CN118014519A

CN118014519A - Digital rural application system and method based on Internet of things

Info

Publication number: CN118014519A
Application number: CN202410188171.5A
Authority: CN
Inventors: 陈小芬
Original assignee: Jiangxi Zhishi Technology Co ltd
Current assignee: Jiangxi Zhishi Technology Co ltd
Priority date: 2024-02-20
Filing date: 2024-02-20
Publication date: 2024-05-10

Abstract

The application relates to the technical field of the Internet of things, and particularly discloses a digital village application system and method based on the Internet of things. Therefore, accurate farmland irrigation management can be provided, the farmer is helped to make a decision, and the growth effect of crops is improved.

Description

Digital rural application system and method based on Internet of things

Technical Field

The application relates to the technical field of the Internet of things, in particular to a digital rural application system and method based on the Internet of things.

Background

The digital village is taken as a rural infrastructure digital transformation project, information is taken as a processing object, a digital technology is taken as a processing means, conscious (generalized conscious concept) products are taken as achievements, all fields of the whole intervening society are taken as markets, and the digital village is a public industry which can promote profits of other industries. Its informatization and intelligent management are very important in the whole project. Specifically, population monitoring, environment monitoring, passenger flow monitoring, agricultural monitoring, land monitoring, energy consumption monitoring, health detection and the like in the rural area belong to the management range.

The existing intelligent agricultural system generally performs single analysis on the moisture or fertilizer of soil and lacks judgment on the growth state of crops; for example, in the patent with application publication No. CN111399508a, an intelligent agricultural system and an intelligent agricultural method are disclosed, the scheme is to analyze soil information and crop information, determine the fertilizing amount of each area, lack the analysis of the growth state of crops after fertilization, and for example, in the patent with application publication No. CN111507857a, a digital agricultural planting system and a method based on the internet of things technology are disclosed, the method realizes intelligent agricultural production by precisely controlling agricultural facilities through robots, saves manpower, reduces human error, enables users to remotely control the farmland planting environment on line, and precisely manages crops, but the method lacks the judgment of the requirement on the state of the crop growth process, so that the existing digital rural management system is necessary to be optimized.

Accordingly, a digital rural application system and method based on the internet of things is desired.

Disclosure of Invention

The present application has been made to solve the above-mentioned technical problems. The embodiment of the application provides a digital rural application system and method based on the Internet of things, which are used for judging whether irrigation is needed or not by utilizing a soil humidity sensor, weather information and crop images through feature fusion and a classifier, so as to realize intelligent farmland irrigation management.

Accordingly, according to one aspect of the present application, there is provided a digital rural application system based on the internet of things, comprising:

the system comprises an Internet of things information acquisition module, a control module and a control module, wherein the Internet of things information acquisition module is used for acquiring soil humidity values of a plurality of preset time points in a preset time period through a soil humidity sensor deployed in soil, acquiring weather information and acquiring crop images through an unmanned aerial vehicle;

The internet of things information processing module is used for extracting soil humidity feature vectors, weather information semantic feature vectors and crop semantic association feature vectors from the soil humidity values, the weather information and the crop images at a plurality of preset time points respectively;

The internet of things information fusion module is used for constructing an irrigation analysis feature vector among the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, and carrying out probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector;

and the information analysis module of the Internet of things is used for enabling the optimized irrigation analysis feature vector to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether irrigation is needed or not.

According to another aspect of the present application, there is provided a digital rural application method based on the internet of things, including:

Acquiring soil humidity values at a plurality of preset time points in a preset time period through a soil humidity sensor arranged in soil, acquiring weather information and acquiring crop images through an unmanned aerial vehicle;

Extracting soil humidity feature vectors, weather information semantic feature vectors and crop semantic association feature vectors from the soil humidity values, the weather information and the crop images at the plurality of predetermined time points respectively;

constructing an irrigation analysis feature vector among the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, and performing probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector;

And the optimized irrigation analysis feature vector is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether irrigation is needed or not.

Compared with the prior art, the digital rural application system and method based on the Internet of things provided by the application have the advantages that the soil humidity sensor is deployed, the crop images are collected, the weather information is combined, the soil humidity characteristics, the weather information semantic characteristics and the crop semantic association characteristics are extracted, irrigation analysis is carried out by fusing the characteristics, and the irrigation analysis characteristics are classified by the classifier so as to determine whether irrigation is needed. Therefore, accurate farmland irrigation management can be provided, the farmer is helped to make a decision, and the growth effect of crops is improved.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing embodiments of the present application in more detail with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and together with the embodiments of the application, and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

Fig. 1 is a block diagram of a digital rural application system based on the internet of things according to an embodiment of the present application.

Fig. 2 is a schematic block diagram of an information processing module of the internet of things in a digital rural application system based on the internet of things according to an embodiment of the present application.

Fig. 3 is a schematic block diagram of a soil moisture extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application.

Fig. 4 is a schematic block diagram of an weather information feature extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application.

Fig. 5 is a schematic block diagram of a crop image feature extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application.

Fig. 6 is a flowchart of a digital rural application method based on the internet of things according to an embodiment of the present application.

