in contextual information through the adaptive memory module. This dynamic key vector acts as a bridge, which not only contains the key features and semantic information of the current planting problem, but also reflects the latest changes and development trends. Next, when we input the semantic cascade encoding vector of the planting problem source text and the planting problem semantic encoding feature vector together with these dynamic key vectors into the interactive feature attention network, we actually use the dynamic key vector as a common attention space to perform feature modulation based on query attention on the original feature vector. This modulation is not just a simple feature combination, but based on the information carried by the dynamic key vector, it enhances the key information in the original feature vector while reducing the influence of noise and irrelevant information. The core of the interactive feature attention network lies in its self-attention mechanism, which allows the model to consider the influence of all other elements when processing each input element, not just the adjacent elements. This global perspective helps capture long-distance dependencies, which is particularly important for understanding complex data. In this process, the network can identify which parts of the features are most critical to the understanding of the current planting problem and give them more weight. After such processing, the optimized planting problem source text semantic cascade encoding vector and the optimized planting problem semantic encoding feature vector are obtained. The two optimized vectors have been aligned in the same semantic space to ensure their consistency and complementarity, which means that both the specific details and general semantic features describing the planting problem have been effectively integrated together to form a more complete and accurate representation. This step greatly improves the efficiency and accuracy of subsequent tasks, enabling the system to provide farmers with more timely and effective planting advice and services based on the optimized feature vectors.

[0088] More specifically, the process of linearly harmonizing the secondary subunits of the above planting problem characteristics can be expressed as follows:

[0089]

[0090] in, represents the semantic concatenated encoding vector of the source text of the problem after optimization. represents the semantic encoding feature vector of the optimized planting problem, represents the weight hyperparameter, The semantically encoded feature vector representing the optimized planting problem.

[0091] Among them, the process of inputting the optimized planting problem source text semantic cascade encoding vector and the optimized planting problem semantic encoding feature vector into the feature linear reconciliation module to obtain the final optimized planting problem semantic encoding feature vector is a key step, which ensures that the features from different sources can be optimally integrated in the same semantic space, thereby providing high-quality and comprehensive input for subsequent tasks. This process aims to further integrate the two optimized feature vectors through linear transformation to generate a comprehensive representation that not only maintains the unique information of each feature vector but also reflects the synergy between the two. Specifically, the task of the feature linear reconciliation module is to adjust the weight parameters so that the importance of each input feature vector is appropriately reflected, while ensuring the semantic consistency and coherence of the final output feature vector. When the optimized planting problem source text semantic cascade encoding vector and the optimized planting problem semantic encoding feature vector are input into the feature linear reconciliation module, the module will balance the contribution of the two vectors according to the weight hyperparameters. This process is not just a simple weighted average, but a deeper level of feature interaction and fusion based on the rich semantic information obtained in the previous processing steps. For example, when analyzing a specific crop growth problem, the source text semantic cascade encoding vector may carry specific details about the farmer's description, while the semantic encoding feature vector provides a higher level of abstract understanding. Through the processing of the feature linear harmonization module, this information is seamlessly combined to form a more comprehensive and accurate feature representation. The resulting optimized planting problem semantic encoding feature vector not only retains the unique information of the original features, but also enhances their complementarity.

[0092] Specifically, the planting question answering unit 132 uses a planting question answering model based on the AIGC (Answer Identification and Generation for CQA) model to input the optimized planting question semantic coding feature vector to generate the answer text of the planting question. The AIGC model is a model specifically used for question-answering systems, which combines the two tasks of answer recognition and answer generation. In the answer recognition stage, the model will identify the key information in the planting question and the location of the answer. In the answer generation stage, the model will generate a specific answer text based on the question and related context information. The planting question answering unit uses the capabilities of the AIGC model to automatically identify the key information in the planting question based on the optimized planting question semantic coding feature vector and generate the corresponding answer text. In this way, automatic answers to planting questions can be achieved, and accurate and high-quality answer results can be provided. In this way, the automatic answer to farmers' planting problems can be achieved based on agricultural information big data, so as to provide farmers with answers and services to planting problems in a more timely and effective manner, and promote the improvement and development of agricultural production.

