Construction method of intelligent agricultural platform based on big data analysis and visual model

Abstract

The invention discloses an intelligent agriculture platform construction method based on big data analysis and a visual model, and the method comprises the steps: arranging a sensor network, and monitoring the key environment parameters of a farmland in real time; using the unmanned aerial vehicle to shoot crop images; according to the shot crop image, identifying a part with pest and disease damage characteristics in the crop image; deducing the age of the crop leaf according to the shot crop image; a cloud platform is built, farmland related data is displayed in an integrated mode through the cloud platform, and decision suggestions are provided based on big data analysis according to all data input into the cloud platform. The system integrates multiple functions such as farmland data display, unmanned aerial vehicle fixed-point automatic cruise, disease and pest recognition and early warning, agricultural question and answer assistant and the like, and provides one-stop intelligent agricultural management service for farmers. The platform provides scientific guidance suggestions for agricultural production based on big data analysis, and assists in improving yield and quality.

Description

Technical Field

[0001] The present invention relates to technical fields such as big data development, machine vision recognition and smart agriculture, and specifically to a method for constructing a smart agriculture platform based on big data analysis and visual models. Background Art

[0002] Agriculture is a pillar industry in my country, which is related to food security and economic development, and plays a pivotal role in national development. At the same time, many problems such as the increasing world population, rising production costs, and reduced resources threaten the development of world agriculture. To solve these problems, it is necessary to vigorously develop smart agriculture as the fundamental way out for agriculture. Smart agriculture based on digitalization and informatization is the future development direction of agriculture. At present, many research results in this field have been achieved, including the launch of some agricultural platforms with intelligence as the direction.

[0003] However, after an in-depth analysis of various technical solutions in the current smart agriculture field, it is found that the current platforms generally have the following deficiencies:

[0004] 1. The phenomenon of information islands is serious: Existing smart agricultural systems often operate independently, and data between various functional modules cannot be effectively shared and communicated, resulting in information fragmentation between various links of agricultural production, making it difficult to form an overall synergistic effect, and affecting agricultural production efficiency and the scientific nature of decision-making.

[0005] 2. Low level of intelligence: Although most systems claim to have intelligent functions, in actual applications, there are often problems such as low intelligence, inaccurate decision support, and low efficiency of automated operations, making it difficult to fully utilize the advantages of smart agriculture.

[0006] 3. Limited customized service capabilities: The agricultural environment, crop types and farmers' needs in different regions vary greatly, and the existing systems often lack sufficient flexibility and customization capabilities, making it difficult to meet the diverse agricultural production needs. Summary of the invention

[0007] Purpose of the invention: The purpose of the present invention is to address the deficiencies in the prior art and to provide a method for constructing a smart agriculture platform based on big data analysis and visual models, so as to construct a smart agriculture platform that integrates multiple functions such as farmland data display, crop pest and disease analysis and prediction, and crop leaf age inference.

[0008] In order to achieve the above object, the present invention provides a method for constructing a smart agriculture platform based on big data analysis and visual model, comprising the following steps:

[0009] S1. Deploy a sensor network to monitor key environmental parameters of farmland in real time; use drones to capture images of crops;

[0010] S2, based on the captured crop images, identify the parts with pest and disease characteristics;

[0011] S3, inferring the leaf age of the crop leaves according to the captured crop images;

[0012] S4, building a cloud platform, inputting the key environmental parameters and crop images obtained in step S1, the characteristics of crop diseases and insect pests identified in step S2, the corresponding crop image calibration positions, and the leaf ages of crop leaves obtained in step S3 into the cloud platform;

[0013] The cloud platform is used to integrate and display farmland-related data, and based on big data analysis, decision-making recommendations are provided according to the data input into the cloud platform.

[0014] A further preferred technical solution of the present invention is that in step S1, a sensor network is arranged to monitor key environmental parameters of farmland in real time, and the specific method is as follows:

[0015] Arrange several different types of wired and / or wireless sensors in the farmland to form a sensor network to monitor the soil moisture, soil nitrogen, phosphorus and potassium content, soil temperature and light intensity of the farmland in real time;

[0016] The data collected by each sensor is cleaned and standardized and then transmitted to the cloud platform.

[0017] Preferably, in step S1, a drone is used to take images of crops, and the specific method is as follows:

[0018] Set the cruising route of the drone, which is equipped with positioning equipment and high-definition cameras;

[0019] The drone takes images of crops along the cruising route and calibrates the positions of the captured crop images based on the drone's positioning information and specific landmarks or crop features on the ground.