Detailed Description

Various exemplary embodiments, features and aspects of the application will be described in detail below with reference to the drawings. In the drawings, like reference numbers indicate identical or functionally similar elements. Although various aspects of the embodiments are illustrated in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better illustration of the application. It will be understood by those skilled in the art that the present application may be practiced without some of these specific details. In some instances, well known methods, procedures, components, and circuits have not been described in detail so as not to obscure the present application.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Fig. 1 illustrates a block diagram schematic of an internet of things based digital rural application according to an embodiment of the present application. As shown in fig. 1, a digital rural application system 100 based on the internet of things according to an embodiment of the present application includes: the internet of things information acquisition module 110 is configured to acquire soil humidity values at a plurality of predetermined time points in a predetermined time period through a soil humidity sensor deployed in soil, acquire weather information, and acquire crop images through an unmanned aerial vehicle; the internet of things information processing module 120 is configured to extract a soil humidity feature vector, a weather information semantic feature vector and a crop semantic association feature vector from the soil humidity values, the weather information and the crop image at the plurality of predetermined time points, respectively; the internet of things information fusion module 130 is configured to construct an irrigation analysis feature vector among the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, and perform probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector; the internet of things information analysis module 140 is configured to pass the optimized irrigation analysis feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether irrigation is needed.

In the embodiment of the present application, the internet of things information acquisition module 110 is configured to acquire soil humidity values at a plurality of predetermined time points in a predetermined time period through a soil humidity sensor deployed in soil, acquire weather information, and acquire crop images through an unmanned aerial vehicle. It will be appreciated that the purpose of acquiring soil moisture values at a plurality of predetermined time points over a predetermined period of time by means of a soil moisture sensor deployed in the soil is to monitor changes in soil moisture in order to determine whether irrigation is required. The soil humidity sensor can acquire soil humidity data in real time, and the change trend of the soil humidity can be obtained by comparing humidity values of different time points, so that whether irrigation is needed or not is determined. The purpose of the weather information is to take into account the influence of external environmental factors on the soil humidity. Weather information may provide rainfall, temperature, humidity, etc. data that may help determine whether there is sufficient rainfall to meet the water demand of the plant and thus determine whether irrigation is needed. The purpose of collecting crop images by unmanned aerial vehicle is to obtain the growth condition of plants. The crop image can provide information such as growth conditions, leaf colors, plant diseases and insect pests and the like of the plants, and the information can be used for judging the health state and the water demand of the plants so as to determine whether irrigation is needed. The unmanned aerial vehicle can acquire crop images in a large area efficiently, and can analyze and identify the crop images through an image processing algorithm, so that decision support is provided.

In the embodiment of the present application, the internet of things information processing module 120 is configured to extract a soil humidity feature vector, a weather information semantic feature vector and a crop semantic association feature vector from the soil humidity values, the weather information and the crop image at the plurality of predetermined time points, respectively. It should be appreciated that the purpose of extracting the soil moisture profile is to convert soil moisture values at various points in time into a comprehensive representation to describe the trend of the soil moisture and the moisture status. The feature vector can comprise indexes such as average humidity, maximum humidity, minimum humidity, humidity change rate and the like, and can reflect the overall state and the change degree of the soil humidity. The purpose of extracting the semantic feature vector of the weather information is to convert the weather data into a semantically representative to describe the features and effects of the weather. The feature vector can comprise indexes such as temperature, humidity, rainfall, wind speed and the like, and can reflect the type, strength and stability of weather, so that the change of soil humidity is interpreted and predicted. The purpose of extracting the crop semantic association feature vector is to convert the information in the crop image into a vector representation to describe the growth state and health of the crop. The characteristic vector can comprise vegetation indexes, leaf colors, plant diseases and insect pests indexes and the like, and can reflect the growth degree, damage degree and water and fertilizer requirement condition of crops, so that the plant diseases and insect pests can be subjected to correlation analysis with soil humidity and weather information. By extracting the feature vectors, complex data can be converted into compact numerical representation, and comparison, analysis and decision making are facilitated. The feature vectors can be used for constructing models, performing tasks such as data mining and machine learning, and helping farmers, agricultural specialists and decision makers to make reasonable decisions such as irrigation scheduling, pest control and the like.

Specifically, in one embodiment of the present application, fig. 2 illustrates a schematic block diagram of an information processing module of the internet of things in the digital rural application system based on the internet of things according to the embodiment of the present application. As shown in fig. 2, in the above-mentioned digital rural application system 100 based on the internet of things, the information processing module 120 of the internet of things includes: a soil moisture feature extraction unit 121 for obtaining the soil moisture feature vector by feature extraction after arranging the soil moisture values of the plurality of predetermined time points into a vector; a weather information feature extraction unit 122 for obtaining the weather information semantic feature vector from the weather information by convolutional encoding; a crop image feature extraction unit 123, configured to encode the crop image to obtain the crop semantic association feature vector.

Specifically, the soil moisture characteristic extraction unit 121 is configured to obtain the soil moisture characteristic vector by feature extraction after arranging the soil moisture values at the plurality of predetermined time points into a vector. It will be appreciated that after arranging the soil moisture values at a plurality of time points into vectors, these values can be combined to provide a more comprehensive and comprehensive representation. This captures the overall trend and change in soil moisture, not just a single point in time value. By arranging soil moisture values at a plurality of time points into a vector, the original data set can be reduced to one vector. Therefore, the dimension of the data can be reduced, and the subsequent analysis and processing are convenient. At the same time, the form of the vector is also more suitable for many applications of machine learning and statistical methods. By feature extraction of the soil moisture value vector, more meaningful and useful features can be extracted. These features may be statistical indicators (e.g., mean, variance, etc.), or domain knowledge defined features (e.g., drought index, humidity rate of change, etc.). Therefore, key characteristics of soil humidity can be further extracted, and noise and redundant information can be removed. By obtaining the soil moisture characteristic vector, comparison and analysis can be conveniently performed. The characteristic vectors of the soil humidity at different time points can be compared, and the variation trend and the periodicity of the humidity can be observed. Soil moisture feature vectors of different sites or different crops can be compared, and differences and correlations between the soil moisture feature vectors can be analyzed. Thus, the method can help the work of agricultural management, decision making, pest control and the like.