[0093] In summary, the agricultural information big data service system 100 based on the embodiment of the present application is explained, which can realize automatic answers to farmers' planting problems based on agricultural information big data, thereby providing farmers with answers and services to planting problems in a more timely and effective manner, and promoting the improvement and development of agricultural production.

[0094] As described above, the agricultural information big data service system 100 based on the embodiment of the present application can be implemented in various terminal devices, such as a server having an agricultural information big data service algorithm based on the embodiment of the present application. In one example, the agricultural information big data service system 100 based on the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the agricultural information big data service system 100 based on the embodiment of the present application can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the agricultural information big data service system 100 based on the embodiment of the present application can also be one of the many hardware modules of the terminal device.

[0095] Alternatively, in another example, the agricultural information big data service system 100 based on the embodiment of the present application and the terminal device may also be separate devices, and the agricultural information big data service system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0096] Figure 4 This is a flowchart of the agricultural information big data service method according to an embodiment of the present application. Figure 5 Schematic diagram of the system architecture of the agricultural information big data service method according to the embodiment of the present application. Figure 4 and Figure 5 As shown, according to the agricultural information big data service method of the embodiment of the present application, it includes: S110, obtaining a voice input of a planting problem input by a farmer; S120, performing voice recognition and understanding on the voice input of the planting problem to obtain semantic features of the planting problem; S130, generating a planting problem answer text based on the planting problem semantic features; and S140, converting the planting problem answer text into a voice signal and playing it through a speaker.

[0097] In a specific example, in the above-mentioned agricultural information big data service method, the planting problem voice input is subjected to voice recognition and understanding to obtain semantic features of the planting problem, including: performing voice recognition on the planting problem voice input to obtain a text description of the planting problem; segmenting the planting problem text description into sentences to obtain a sequence of planting problem description sentences; vectorizing and encoding the sequence of planting problem description sentences to obtain a sequence of planting problem description sentence vectors; and semantically encoding the sequence of planting problem description sentence vectors to obtain a planting problem semantic encoding feature vector as the planting problem semantic feature.

[0098] In a specific example, in the above-mentioned agricultural information big data service method, the sequence of the planting problem description sentences is vectorized and encoded to obtain a sequence of planting problem description sentence vectors, including: converting the sequence of the planting problem description sentences into a sequence of the planting problem description sentence vectors through a FastText model.

[0099] The specific operations of each step in the above agricultural information big data service method have been referenced above. Figures 1 to 4 The agricultural information big data service system 100 has been introduced in detail, and therefore, its repeated description will be omitted.

[0100] According to another aspect of the present application, a non-volatile computer-readable storage medium is provided, on which computer-readable instructions are stored. When the instructions are executed by a computer, the above-mentioned method can be executed.

[0101] The program part of the technology can be considered as a "product" or "manufactured product" in the form of executable code and / or related data, which is participated in or realized by computer-readable media. Tangible and permanent storage media can include any memory or storage used by any computer, processor, or similar device or related module. For example, various semiconductor memories, tape drives, disk drives or any similar devices that can provide storage functions for software.