[0020] Preferably, the method of identifying the portion of the crop image having pest and disease characteristics according to the captured crop image in step S2 is as follows:

[0021] S21, respectively constructing a classification model and a target detection model, and training them based on pest and disease image data;

[0022] S22, preprocessing the crop image obtained in step S1;

[0023] S23, using the trained classification model and target detection model to detect the preprocessed crop images, and identifying the parts with pest and disease characteristics;

[0024] S24, transmitting the crop image with pest and disease characteristics and the calibrated position corresponding to the image to the cloud platform.

[0025] Preferably, the classification model and the target detection model are constructed separately in step S21, and are trained according to the pest image data. The specific method is as follows:

[0026] First, the pest and disease image data is standardized, including image scaling, cropping, and data enhancement;

[0027] The classification model adopts the ResNet architecture and introduces the channel attention mechanism and SimAM attention mechanism to enhance the network's attention to important features; the target detection model adopts the YOLOv5 framework and introduces the CBAM attention mechanism;

[0028] The training process of the model is based on gradient descent, combined with a learning rate scheduling strategy to gradually optimize the model performance; in the classification model training, the cross entropy loss function is adopted, and the Adam optimizer is used for training; in the target detection model training, a multi-task loss function based on CIOU loss is adopted, combined with classification, positioning and IoU loss for optimization.

[0029] Preferably, in step S3, the age of the crop leaves is inferred based on the captured crop images, and the specific method is as follows:

[0030] Based on the captured crop images, the KMeans clustering algorithm is used to analyze the vein structure, and the Roberts operator is used for edge detection. By extracting the veins and leaf edge positions, the vein orientation is calculated, and the age of the crop leaves is inferred based on the different vein orientations and the current month.

[0031] Preferably, in step S4, based on big data analysis, decision suggestions are provided according to various data input into the cloud platform, including:

[0032] Provide irrigation, fertilization and pesticide application plan recommendations based on key environmental parameters of the farmland and the leaf age of the crops;

[0033] Provide analysis of the types and causes of crop pests and diseases, and provide pesticide application recommendations based on key environmental parameters of farmland and crop images with pest and disease characteristics;

[0034] Pest and disease warnings are issued based on crop images with pest and disease characteristics and the calibrated locations corresponding to the images.

[0035] Preferably, the cloud platform in step S4 is also integrated with an AI question-answering assistant developed based on a large model, and the AI ​​question-answering assistant has a built-in database containing agricultural production knowledge.

[0036] Beneficial effects: The present invention breaks down information silos and realizes data interconnection and interoperability: by building a unified smart agricultural platform, it realizes real-time collection, processing and sharing of data from all links of agricultural production, breaking down information barriers; on the hardware side, drones collect crop images, and sensor systems monitor in real time and transmit back precise values ​​such as soil nutrition and entropy status; on the software side, the platform presents data and diagnoses pests and diseases, ultimately achieving fine control of the production cycle of its own faults, and promoting close collaboration and efficient coordination of all links in agricultural production.

[0037] The platform constructed by the present invention greatly improves the level of intelligence and optimizes agricultural production decisions: it uses big data analysis and computer vision models to provide accurate decision-making support for agricultural production, realizes scientific planting planning, timely early warning of pests and diseases, and refined resource management, thereby improving agricultural production efficiency and output quality.

[0038] The present invention enhances customized service capabilities to meet diversified needs. It can provide customized smart agriculture solutions based on the agricultural characteristics and needs of farmers in different regions. It provides professional technical support and services throughout the entire process from solution design, equipment selection, installation and commissioning to later operation and maintenance to ensure the effective implementation and continuous operation of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flowchart of a method for constructing a smart agriculture platform based on big data analysis and visual models. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0041] Embodiment: This embodiment provides a method for constructing a smart agriculture platform based on big data analysis and visual models, such as Figure 1 As shown, the following steps are included:

[0042] S1. Deploy a sensor network to monitor key environmental parameters of farmland in real time; use drones to capture images of crops;

[0043] S2, based on the captured crop images, identify the parts with pest and disease characteristics;

[0044] S3, inferring the leaf age of the crop leaves according to the captured crop images;

[0045] S4, building a cloud platform, inputting the key environmental parameters and crop images obtained in step S1, the characteristics of crop diseases and insect pests identified in step S2, the corresponding crop image calibration positions, and the leaf ages of crop leaves obtained in step S3 into the cloud platform;

[0046] The cloud platform is used to integrate and display farmland-related data, and based on big data analysis, decision-making recommendations are provided according to the data input into the cloud platform.