Further, fig. 3 illustrates a block diagram schematic diagram of a soil moisture extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application. As shown in fig. 3, in the internet of things information processing module 120 of the digital rural application system 100 based on the internet of things, the soil moisture feature extraction unit 121 includes: a humidity value arrangement vector subunit 1211 for arranging the soil humidity values at the plurality of predetermined time points as a soil humidity input vector; a multiscale feature extraction subunit 1212 is configured to pass the soil moisture input vector through a moisture extractor based on a multiscale neighborhood feature extraction module to obtain the soil moisture feature vector.

Specifically, the humidity value arrangement vector subunit 1211 is configured to arrange the soil humidity values at the plurality of predetermined time points into a soil humidity input vector. It will be appreciated that the change in soil moisture values is generally time dependent, and that arranging them into an input vector may preserve the time relationship. By encoding the time points as dimensions of vectors or by time intervals, the model can learn the time dependence of soil humidity, capturing the trend and periodicity of humidity over time. After the soil humidity values at a plurality of time points are arranged as the input vector, the feature extraction can be conveniently performed. In time series data, various statistical indicators (e.g., mean, variance, maximum, minimum, etc.) and other features (e.g., trends, periodic features, etc.) may be extracted from different points in time. These features can be used for training and prediction of models. Many machine learning and statistical analysis models are trained and predicted based on vector-form inputs. Arranging soil moisture values at multiple points in time as input vectors may make these models easier to understand and process soil moisture data. After the soil humidity values at a plurality of time points are arranged as input vectors, the data at different time points can be conveniently integrated into a matrix or a data frame for batch processing and analysis. Thus, preprocessing steps such as data cleaning, missing value processing, feature selection and the like can be performed more efficiently.

Specifically, the multi-scale feature extraction subunit 1212 is configured to pass the soil moisture input vector through a moisture extractor based on a multi-scale neighborhood feature extraction module to obtain the soil moisture feature vector. It should be appreciated that variations in soil moisture may exist in patterns on different scales and time scales. By using a multi-scale neighborhood feature extraction module, features of soil moisture can be captured from different scales. Thus, the humidity change trend and the correlation under different time scales can be obtained, so that the characteristics of the soil humidity can be more comprehensively described. The humidity extractor based on the multi-scale neighborhood feature extraction module can extract local and global features at the same time. Local features represent local changes in soil humidity on a certain time scale, such as humidity fluctuations over a short period of time; global characteristics represent trends and changes in soil humidity over time, such as long-term humidity trends. Thus, different aspects of soil humidity can be comprehensively considered, and the characteristic expression capability is improved. By the humidity extractor based on the multi-scale neighborhood feature extraction module, nonlinear transformation and feature enhancement can be performed on the input soil humidity input vector. Thus, the characteristics with more discrimination and expression capability can be extracted, and different types of soil humidity can be better distinguished. For example, a Convolutional Neural Network (CNN) or a Recurrent Neural Network (RNN) model may be used to extract features and learn higher-level abstract features through a deep learning method. By a humidity extractor based on a multi-scale neighborhood feature extraction module, the original soil humidity input vector can be converted into a more compact and high-dimensional feature vector. Therefore, the dimension of the features can be reduced, the complexity of the data is reduced, and the efficiency and the robustness of the subsequent model are improved.

Specifically, the multi-scale feature extraction subunit comprises: a first scale humidity coding secondary subunit, configured to perform one-dimensional convolution coding on the soil humidity input vector with a one-dimensional convolution kernel having a first scale by using a first convolution layer of the humidity extractor based on the multi-scale neighborhood feature extraction module to obtain a first scale humidity feature vector, where the first convolution layer has a first one-dimensional convolution kernel having a first length; a second scale humidity encoding secondary subunit, configured to perform one-dimensional convolution encoding on the soil humidity input vector with a one-dimensional convolution kernel having a second scale using a second convolution layer of the humidity extractor based on the multi-scale neighborhood feature extraction module to obtain a second scale humidity feature vector, where the second convolution layer has a second one-dimensional convolution kernel having a second length, and the first length is different from the second length; and the multi-scale humidity cascade secondary subunit is used for cascading the first-scale humidity characteristic vector and the second-scale humidity characteristic vector to obtain the soil humidity characteristic vector.

Accordingly, in a specific example of the present application, the weather information feature extraction unit 122 is configured to obtain the weather information semantic feature vector from the weather information through convolutional encoding. It should be appreciated that weather information typically contains data for a number of aspects, such as temperature, humidity, wind speed, etc. Features with semantic meaning can be extracted from the original weather information through convolutional coding. Convolutional Neural Networks (CNNs) are an efficient feature extractor that can capture local patterns and features in data through convolutional operations. By applying convolutional coding, important features in weather information, such as trend of temperature, change of humidity, etc., can be extracted. There may be some correlation and dependency between different features in the weather information. Convolutional encoding may capture local correlations between features through a sliding window operation of a convolutional kernel. The interaction between different features can be better understood, and some important modes and rules in weather information are captured. Weather information typically contains a large number of dimensions, such as temperature, humidity, etc. data per hour. Through convolution coding, the original weather information can be subjected to dimension reduction and compression, and a more compact semantic feature vector is obtained. Therefore, the dimension of the features can be reduced, the complexity of the data is reduced, and the efficiency and the robustness of the subsequent model are improved. Convolutional coding can learn from a large-scale dataset to a generic representation of features by way of pre-training. Such feature representations may be migrated to other tasks, including semantic feature extraction of weather information. By using a pre-trained convolutional coding model, the existing knowledge and modes can be fully utilized, and the characteristic expression capability of weather information is improved.