[0102] The present application uses specific words to describe the embodiments of the present application. For example, "first / second embodiment", "one embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or multiple times in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0103] In addition, it will be appreciated by those skilled in the art that various aspects of the present application may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present application may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0104] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0105] The above is an explanation of the present invention and should not be considered as a limitation thereof. Although several exemplary embodiments of the present invention have been described, it will be readily appreciated by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined in the claims. It should be understood that the above is an explanation of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. Agricultural information big data service system, characterized by: include: A voice input module is used to obtain voice input of planting problems input by farmers; A planting question semantic understanding module, used for performing speech recognition and understanding on the planting question speech input to obtain the planting question semantic features; A planting problem answering text generation module, used for generating a planting problem answering text based on the planting problem semantic features; An answer voice playback module, used to convert the planting question answer text into a voice signal and play it through a speaker; The planting problem semantic understanding module includes: A question speech recognition unit, used for performing speech recognition on the planting question speech input to obtain a text description of the planting question; A sentence granularity division unit, used for dividing the text description of the planting problem into sentences to obtain a sequence of planting problem description sentences; A sentence granularity vectorized encoding unit, used for vectorizing and encoding the sequence of the planting problem description sentences to obtain a sequence of planting problem description sentence vectors; A planting problem semantic encoding unit, used for semantically encoding the sequence of the planting problem description sentence vectors to obtain a planting problem semantic encoding feature vector as the planting problem semantic feature; The sentence granularity vectorized encoding unit is used to: Converting the sequence of the planting problem description sentences into a sequence of the planting problem description sentence vectors through the FastText model; The planting problem semantic encoding unit is used to: Passing the sequence of the planting problem description sentence vectors through a transformer-based context encoder to obtain the planting problem semantic encoding feature vector; The planting problem answering text generation module comprises: A feature semantic residual compensation unit, used to perform residual compensation on the feature semantics of the context-related coding using the text semantics of the source text formed by the sequence of the planting problem description sentence vectors to obtain an optimized planting problem semantic coding feature vector; A planting question answering unit, used for passing the optimized planting question semantic coding feature vector through a planting question answering model based on an AIGC model to obtain a text of the planting question answering; The feature semantic residual compensation unit comprises: A source text semantic cascade subunit, used for cascading the sequence of the planting problem description sentence vectors to obtain a planting problem source text semantic cascade encoding vector; The planting problem dynamic memory enhancement subunit is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into the dynamic memory enhancement feature interaction module to obtain an optimized planting problem semantic coding feature vector; The planting problem dynamic memory enhancement subunit comprises: A planting problem feature collaborative enhancement secondary subunit is used to input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into a feature collaborative enhancement network to obtain an implicit collaborative coding vector between planting problem features; An adaptive memory processing secondary subunit is used to write the implicit collaborative coding vector between the planting problem features into the adaptive memory module to obtain a dynamic key vector of the planting problem; The planting problem interactive feature coding secondary subunit is used to extract the planting problem dynamic key vector from the adaptive memory module, and input the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector into the interactive feature attention network to obtain the optimized planting problem source text semantic cascade coding vector and the optimized planting problem semantic coding feature vector; The planting problem feature linear harmonization secondary subunit is used to input the optimized planting problem source text semantic cascade coding vector and the optimized planting problem semantic coding feature vector into the feature linear harmonization module to obtain the optimized planting problem semantic coding feature vector.

2. The agricultural information big data service system according to claim 1, characterized in that: The planting problem features synergistically enhance the secondary subunits for: Adding the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem joint semantic representation vector; Subtracting the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector by position to obtain a planting problem semantic deviation representation vector; Performing position point multiplication on the planting problem source text semantic cascade coding vector and the planting problem semantic coding feature vector to obtain a planting problem interactive semantic representation vector; The planting problem joint semantic representation vector, the planting problem semantic deviation representation vector and the planting problem interactive semantic representation vector are cascaded to obtain a planting problem multimodal semantic representation vector, and the planting problem multimodal semantic representation vector is point convolved and then input into a leakage corrected linear unit to obtain an implicit collaborative coding vector between planting problem features.

3. The agricultural information big data service system according to claim 2, characterized in that: The planting problem interactive feature encoding secondary subunit is used to: Determine the planting problem source text semantic query embedding matrix, the planting problem source text semantic query bias vector, the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector; Based on the planting problem source text semantic query embedding matrix and the planting problem source text semantic query bias vector, performing a first linear transformation on the planting problem source text semantic cascade encoding vector to obtain a planting problem source text semantic query vector; Based on the planting problem source text semantic value embedding matrix and the planting problem source text semantic value bias vector, performing a second linear transformation on the planting problem source text semantic cascade encoding vector to obtain a planting problem source text semantic value vector; The planting question dynamic key vector, the first linear transformation to obtain the planting question source text semantic query vector and the planting question source text semantic value vector are input into an interactive feature attention network based on the first converter structure to obtain an optimized planting question source text semantic cascade encoding vector.

Images