[0047] Specifically, in step S1, a sensor network is deployed to monitor key environmental parameters of farmland in real time. The specific method is as follows:

[0048] Arrange several different types of wired and / or wireless sensors in the farmland to form a sensor network to monitor the soil moisture, soil nitrogen, phosphorus and potassium content, soil temperature and light intensity of the farmland in real time;

[0049] The data collected by each sensor is cleaned and standardized and then transmitted to the cloud platform.

[0050] The present invention collects data such as soil moisture, temperature, pH value, light intensity, etc. through sensors deployed in farmland, and transmits it to the cloud platform in real time through the Internet of Things. Farmers can access this data through smart devices, and the platform displays it in the form of charts or dashboards for farmers to understand intuitively. If the soil moisture sensor shows that a certain area is dry, the platform will automatically highlight it in red on the dashboard to remind farmers to pay attention to irrigation.

[0051] In addition, for the part where the sensor data is collected and sent to the platform, the present invention has the following feasible technical implementation routes: 1. Use a microcontroller to connect the soil sensor, periodically read the soil temperature and humidity, nitrogen, phosphorus and potassium data, and perform simple data processing, such as calibration and mean calculation, and then send the collected data to the cloud server through the wireless module; 2. Select soil sensor nodes with wireless communication functions and place them in the farmland. These nodes communicate with the gateway through a self-organized wireless sensor network, and the gateway is responsible for collecting, processing and summarizing data. In addition, single-board computers (such as other models of Raspberry Pi) are used to collect data, and edge computing technology is used for real-time data processing and analysis, but this solution has too high requirements for resource occupation and bandwidth, and its application value is not high.

[0052] In step S1, the crop images are photographed by using a drone, and the specific method is as follows:

[0053] Set the cruising route of the drone, which is equipped with positioning equipment and high-definition cameras;

[0054] The drone takes images of crops along the cruising route and calibrates the positions of the captured crop images based on the drone's positioning information and specific landmarks or crop features on the ground.

[0055] The drone of the present invention is equipped with GPS positioning and a high-definition camera. It automatically cruises farmland according to a preset route to collect information on crop growth and pests and diseases. The drone can identify specific landmarks or crop features to achieve precise positioning and data collection. During the cruise, the drone will periodically take high-definition images, automatically record the locations of pests and diseases, and upload them to the platform for identification by the identification system.

[0056] The specific method of identifying the portion of the crop image having pest and disease characteristics according to the captured crop image in step S2 is as follows:

[0057] S21. Build a classification model and a target detection model respectively, and train them based on pest and disease image data; including:

[0058] First, the pest and disease image data is standardized, including image scaling, cropping, and data enhancement;

[0059] The classification model adopts the ResNet architecture and introduces the channel attention mechanism and SimAM attention mechanism to enhance the network's attention to important features; the target detection model adopts the YOLOv5 framework and introduces the CBAM attention mechanism;

[0060] The model training process is based on gradient descent, combined with a learning rate scheduling strategy to gradually optimize the model performance. In the classification model training, the cross entropy loss function is used, and the Adam optimizer is used for training. In the object detection model training, a multi-task loss function based on CIOU loss is used, combined with classification, positioning, and IoU loss for optimization.

[0061] S22, preprocessing the crop image obtained in step S1;

[0062] S23, using the trained classification model and target detection model to detect the preprocessed crop images, and identifying the parts with pest and disease characteristics;

[0063] S24, transmitting the crop image with pest and disease characteristics and the calibrated position corresponding to the image to the cloud platform.

[0064] The present invention combines deep learning and image recognition technology. The images taken by the drone are analyzed in real time to identify the characteristics of pests and diseases. Based on the recognition results, the system sends early warning information to farmers in a timely manner. If the system identifies the presence of rice leafhoppers in a certain rice field, the farmer will be notified immediately.

[0065] The present invention constructs a multi-model voting mechanism based on deep learning, especially supervised learning in the field of image classification. After a large number of comparative experiments and ablation experiments, the present invention selects the ResNet model, which significantly improves the performance of pest image classification tasks by adding a hybrid mechanism of channel attention (CAM) and spatial attention (SAM). This hybrid attention mechanism can enhance the model's ability to capture image features, especially in the subtle differences of pest features, thereby improving the accuracy of classification.