Further, fig. 4 illustrates a block diagram of an weather information feature extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application. As shown in fig. 4, in the internet of things information processing module 120 of the digital rural application system 100 based on the internet of things, the weather information feature extraction unit 122 includes: a semantic understanding sub-unit 1221 for obtaining weather information feature vectors from the weather information using a semantic understanding model; a binarization subunit 1222 for binarizing the weather information based on conditions to obtain a weather information binarized feature vector; a vector multiplication subunit 1223 for multiplying the weather information feature vector by a transpose of the weather information binarized feature vector to obtain a weather information feature matrix for expressing the weather information; a convolutional encoding subunit 1224, configured to input the weather information feature matrix into a weather information extractor based on the first convolutional neural network model to obtain the weather information semantic feature vector.

Specifically, the semantic understanding subunit 1221 is configured to obtain a weather information feature vector from the weather information using a semantic understanding model. It should be understood that the weather information includes abundant semantic information, such as weather status (sunny day, cloudy day, rainy day, etc.), temperature range (high temperature, low temperature, etc.), humidity level, etc. The semantic understanding model may help understand and interpret these semantic information, thereby better expressing the meaning of the weather information. Weather information can be converted into feature vectors with certain semantic meanings through a semantic understanding model. Weather information is typically context-dependent, with respect to geographic location, time, season, and the like. The semantic understanding model may correlate weather information with related semantic information by modeling context information. This allows a better understanding of the meaning and underlying semantic relationships of weather information and encodes it into feature vectors. The semantic understanding model may learn abstract features in the weather information and is not limited to the original observations. Through a deep learning method, the semantic understanding model can extract abstract features with higher level including features in aspects of semantics, grammar, context and the like through multi-level feature extraction and representation learning. These abstract features may better capture the meaning and relevance of weather information. Weather information may have incomplete data, such as missing values, noise, etc. The semantic understanding model can fill in missing values, correct noise and the like by modeling and reasoning weather information, so that more complete and accurate characteristic representation is obtained. Thus, the accuracy and the robustness of the subsequent tasks can be improved.

Accordingly, the semantic understanding subunit comprises: the word vector conversion secondary subunit is used for mapping each word in the weather information into a word vector by using a word embedding layer of the semantic understanding model so as to obtain a word vector sequence of the weather information; the semantic processing secondary subunit is used for processing the weather information word vector sequence by using a Bert model of the semantic understanding model to obtain a weather information word feature vector sequence; and the context coding secondary subunit is used for performing context coding on the weather information word feature vector sequence by using the bidirectional LSTM network of the semantic understanding model so as to obtain the weather information feature vector.

Specifically, the binarizing subunit 1222 is configured to binarize the weather information based on conditions to obtain a weather information binarized feature vector. It should be appreciated that weather information typically contains data in multiple dimensions, such as temperature, humidity, wind speed, and the like. Binarizing these successive values may simplify it into a more compact representation. Binarization converts the value of each dimension into two discrete values, such as high/low temperature, wet/dry, windy/windless, etc. This can greatly reduce the dimensions of the feature space, making the feature vector more compact and easy to handle. In weather information, certain features may be more important for a particular task or application. By condition-based binarization, these key features can be highlighted and retained, while other less important information is ignored. For example, for some applications, it may be more critical to only focus on whether it is raining, blowing, etc., and the particular temperature and humidity values may not be important. Different tasks and applications may have different demands on weather information. By condition-based binarization, feature vectors can be tailored to specific requirements. Specifically, binarized conditions are set, which may be determined based on thresholds, ranges, or other rules. The continuous values are mapped to discrete values, such as 0 and 1. For example, if the temperature is greater than a certain threshold (e.g., 30 degrees celsius), it is marked as high temperature (1), otherwise it is marked as low temperature (0). Of course, the choice of threshold may vary depending on the region, season and specific application.

Specifically, the vector multiplication subunit 1223 is configured to multiply the weather information feature vector by a transpose of the weather information binarized feature vector to obtain a weather information feature matrix for expressing the weather information. It should be appreciated that continuous weather information is converted into discrete feature representations for better analysis and processing. By binarizing the weather information, the continuous features can be converted into binary discrete values, and the relationship between the features can be captured better. Multiplying the weather information feature vector by the transpose of the weather information binarized feature vector can provide a more compact, easier to process and analyze weather information feature matrix by converting continuous weather information into discrete feature representations. Such feature matrices may help us better understand relationships and patterns in weather data and support various weather information analysis tasks.

Specifically, the convolutional encoding subunit 1224 is configured to input the weather information feature matrix into a weather information extractor based on a first convolutional neural network model to obtain the weather information semantic feature vector. It should be appreciated that inputting the weather information feature matrix into the weather information extractor based on the first convolutional neural network model may help us obtain semantic feature vectors of the weather information. According to the method, through the feature extraction capability of the convolutional neural network, abstract representations with higher levels can be learned from the weather information feature matrix, and richer semantic information can be captured. In particular, convolutional neural networks (Convolutional Neural Network, CNN) are a type of neural network model that is widely used for image processing and pattern recognition tasks. The method performs feature extraction and classification on input data in an end-to-end manner through components such as a convolution layer, a pooling layer, a full connection layer and the like. In the weather information processing, we can consider the weather information feature matrix as a two-dimensional image data in which each element represents a relationship between features. By inputting the weather information feature matrix into the CNN-based weather information extractor, the network can learn the spatial and structural features in the weather information. Through the convolution layer, the network can automatically learn the local mode and the feature in the weather information feature matrix. The convolution operation can share the weight at different positions, so that the parameter quantity of the network is reduced, and the efficiency of feature extraction is improved. The pooling layer may downsample the feature map, preserve important features, and reduce the size of the feature map. The full connection layer can map the extracted features to semantic space to obtain semantic feature vectors of weather information. Thus, by inputting the weather information feature matrix into the CNN-based weather information extractor, we can extract feature vectors with more semantic meaning from the original weather information. The semantic feature vectors can be used for tasks such as classification, prediction and similarity calculation of weather information, and the analysis and application capacity of the weather data are improved.