[0066] In order to further optimize the target monitoring task, based on the principle of accuracy priority rather than speed priority, the present invention adopts the yolov5 model and introduces the convolutional block attention mechanism (CBAM). This attention mechanism helps the model to more accurately identify the pest and disease areas in the image, and maintain a high detection accuracy even in complex and changeable natural environments.

[0067] In the process of model selection, some other classic image classification models, such as SEnet and GoogLeNet, were also tried, and they were combined with different attention mechanisms. However, the effects of these solutions in practical applications are not ideal. In addition, other models of the yolo series, such as yolov8, and the Transformer-based DETR model were also explored for the target monitoring task of rice pests and diseases. However, after rigorous testing and evaluation, it was found that the performance of these models in this project was not as good as the solution finally adopted by the present invention.

[0068] Classification process: After the classification model inputs the pest image, it first extracts low-level and high-level features through a series of convolutional layers. Subsequently, the channel attention mechanism and SimAM module further optimize the weights of the features, and finally classifies through the fully connected layer to output the most likely pest category.

[0069] Detection process: The target detection model receives the input image, extracts features through the YOLOv5 backbone network, and then weights the feature map in combination with the CBAM attention mechanism to output the bounding box and category information of the pest target. The model detects at multiple scales, ensuring high-precision recognition of targets of different sizes.

[0070] In step S3, the age of the crop leaves is inferred based on the captured crop images. The specific method is as follows:

[0071] Based on the captured crop images, the KMeans clustering algorithm is used to analyze the vein structure, and the Roberts operator is used for edge detection. By extracting the veins and leaf edge positions, the vein orientation is calculated, and the age of the crop leaves is inferred based on the different vein orientations and the current month.

[0072] In step S4, based on big data analysis, decision suggestions are provided according to various data input into the cloud platform, including:

[0073] Provide irrigation, fertilization and pesticide application plan recommendations based on key environmental parameters of the farmland and the leaf age of the crops;

[0074] Provide analysis of the types and causes of crop pests and diseases, and provide pesticide application recommendations based on key environmental parameters of farmland and crop images with pest and disease characteristics;

[0075] Pest and disease warnings are issued based on crop images with pest and disease characteristics and the calibrated locations corresponding to the images.

[0076] The present invention integrates multiple functions such as farmland data display, automatic cruise of drones at fixed points, pest and disease identification and early warning, and agricultural question-and-answer assistant, providing farmers with one-stop smart agricultural management services. The platform provides scientific guidance and suggestions for agricultural production based on big data analysis, helping to improve yield and quality.

[0077] The cloud platform in step S4 is also integrated with an AI question-answering assistant developed based on a large model, and the AI ​​question-answering assistant has a built-in database containing agricultural production knowledge.

[0078] The platform constructed by the present invention integrates an intelligent question-and-answer system, where farmers can ask questions through voice or text to obtain information and suggestions related to agricultural production. The question-and-answer assistant can understand the farmers' questions and provide answers based on big data and professional knowledge.

[0079] In summary, the platform constructed by the present invention comprehensively covers all aspects of crop growth, including growth status assessment, full process monitoring and intelligent agricultural machinery control, etc., providing a one-stop solution for agricultural production and greatly simplifying the management process of agricultural production.

[0080] The present invention integrates sensors to monitor soil nutrition, full-process monitoring cameras and drone fixed-point monitoring facilities, combined with AI data analysis and multi-dimensional comprehensive evaluation algorithms, to achieve accurate monitoring and evaluation of crop growth status, pests and diseases, and leaf age. This innovation has significantly improved the level of intelligent agricultural production and provided farmers with a more scientific and efficient planting management solution.

[0081] The present invention uses multiple sets of high-precision sensors to monitor and transmit accurate values ​​such as soil nutrition and entropy in real time, providing strong support for the refined monitoring of farmland environment. These accurate data provide a reliable basis for subsequent comprehensive evaluation, decision-making and optimization management, further improving the accuracy and efficiency of agricultural production.

[0082] The integrated access platform (cloud platform) constructed by the present invention not only realizes functions such as data visualization and real-time control of agricultural machinery, but also uses big data and artificial intelligence technology to conduct intelligent management and early warning of the entire planting process. This innovation effectively reduces the risks of agricultural production, improves the stability and benefits of production, and at the same time reduces the need for manual intervention and reduces production costs.

[0083] The platform constructed by the present invention can be used with advanced drones and intelligent agricultural equipment to achieve accurate monitoring and management of farmland. Drones can be used for tasks such as crop growth monitoring, pest inspection, and precise fertilization and pesticide application to improve work efficiency and accuracy; intelligent equipment covers multiple links such as irrigation, sowing, and harvesting, realizing full automation of agricultural production.

[0084] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for constructing a smart agriculture platform based on big data analysis and visual models, characterized in that: The steps include: S1. Deploy a sensor network to monitor key environmental parameters of farmland in real time; use drones to capture images of crops; S2, based on the captured crop images, identify the parts with pest and disease characteristics; S3, inferring the leaf age of the crop leaves according to the captured crop images; S4, building a cloud platform, inputting the key environmental parameters and crop images obtained in step S1, the characteristics of crop diseases and insect pests identified in step S2, the corresponding crop image calibration positions, and the leaf ages of crop leaves obtained in step S3 into the cloud platform; The cloud platform is used to integrate and display farmland-related data, and based on big data analysis, decision-making recommendations are provided according to the data input into the cloud platform.

2. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 1, characterized in that: In step S1, a sensor network is deployed to monitor key environmental parameters of farmland in real time. The specific method is as follows: Arrange several different types of wired and / or wireless sensors in the farmland to form a sensor network to monitor the soil moisture, soil nitrogen, phosphorus and potassium content, soil temperature and light intensity of the farmland in real time; The data collected by each sensor is cleaned and standardized and then transmitted to the cloud platform.

3. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 1, characterized in that: In step S1, the crop images are photographed by using a drone, and the specific method is as follows: Set the cruising route of the drone, which is equipped with positioning equipment and high-definition cameras; The drone takes images of crops along the cruising route and calibrates the positions of the captured crop images based on the drone's positioning information and specific landmarks or crop features on the ground.

4. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 3, characterized in that: The specific method of identifying the portion of the crop image having pest and disease characteristics according to the captured crop image in step S2 is as follows: S21, respectively constructing a classification model and a target detection model, and training them based on pest and disease image data; S22, preprocessing the crop image obtained in step S1; S23, using the trained classification model and target detection model to detect the preprocessed crop images, and identifying the parts with pest and disease characteristics; S24, transmitting the crop image with pest and disease characteristics and the calibrated position corresponding to the image to the cloud platform.

5. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 4, characterized in that: In step S21, a classification model and a target detection model are constructed respectively, and trained according to the pest image data. The specific method is as follows: First, the pest and disease image data is standardized, including image scaling, cropping, and data enhancement; The classification model adopts the ResNet architecture and introduces the channel attention mechanism and SimAM attention mechanism to enhance the network's attention to important features; the target detection model adopts the YOLOv5 framework and introduces the CBAM attention mechanism; The model training process is based on gradient descent, combined with a learning rate scheduling strategy to gradually optimize model performance. In the classification model training, the cross entropy loss function is adopted, and the Adam optimizer is used for training. In the object detection model training, a multi-task loss function based on CIOU loss is used, combined with classification, positioning and IoU loss for optimization.

6. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 3, characterized in that: In step S3, the age of the crop leaves is inferred based on the captured crop images. The specific method is as follows: Based on the captured crop images, the KMeans clustering algorithm is used to analyze the vein structure, and the Roberts operator is used for edge detection. By extracting the veins and leaf edge positions, the vein orientation is calculated, and the age of the crop leaves is inferred based on the different vein orientations and the current month.

7. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 1, characterized in that: In step S4, based on big data analysis, decision suggestions are provided according to various data input into the cloud platform, including: Provide irrigation, fertilization and pesticide application plan recommendations based on key environmental parameters of the farmland and the leaf age of the crops; Provide analysis of the types and causes of crop pests and diseases, and provide pesticide application recommendations based on key environmental parameters of farmland and crop images with pest and disease characteristics; Pest and disease warnings are issued based on crop images with pest and disease characteristics and the calibrated locations corresponding to the images.

8. The method for constructing a smart agriculture platform based on big data analysis and visual model according to claim 1, characterized in that: The cloud platform in step S4 is also integrated with an AI question-answering assistant developed based on a large model, and the AI ​​question-answering assistant has a built-in database containing agricultural production knowledge.

Images