Accordingly, in a specific example of the present application, the crop image feature extraction unit 123 is configured to encode the crop image to obtain the crop semantic association feature vector. It should be appreciated that by encoding the crop image, we can use feature extraction methods in deep learning, such as Convolutional Neural Networks (CNNs) or pre-trained visual models, to convert the image into high-dimensional feature vectors. These feature vectors capture semantic information in the image, such as visual features of shape, texture, color, etc. of the object. Encoding may convert high-dimensional image data into low-dimensional feature vectors. Thus, the dimension of the data can be reduced, redundant information is removed, and the feature vector is more compact and efficient. The coding method utilizes a deep learning model, so that semantic information in an image can be learned. These feature vectors can better represent key features in the crop image, such as variety, growth status, disease condition, etc. of the crop. The coding method can learn the feature representation with more generalization capability. This means that the feature vectors obtained by encoding can be effectively applied to different crop image datasets, not just to datasets used in training.

Further, fig. 5 illustrates a block diagram schematic diagram of a crop image feature extraction unit in a digital rural application system based on the internet of things according to an embodiment of the present application. As shown in fig. 5, in the internet of things information processing module 120 of the digital rural application system 100 based on the internet of things, the crop image feature extraction unit 123 includes: a gray image subunit 1231 configured to convert the crop image into a gray image to obtain a crop gray image; a gray image corrector unit 1232 for correcting the crop gray image to obtain a crop gray corrected image; a gray co-occurrence matrix subunit 1233 configured to obtain a plurality of crop gray feature statistics from the crop gray corrected image based on a gray co-occurrence matrix; a context encoding subunit 1234 configured to pass the plurality of crop gray feature statistics through a context encoder comprising an embedded layer to obtain a plurality of crop gray feature statistical semantic feature vectors; and the bidirectional long-short term subunit 1235 is configured to obtain the crop semantic association feature vector through a bidirectional long-short term memory neural network model after one-dimensional arrangement of the plurality of crop gray feature statistical semantic feature vectors.

Specifically, the grayscale image subunit 1231 is configured to convert the crop image into a grayscale image to obtain a crop grayscale image. It will be appreciated that the greyscale image contains only one colour channel, with less computational complexity than the three channels of the colour image. This may increase computational efficiency for some computationally intensive image processing tasks, such as feature extraction, classification, and detection. In some cases, the color of the crop image may be affected by factors such as lighting conditions, camera settings, etc., resulting in unstable color information. Converting the image into a gray scale image can remove interference of colors, so that subsequent image processing is more stable and reliable. The gray image mainly contains brightness information, and can better highlight the texture and shape characteristics of the crop image. These features are important for the tasks of crop identification, disease detection, growth state assessment, etc. The storage space and transmission overhead of gray scale images are smaller than for color images. This may reduce the cost of storage and transmission for resource-constrained environments or applications requiring large-scale processing of crop image data.

Specifically, the gray-scale image corrector unit 1232 is configured to correct the crop gray-scale image to obtain a crop gray-scale corrected image. It will be appreciated that correcting the gray scale image may adjust the brightness and contrast of the image so that details in the image are more clearly visible. Through correction, the brightness range of the crop gray level image is more suitable, the condition of over darkness or over brightness is avoided, and the visual effect of the image is improved. The gray scale image of a crop may be affected by ambient light, camera noise, etc., resulting in some background noise in the image. By correction, these noises can be reduced or removed, making the main information of the crop more prominent. Correcting gray scale images may result in consistent brightness and contrast between different images. This may provide a more stable and reliable input for subsequent image processing tasks such as feature extraction, classification, and detection. Correcting the gray scale image can improve the accuracy of subsequent crop image analysis tasks. For example, if tasks such as disease detection or leaf area measurement are to be performed, the corrected image can more accurately reflect the actual state of the crop, and the reliability of the analysis result is improved.

Specifically, the gray level co-occurrence matrix subunit 1233 is configured to obtain a plurality of crop gray level feature statistics from the crop gray level correction image based on the gray level co-occurrence matrix. It should be appreciated that gray level co-occurrence matrix is a statistical method for describing the spatial relationship between different gray levels in an image. By counting the frequency of occurrence and the positional relationship of gray values between each pair of pixels, a co-occurrence matrix between gray levels can be obtained. This matrix provides probability and relative position information of the occurrence of different grey levels in the image, which can be used to characterize the grey scale distribution of the crop. The gray level co-occurrence matrix may reflect texture information in the image. Texture features between different gray levels can be obtained by calculating statistical features of the co-occurrence matrix, such as Contrast (Contrast), uniformity (Homogeneity), entropy (Entropy), etc. These features may be used to describe texture complexity, granularity, smoothness, etc. information of the crop image. Feature statistics based on gray level co-occurrence matrices can be used to distinguish different classes of crops. The gray distribution and the texture features of different crops are usually different, so that the feature vector capable of effectively distinguishing different crops can be extracted by calculating the features of the gray co-occurrence matrix. These features may be used as inputs to a classifier or machine learning model to enable automatic identification and classification of crops.

Specifically, the context encoding subunit 1234 is configured to pass the plurality of crop gray feature statistics through a context encoder that includes an embedded layer to obtain a plurality of crop gray feature statistical semantic feature vectors. It should be appreciated that the plurality of crop gray scale characteristic statistics are typically represented in a matrix or vector form, with a relatively high dimension. By using an embedded layer, high-dimensional features can be mapped into a low-dimensional space, thereby achieving a reduction in dimensions. This helps to reduce the storage space and computational complexity of the feature vectors and improves the efficiency of subsequent tasks. The function of the embedding layer is to translate the original features into representations in semantic space. By inputting a plurality of crop grayscale feature statistics into the context encoder, semantic relationships and interactions between features can be learned. The embedding layer may encode the relevance and importance between different features as distance and direction relationships in the semantic feature vector. The plurality of crop gray scale feature statistics comprise information of different aspects of the crop image, such as gray scale distribution, texture features, and the like. By inputting these features into the context encoder, fusion and integration of the features can be achieved. The context encoder may learn interactions and dependencies between features to generate more representative and comprehensive semantic feature vectors. The context encoder is typically part of a deep learning model, and abstract features can be extracted by multi-layer nonlinear transformations. These abstract features may capture higher level semantic information and features in the crop image. By inputting a plurality of crop gray scale feature statistics into the context encoder, higher level semantic features can be progressively extracted, thereby better expressing the features and semantic meaning of the crop image.

Specifically, the bidirectional long-short term subunit 1235 is configured to obtain the crop semantic association feature vector by performing one-dimensional arrangement on the plurality of crop gray feature statistical semantic feature vectors and then passing through a bidirectional long-short term memory neural network model. It should be understood that a plurality of crop gray scale feature statistical semantic feature vectors can be regarded as a sequence after one-dimensional arrangement. The sequence modeling mode can capture the time sequence relation and the dependency relation between crop characteristics. Through the two-way long-short-term memory neural network model, the forward and backward context information can be considered at the same time, so that semantic association between crop features can be better understood. The two-way long-short-term memory neural network model can effectively capture the context information in the sequence data. By introducing a bi-directional structure in the model, feature vectors before and after the current moment can be taken into account simultaneously. By doing so, the relationships between crop features can be more fully understood, and feature vectors with more semantic relevance can be generated. Long-term memory (LSTM) is a Recurrent Neural Network (RNN) model suitable for processing sequence data, with memory units to store and update information. By using LSTM cells, the two-way long and short term memory neural network model can effectively handle long term dependencies, i.e., relationships between locations farther in the sequence. This is very important for modeling the semantic relevance of crop features. The two-way long-short term memory neural network model can be used as a feature extractor to learn higher-level semantic features from an input sequence. By inputting a plurality of crop gray feature statistical semantic feature vectors into the two-way long-short-term memory neural network model, crop feature representation with better expressive force and semantic relevance can be obtained, so that subsequent crop analysis and processing tasks are better supported.

Accordingly, the bidirectional long-short term subunit comprises: a one-dimensional arrangement secondary subunit, configured to perform one-dimensional arrangement on the plurality of crop gray feature statistical semantic feature vectors to obtain a sequence of crop gray feature statistical semantic feature vectors; and the sequence coding secondary subunit is used for carrying out context semantic coding on the sequence of the crop gray feature statistical semantic feature vector by using the two-way long-short-term memory neural network model so as to obtain the crop semantic association feature vector.

In the embodiment of the present application, the information fusion module 130 of the internet of things is configured to construct an irrigation analysis feature vector among the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, and perform probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector. It should be appreciated that soil moisture, weather information and crop characteristics are all important factors affecting irrigation requirements. The information of various kinds can be comprehensively considered by fusing the information, so that a more comprehensive and accurate irrigation analysis characteristic vector is obtained. The soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector come from different data sources and angles and have different information contents. Fusing them together can make full use of the complementarity between them, providing more abundant and diversified feature information. By fusing a plurality of feature vectors, the predictive performance of irrigation requirements can be improved. The combination of different features can provide more information, helping the model to better understand and capture patterns and laws of irrigation needs. The fusion feature vector can combine a plurality of features into one feature vector, so that the dimension of the features is reduced. This helps to simplify the input and computation of the model, improving the efficiency and interpretability of the model.

In particular, in the technical scheme of the application, the soil humidity characteristic vector is constructed according to soil humidity values of a plurality of preset time points in a preset time period. These humidity values are arranged in a time dimension as soil humidity input vectors and processed by a humidity extractor based on a multi-scale neighborhood feature extraction module. The method and data representation used in this process is different from other feature extraction modules, so the resulting feature vectors may have different dimensions. The weather information semantic feature vector is extracted from the weather information through a semantic understanding model. Weather information may include various meteorological parameters such as temperature, humidity, wind speed, etc. The semantic understanding model processes the weather information to obtain feature vectors related to weather semantics. Since the dimensions of weather information may be different from those of soil moisture feature vectors and crop semantic association feature vectors, there is a case where there is a dimensional misalignment between them. The crop semantic association feature vector is obtained by processing a crop image. First, a crop image is converted into a gray-scale image, and then correction is performed to obtain a crop gray-scale correction image. Next, a plurality of crop gray-scale feature statistics are derived from the crop gray-scale corrected image based on the gray-scale co-occurrence matrix. The feature statistics are processed by a context encoder comprising an embedded layer to obtain a crop gray feature statistical semantic feature vector. Since the dimension of the crop gray scale feature statistical semantic feature vector may be different from the dimension of other feature vectors, there is a case that the dimensions are not aligned. When the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector are fused to obtain an irrigation analysis feature vector, the feature vector obtained by fusion may have an out-of-distribution feature value due to misalignment of the dimensions of the feature vector. This is because misalignment of the dimensions during fusion may result in some portions of the feature vector not being properly aligned, thereby introducing irrelevant or erroneous information that can affect the quality of the irrigation analysis feature vector. To address this problem, probability density domain-dependent migration hyperconcentration projection metrics are performed on the irrigation analysis feature vectors.

Accordingly, in one embodiment of the present application, the information fusion module 130 of the internet of things includes: the fusion unit is used for fusing the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector to obtain an irrigation analysis feature vector; the optimizing unit is used for carrying out probability density domain related migration super convex projection measurement on the irrigation analysis feature vector so as to obtain an optimized irrigation analysis feature vector;

wherein, the optimizing unit is used for: and carrying out probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector according to the following formula:

Wherein V represents the irrigation analysis feature vector, V _i represents the feature value of the irrigation analysis feature vector, softmax represents the normalized exponential function, log represents the logarithmic function value based on 2, β represents the hyper-parameter, and V _i' represents the feature value of the optimized irrigation analysis feature vector.

That is, in the process of obtaining the irrigation analysis feature vector by fusing the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, there is a dimensional misalignment between the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, which results in the fused irrigation analysis feature vector having an out-of-distribution feature value.

According to the technical scheme, probability density domain related migration super-convex projection measurement is carried out on the irrigation analysis feature vector, the irrigation analysis feature vector is mapped to a super-convex space from an original space through a probability density domain related migration super-convex projection measurement function, feature vectors with different data densities have the same dimension and distribution in the super-convex space, and the problem that the dimension between the global feature vector of a tourist attraction and the related feature vector of a point cloud coordinate is not aligned and distributed inconsistently can be eliminated through super-convex space mapping, so that the irrigation analysis feature vector obtained through fusion does not have an out-of-distribution feature value, but has more compact and robust feature representation. The method comprises the steps of determining a probability density domain related migration super convex projection metric function, and determining the characteristic vector of the irrigation analysis by using the probability density domain related migration super convex projection metric function.

In the embodiment of the present application, the internet of things information analysis module 140 is configured to pass the optimized irrigation analysis feature vector through a classifier to obtain a classification result, where the classification result is used to indicate whether irrigation is needed. It will be appreciated that by inputting the optimized irrigation analysis feature vector into the classifier, complex feature information can be converted into a simple classification result, i.e., whether irrigation is required. This simplifies the decision making process and makes it easier for the decision maker to understand and take corresponding actions. The use of irrigation analysis feature vectors in combination with classifiers may enable automated irrigation decisions. The classifier can automatically judge whether irrigation is needed according to the input feature vector, and manual intervention is not needed. Thus, time and labor cost can be saved, and the decision efficiency and accuracy can be improved. The optimized irrigation analysis feature vector is classified by a classifier and can be used as a problem of supervised learning. By collecting a certain amount of sample data, including feature vectors and corresponding labels whether irrigation is required, the classifier model can be trained and optimized so that it can better classify and predict new feature vectors. The irrigation system can flexibly adapt to different irrigation requirements and conditions by classifying through the classifier. And proper classification algorithm and parameter setting can be selected according to specific farmland environment and crop requirements so as to obtain a classification result which is most suitable for practical conditions. In addition, the model of the classifier can be extended and updated as needed to accommodate new features and requirements.

Accordingly, in one embodiment of the present application, the information analysis module 140 of the internet of things is configured to: processing the optimized irrigation analysis feature vector using the classifier in the following formula to obtain the classification result; wherein, the formula is: softmax { (W _n,B_n):…:(W₁,B₁) |x }, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, X is an optimized irrigation analysis feature vector, softmax represents a softmax function, and O represents the classification result.

In summary, according to the digital village application system and method based on the internet of things provided by the embodiment of the application, by deploying the soil humidity sensor and collecting the crop images, combining the weather information, extracting the soil humidity characteristics, the weather information semantic characteristics and the crop semantic association characteristics, fusing the characteristics for irrigation analysis, and classifying the irrigation analysis characteristics through the classifier to determine whether irrigation is needed. Therefore, accurate farmland irrigation management can be provided, the farmer is helped to make a decision, and the growth effect of crops is improved.

As described above, the digital rural application system 100 based on the internet of things according to the embodiment of the present application may be implemented in various terminal devices, for example, a server of the digital rural application system based on the internet of things, and the like. In one example, the digital rural application system 100 based on the internet of things may be integrated into the terminal device as one software module and/or hardware module. For example, the digital country application 100 based on the internet of things may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the digital rural application 100 based on the internet of things can be one of a plurality of hardware modules of the terminal device.

Alternatively, in another example, the internet of things-based digital country application 100 and the terminal device may be separate devices, and the internet of things-based digital country application 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information in a agreed data format.

Fig. 6 is a flowchart of a digital rural application method based on the internet of things according to an embodiment of the present application. As shown in fig. 6, the digital village application method based on the internet of things according to the embodiment of the application includes the steps of: s110, acquiring soil humidity values at a plurality of preset time points in a preset time period through a soil humidity sensor arranged in soil, acquiring weather information and acquiring crop images through an unmanned aerial vehicle; s120, extracting soil humidity feature vectors, weather information semantic feature vectors and crop semantic association feature vectors from the soil humidity values, the weather information and the crop images of the plurality of preset time points respectively; s130, constructing an irrigation analysis feature vector among the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector, and performing probability density domain related migration super convex projection measurement on the irrigation analysis feature vector to obtain an optimized irrigation analysis feature vector; and S140, the optimized irrigation analysis feature vector passes through a classifier to obtain a classification result, wherein the classification result is used for indicating whether irrigation is needed or not.

Here, it will be understood by those skilled in the art that the specific operations of the respective steps in the above-described internet of things-based digital rural application method have been described in detail in the above description of the internet of things-based digital rural application system with reference to fig. 1 to 5, and thus, repetitive descriptions thereof will be omitted.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.

In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. The utility model provides a digital rural application system based on thing networking which characterized in that includes:

2. The digital rural application system based on the internet of things according to claim 1, wherein the information processing module of the internet of things comprises:

a soil humidity characteristic extraction unit, configured to arrange soil humidity values at the plurality of predetermined time points into vectors, and obtain the soil humidity characteristic vector through characteristic extraction;

The weather information feature extraction unit is used for obtaining the weather information semantic feature vector from the weather information through convolutional encoding;

And the crop image feature extraction unit is used for encoding the crop image to obtain the crop semantic association feature vector.

3. The internet of things-based digital rural application system according to claim 2, wherein the soil moisture profile extraction unit comprises:

A humidity value arrangement vector subunit configured to arrange the soil humidity values at the plurality of predetermined time points into a soil humidity input vector;

And the multiscale feature extraction subunit is used for enabling the soil humidity input vector to pass through a humidity extractor based on a multiscale neighborhood feature extraction module to obtain the soil humidity feature vector.

4. The internet of things-based digital rural application system according to claim 3, wherein the multi-scale feature extraction subunit comprises:

A first scale humidity coding secondary subunit, configured to perform one-dimensional convolution coding on the soil humidity input vector with a one-dimensional convolution kernel having a first scale by using a first convolution layer of the humidity extractor based on the multi-scale neighborhood feature extraction module to obtain a first scale humidity feature vector, where the first convolution layer has a first one-dimensional convolution kernel having a first length;

A second scale humidity encoding secondary subunit, configured to perform one-dimensional convolution encoding on the soil humidity input vector with a one-dimensional convolution kernel having a second scale using a second convolution layer of the humidity extractor based on the multi-scale neighborhood feature extraction module to obtain a second scale humidity feature vector, where the second convolution layer has a second one-dimensional convolution kernel having a second length, and the first length is different from the second length;

and the multi-scale humidity cascade secondary subunit is used for cascading the first-scale humidity characteristic vector and the second-scale humidity characteristic vector to obtain the soil humidity characteristic vector.

5. The digital rural area application system based on the internet of things according to claim 4, wherein the weather information feature extraction unit comprises:

a semantic understanding subunit for obtaining weather information feature vectors from the weather information using a semantic understanding model;

a binarization subunit, configured to binarize the weather information based on conditions to obtain a weather information binarization feature vector;

a vector multiplication subunit for multiplying the weather information feature vector by a transpose of the weather information binarized feature vector to obtain a weather information feature matrix for expressing the weather information;

And the convolution coding subunit is used for inputting the weather information feature matrix into a weather information extractor based on the first convolution neural network model to obtain the weather information semantic feature vector.

6. The internet of things-based digital rural application system according to claim 5, wherein the semantic understanding subunit comprises:

the word vector conversion secondary subunit is used for mapping each word in the weather information into a word vector by using a word embedding layer of the semantic understanding model so as to obtain a word vector sequence of the weather information;

The semantic processing secondary subunit is used for processing the weather information word vector sequence by using a Bert model of the semantic understanding model to obtain a weather information word feature vector sequence;

and the context coding secondary subunit is used for performing context coding on the weather information word feature vector sequence by using the bidirectional LSTM network of the semantic understanding model so as to obtain the weather information feature vector.

7. The internet of things-based digital rural application system according to claim 6, wherein the crop image feature extraction unit comprises:

A gray image subunit, configured to convert the crop image into a gray image to obtain a crop gray image;

a gray image correction subunit, configured to correct the crop gray image to obtain a crop gray correction image;

a gray level co-occurrence matrix subunit configured to obtain a plurality of crop gray level feature statistics from the crop gray level correction image based on a gray level co-occurrence matrix;

A context coding subunit, configured to pass the plurality of crop gray feature statistics through a context encoder that includes an embedded layer to obtain a plurality of crop gray feature statistical semantic feature vectors;

and the bidirectional long-short term subunit is used for carrying out one-dimensional arrangement on the plurality of crop gray feature statistical semantic feature vectors and then obtaining the crop semantic association feature vectors through a bidirectional long-short term memory neural network model.

8. The internet of things-based digital rural application system according to claim 7, wherein the bi-directional long-short term sub-unit comprises:

a one-dimensional arrangement secondary subunit, configured to perform one-dimensional arrangement on the plurality of crop gray feature statistical semantic feature vectors to obtain a sequence of crop gray feature statistical semantic feature vectors;

And the sequence coding secondary subunit is used for carrying out context semantic coding on the sequence of the crop gray feature statistical semantic feature vector by using the two-way long-short-term memory neural network model so as to obtain the crop semantic association feature vector.

9. The digital rural application system based on the internet of things of claim 8, wherein the information fusion module of the internet of things comprises:

The fusion unit is used for fusing the soil humidity feature vector, the weather information semantic feature vector and the crop semantic association feature vector to obtain an irrigation analysis feature vector;

the optimizing unit is used for carrying out probability density domain related migration super convex projection measurement on the irrigation analysis feature vector so as to obtain an optimized irrigation analysis feature vector;

10. The digital rural application method based on the Internet of things is characterized by comprising the following steps of: