Farmland soil health comprehensive detection method and device based on deep learning

Abstract

The invention provides a farmland soil health comprehensive detection method and device based on deep learning, and belongs to the technical field of soil health detection.The method comprises the steps that data collection is conducted on current farmland soil through a data collection device, and collected data is obtained and preprocessed; obtaining soil surface image data, soil environment data and Beidou satellite navigation data; processing the soil surface image data by using a deep learning model to obtain a farmland surface detection result and a soil drought classification result; analyzing the soil environment data, combining a soil drought classification result to obtain accurate classification of drought grades, and obtaining time-space positioning information of current farmland soil according to Beidou satellite navigation data; and integrating the farmland surface detection result, the accurate classification of the drought grades and the time-space positioning information of the current farmland soil to obtain summarized data. According to the invention, high-precision and high-efficiency comprehensive detection of farmland soil health is realized, and various environmental parameters can be comprehensively detected.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil health detection, and in particular, relates to a method and device for comprehensive detection of farmland soil health based on deep learning. Background Art

[0002] Farmland soil is the foundation for the survival of crops and the basis for human survival. First of all, water in the soil is indispensable in the agricultural production process. On the one hand, due to climate and environmental changes, the moisture content of farmland is in an unstable state. On the other hand, the construction and management of farmland irrigation facilities are not perfect, which often leads to waste of water resources or soil drought due to untimely irrigation. In large-scale agricultural production, real-time, accurate and automatic collection of soil moisture information and judgment of farmland soil drought level are not only crucial for water-saving irrigation and increasing crop yields, but also can greatly reduce the input of manpower and material resources.

[0003] Secondly, due to the continuous expansion of human activities and the acceleration of human industrialization, the problem of heavy metal pollution in soil has become increasingly prominent. Heavy metals are metal elements with a relative density greater than 5g / cm3, such as copper, lead, chromium, manganese, nickel, cadmium, etc. According to the different sources of heavy metal pollution in soil, the pollution sources are generally divided into two aspects: one is natural factors and the other is human factors. Heavy metal pollution exists for a long time and is difficult to be effectively degraded in the soil. It is often enriched into the human body through the food chain, causing potential harm to human health. Real-time and accurate detection of heavy metal content in farmland soil is of great significance to the safe cultivation of crops.

[0004] In addition, a large amount of pesticides are used in agricultural production, which has made great contributions to eliminating pests and diseases, eradicating weeds, and increasing agricultural output; but with the increase in the use of pesticides, the residual pesticide pollution in the soil environment has become increasingly serious. In 2014, relevant surveys showed that the total point-to-point excess rate of soil in my country was 16.1%, and the poor quality rate of cultivated land was 27.9%. The soil pollution situation is particularly severe in some areas. Studies have shown that only 20% to 30% of the pesticides applied each year act on the target organisms, and the remaining 70% to 80% will enter the environment. The sources of pesticides in the soil are divided into direct and indirect sources: a small part of pesticides that directly act on the soil and pesticides applied to fields and other places fall on the surface of crops, and most of the rest remain in the soil or drift in the atmosphere; sewage, wastewater or animal and plant residues are contaminated by pesticides and then irrigated or fall into the soil, which is an indirect source of pesticide residues in the soil. Soil not only participates in the exchange of matter and energy in terrestrial ecosystems, but also participates in material exchange through the atmosphere and water cycle. Pesticide control in soil poses great challenges to the research and protection of soil safety due to the hysteresis and concealment of pesticides in soil. Analyzing and testing soil is currently the only way to determine. The concentration of pesticides in soil is at trace levels. In order to meet the pesticide residue limit detection standards, it is very necessary to use efficient, fast and convenient methods to detect pesticide residues in soil.

[0005] Due to poor management, weeds, debris, wild animals, etc. often appear on the surface of farmland soil, and pests and diseases may appear on the surface of crops. These factors will affect the growth and development of crops. Manual inspection of farmland requires frequent inspections, which results in a large investment of manpower and time. Automated detection of weeds, debris, wild animals, and crop pests and diseases can help farmland managers understand the growth dynamics of crops, save manpower, and improve efficiency.

[0006] In the existing technology, soil moisture monitoring has basically achieved instant communication and network transmission. Equipment such as soil moisture monitors have been widely used, and data monitoring and release systems have been gradually built and improved. In terms of monitoring and forecasting soil moisture, the overall work level is relatively high. In the existing technology, spectral analysis, fluorescence spectroscopy, electrochemical analysis and biosensor methods are used for heavy metal inspection. The pesticide residue analysis and detection technology is divided into chemical detection, biochemical detection and biological detection, but the most widely used is instrumental analysis, which has high sensitivity and With high accuracy, fast and convenient, the existing technology uses solid phase extraction / high performance liquid chromatography-tandem mass spectrometry to simultaneously detect 24 pesticide residues in environmental water samples, with a recovery rate of 65.9% to 127.8%; for the identification of weeds, debris, wild animals, pests and diseases, and crop growth status, the existing technology uses a classification system for outdoor weed and vegetable images; the system uses gray-level co-occurrence matrix (GLCM) to extract weed texture features, and combines principal component analysis (PCA) to reduce the dimension to 2D feature space, and finally uses support vector machine (SVM) to achieve classification. Experimental results show that the classification accuracy of the system can reach more than 90%, with high recognition performance.

[0007] At present, there is no unified comprehensive detection method and device for farmland soil health that can simultaneously detect soil moisture content, heavy metal content, pesticide residue content, weeds, debris, wild animals, pests and diseases, and crop growth status. Secondly, due to the complex soil environment and uneven distribution, the traditional detection method does not sample uniformly and has large errors. In addition, the traditional detection method does not collect sufficient data, and the existing detection instruments have a single function and only measure the sampling data from a single indicator. In addition, the existing detection technology requires chemical analysis of the sampling data, which consumes a certain amount of time and cost, so real-time detection cannot be achieved. Finally, the existing technology cannot give comprehensive results, and lacks the necessary data summary to help managers make decisions to solve corresponding problems. Summary of the invention

[0008] In response to the above-mentioned deficiencies in the prior art, the present invention provides a method and device for comprehensive detection of farmland soil health based on deep learning, which solves the problems of insufficient accuracy, low efficiency and inability to comprehensively detect multiple environmental parameters in existing detection methods and devices.

[0009] In order to achieve the above objectives, the technical solution adopted by the present invention is as follows: On the one hand, the present invention provides a method for comprehensive detection of farmland soil health based on deep learning, comprising the following steps:

[0010] S1. Use a data acquisition device to collect data on the current farmland soil to obtain collected data, and pre-process the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data;

[0011] S2. Label the soil surface image data to obtain the farmland surface data set, and use the YOLACT++ model to process the farmland surface data set to obtain the farmland surface detection results and soil drought classification results;

[0012] S3. Analyze soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, and combine soil drought classification results to obtain accurate drought grade classification. According to Beidou satellite navigation data, obtain the current spatial and temporal positioning information of farmland soil;

[0013] S4. Integrate the farmland surface detection results, precise classification of drought levels, heavy metal content, pesticide residue content, and the current spatiotemporal positioning information of farmland soil to obtain summary data and complete the comprehensive detection of farmland soil health.

[0014] The beneficial effects of the present invention are as follows: the present invention realizes comprehensive detection of multiple environmental parameters and improves the accuracy and efficiency of multiple detection tasks by combining deep learning technology with traditional environmental detection methods; the present invention performs target detection and segmentation on soil surface images through deep learning methods, and combines it with multiple tasks such as soil drought classification and soil environmental data analysis, thereby realizing the ability to automatically identify and classify different environmental factors in farmland, and conduct comprehensive evaluation on this basis to comprehensively monitor the soil environment; the present invention reduces manual intervention through intelligent automated detection means, improves the speed and accuracy of the detection process, and contributes to the intelligence and precision of soil management and environmental protection in agricultural production.

[0015] Furthermore, the S1 comprises the following steps:

[0016] S101. Use a soil detector to obtain soil sample data, use a high-speed linear array camera to obtain a high-definition color system image of the soil surface, use a laser scanner to obtain three-dimensional point cloud data, use an infrared camera to obtain an infrared image, and use a Beidou positioning terminal to obtain the current farmland soil positioning and navigation data;

[0017] S102, integrating soil sample data, soil surface high-definition color system images, three-dimensional point cloud data, infrared images, and current farmland soil positioning and navigation data to obtain collected data;

[0018] S103, performing data cleaning, data integration, data transformation and data reduction processing on the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data.

[0019] The beneficial effects of the above further scheme are as follows: the present invention collects data through soil detectors, high-speed linear array cameras, laser scanners, infrared cameras and Beidou positioning terminals, thereby improving the real-time and comprehensiveness of multi-dimensional data collected from farmland soil environment; and adopts precise data preprocessing and cleaning methods to improve the quality and accuracy of data input into the comprehensive farmland soil health detection method based on deep learning.

[0020] Furthermore, the S2 comprises the following steps:

[0021] S201, according to the soil surface image data, the data annotation contents are divided into: image-level annotations including weed categories, debris categories, wild animal categories, pest and disease categories, crop growth cycle status categories, and drought level categories, and instance-level annotations including position coordinates of weeds and debris, crop pest and disease location coordinates, and wild animal location coordinates;

[0022] S202, using an image annotation tool to perform position annotation, category annotation, and pixel-level annotation on image-level annotation and instance-level annotation to obtain annotated soil surface image data, and integrating the annotated soil surface image data to obtain a farmland surface data set;

[0023] S203, inputting the farmland surface dataset into the YOLACT++ model, and extracting features using a feature pyramid network to obtain image features;

[0024] S204, inputting the image features into the prediction head, and obtaining the prediction features through non-maximum suppression; inputting the image features into the global mask to generate the image mask; performing feature fusion on the prediction features and the image mask to obtain the fusion features;

[0025] S205, the fused features are clipped and thresholded in sequence to obtain farmland surface detection results and soil drought classification results.

[0026] The beneficial effects of the above further scheme are as follows: the present invention realizes accurate pixel-level target area annotation by performing detailed data annotation and division of soil surface image data; and adopts YOLACT++ target detection and segmentation method to perform target detection and soil drought classification of farmland surface data set, thereby improving the accuracy and efficiency in target detection and segmentation tasks, and realizing comprehensive real-time and rapid detection of farmland soil surface.

[0027] Furthermore, the S3 comprises the following steps:

[0028] S301, analyzing soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, combining soil drought classification results and soil moisture content, and using comparative analysis to obtain accurate classification of drought levels;

[0029] S302: According to Beidou satellite navigation data, the spatiotemporal positioning information of the current farmland soil is obtained by using spatiotemporal positioning analysis.

[0030] The beneficial effects of the above further scheme are as follows: the present invention achieves a more accurate soil drought classification and provides an accurate soil health assessment by utilizing the classification results of the deep learning model and combining it with soil environmental data for comprehensive analysis; and introduces spatiotemporal positioning analysis to optimize the spatial accuracy of the detection results based on Beidou satellite data, thereby improving the credibility and quality of the acquired data.

[0031] Furthermore, the step S4 is specifically as follows:

[0032] S401, storing the farmland surface detection results, accurate classification of drought levels, heavy metal content, pesticide residue content, and spatiotemporal positioning information of the current farmland soil in a database management system, and displaying them through data visualization;

[0033] S402: Summarize the data stored in the database management system to obtain summary data, and use the data analysis system to generate agricultural management strategies to complete comprehensive detection of farmland soil health.

[0034] The beneficial effects of the above further scheme are as follows: the present invention classifies and summarizes the acquired data, thereby reducing the impact of the complexity and diversity of the data and the large amount of data, which makes it inconvenient to classify and manage; it realizes the accurate classification and detection of multiple indicators such as soil moisture content, heavy metal content, pesticide residue content, weeds, debris, pests and diseases, crop growth status and drought level; it improves the supervision of the soil environment and realizes early warning; the present invention standardizes the management of the test result data and improves the management efficiency; it displays the test results through visualization and generates agricultural management strategies, based on which managers can propose targeted measures, thereby improving the scientific nature and sustainable development of agricultural production.

[0035] In order to achieve the above-mentioned object, according to a second aspect of the present invention, a deep learning-based comprehensive detection device for farmland soil health is provided, comprising: a high-definition camera, an inclinometer, a motor controller, an automatic adjustment telescopic rod, a rangefinder, an environmental detector, a digital leveling device, a positioning and orientation system stabilizing gimbal, a laser scanner, a control processing device, a processing terminal and a Beidou positioning terminal;

[0036] The automatically adjustable telescopic rod is located at the bottom of the farmland soil health comprehensive detection device and is equipped with wheels;

[0037] The positioning and orientation system stabilizes the pan / tilt head above the automatically adjustable telescopic rod;

[0038] The high-definition camera, inclinometer, motor controller, rangefinder, environmental detector, digital leveling device, laser scanner, control processing device, processing terminal and Beidou positioning terminal are installed on the stable gimbal of the positioning and orientation system;

[0039] The high-definition camera is located at the left rear of the stable platform of the positioning and orientation system; the inclinometer is located at the upper left corner of the stable platform of the positioning and orientation system;

[0040] The motor controller is located on the left side of the high-definition camera; the rangefinder is located in front of the stable platform of the positioning and orientation system; the environmental detector is located in front of the right side of the stable platform of the positioning and orientation system; the digital leveling device is located on the right side of the environmental detector, at the upper right corner of the stable platform of the positioning and orientation system;

[0041] The laser scanner is located at the right rear of the stable platform of the positioning and orientation system; the control processing device is located in the middle of the stable platform of the positioning and orientation system; the processing terminal is located at the rear of the control processing device; and the Beidou positioning terminal is located at the right side of the processing terminal;

[0042] The control processing device is respectively connected with a high-definition camera, an inclinometer, a motor control, a rangefinder, an environmental detector, a digital leveling device, a laser scanner, a processing terminal and a Beidou positioning terminal.

[0043] Further, the soil environment detector is a soil detector, connected to a drilling device, and enters the soil through the drilling device, and is used to collect and analyze data every three meters, collect data, and analyze multiple components at the same time;

[0044] The laser scanner is used to collect three-dimensional point cloud data;

[0045] The high-definition camera includes an array camera and an infrared camera, which is used to collect high-definition images and infrared images;

[0046] The Beidou positioning terminal is used to collect positioning and navigation data of the farmland soil health comprehensive detection device;

[0047] The control processing device is used to receive the collected data and transmit it to the processing terminal;

[0048] The processing terminal includes a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the processing terminal executes the comprehensive detection method of farmland soil health based on deep learning;

[0049] The motor controller is used to power the farmland soil health comprehensive detection device.

[0050] The beneficial effects of the above scheme are as follows: the present invention integrates a variety of sensor equipment into a comprehensive farmland soil health detection device, and embeds the proposed comprehensive farmland soil health detection method based on deep learning, thereby realizing automatic data collection, processing and analysis; the present invention effectively improves the automation level and efficiency of farmland soil health detection through automation components and their integration technology, especially the part that combines the device with the deep learning algorithm; and realizes high-precision, high-efficiency and effective comprehensive detection of multiple environmental parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The figure is a flow chart of the method of the present invention.

[0052] Figure 2 This is a flow chart of the comprehensive detection of farmland soil surface in this embodiment.

[0053] Figure 3 This is a structural diagram of a comprehensive farmland soil health detection device based on deep learning in this embodiment.

[0054] Among them, 1. High-definition camera, 2. Inclinometer, 3. Motor controller, 4. Automatic adjustment telescopic rod, 5. Rangefinder, 6. Environmental detector, 7. Digital leveling device, 8. POS stabilized gimbal, 9. Laser scanner, 10. Control processing device, 11. Processing terminal, 12. Beidou positioning terminal. DETAILED DESCRIPTION

[0055] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0056] Before describing this embodiment, the following terms are explained:

[0057] YOLACT++: A single-stage real-time object detection and segmentation model.

[0058] POS stabilized gimbal: positioning and orientation system stabilized gimbal;

[0059] RGB: color system;

[0060] Labelme - an image annotation tool.

[0061] FPN network: Feature Pyramid Network;

[0062] NMS: non-maximum suppression;

[0063] mask: mask.

[0064] Example 1

[0065] The present invention aims to solve a number of technical problems existing in the comprehensive detection of farmland soil health, including: difficulties in detecting soil moisture content, heavy metal content, pesticide residue content, weeds, debris, wild animals, pests and diseases, and crop growth status; the current detection methods have problems such as insufficient accuracy, low efficiency, and inability to comprehensively detect multiple environmental parameters; in order to solve these problems, the present invention proposes a method and device for comprehensive detection of farmland soil health based on deep learning; the device combines deep learning technology with traditional environmental detection methods, which can effectively improve the accuracy and efficiency of multiple detection tasks. Specifically, the deep learning algorithm can automatically identify and classify different environmental factors in farmland by analyzing the multi-dimensional data collected by sensors, and conduct a comprehensive evaluation on this basis to achieve comprehensive monitoring of the soil environment. The technical solution of the invention reduces manual intervention through intelligent and automated detection methods, improves the speed and accuracy of the detection process, and contributes to the intelligent and precise soil management and environmental protection in agricultural production.

[0066] like Figure 1 As shown, the present invention provides a comprehensive detection method for farmland soil health based on deep learning, and its implementation method is as follows:

[0067] S1. Use a data acquisition device to collect data on the current farmland soil to obtain collected data, and pre-process the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data. The specific steps are as follows:

[0068] S101. Use a soil detector to obtain soil sample data, use a high-speed linear array camera to obtain a high-definition color system image of the soil surface, use a laser scanner to obtain three-dimensional point cloud data, use an infrared camera to obtain an infrared image, and use a Beidou positioning terminal to obtain the current farmland soil positioning and navigation data;

[0069] S102, integrating soil sample data, soil surface high-definition color system images, three-dimensional point cloud data, infrared images, and current farmland soil positioning and navigation data to obtain collected data;

[0070] S103, performing data cleaning, data integration, data transformation and data reduction processing on the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data.

[0071] In this embodiment, data acquisition is performed using a variety of data acquisition devices, including soil detectors, high-speed linear array cameras, laser scanners, infrared cameras, and Beidou positioning terminals. The soil detector is used to acquire and analyze soil sample data, including water content, heavy metal content, and pesticide residues; the high-speed linear array camera is used to acquire high-definition RGB images of the soil surface; the laser scanner is used to acquire three-dimensional point cloud data; the infrared camera is used to acquire infrared images; the Beidou positioning terminal is used to receive the positioning and navigation data of the detection robot; the soil sample data, the high-definition color system image of the soil surface, the three-dimensional point cloud data, the infrared image, and the positioning and navigation data of the current farmland soil are integrated to obtain the acquired data.

[0072] In this embodiment, due to the complexity of the sampling environment, the collected data has problems such as incompleteness and inconsistency, so it is necessary to preprocess the collected data to improve the quality of the data and ensure the integrity and accuracy of the data. The preprocessing method includes: data cleaning, data integration, data transformation and data reduction;

[0073] After preprocessing the collected data, soil surface image data, soil environment data and Beidou satellite navigation data are obtained.

[0074] S2. Label the soil surface image data to obtain the farmland surface data set. Use the YOLACT++ model to process the farmland surface data set to obtain the farmland surface detection results and soil drought classification results. The specific steps are as follows:

[0075] S201, according to the soil surface image data, the data annotation contents are divided into: image-level annotations including weed categories, debris categories, wild animal categories, pest and disease categories, crop growth cycle status categories, and drought level categories, and instance-level annotations including position coordinates of weeds and debris, crop pest and disease location coordinates, and wild animal location coordinates;

[0076] S202, using an image annotation tool to perform position annotation, category annotation, and pixel-level annotation on image-level annotation and instance-level annotation to obtain annotated soil surface image data, and integrating the annotated soil surface image data to obtain a farmland surface data set;

[0077] S203, inputting the farmland surface dataset into the YOLACT++ model, and extracting features using a feature pyramid network to obtain image features;

[0078] S204, inputting the image features into the prediction head, and obtaining the prediction features through non-maximum suppression; inputting the image features into the global mask to generate the image mask; performing feature fusion on the prediction features and the image mask to obtain the fusion features;

[0079] S205, the fused features are clipped and thresholded in sequence to obtain farmland surface detection results and soil drought classification results.

[0080] In this embodiment, in order to accurately identify weeds, debris, wild animals, pests and diseases, as well as crop growth status and drought level, etc., it is necessary to manually annotate the acquired RGB image data; the annotation content is divided into two types: image-level annotation and instance-level annotation; the image-level annotation includes: weed category, debris category, wild animal category, pest and disease category, crop growth cycle status category, drought level category, etc.; the instance-level annotation mainly includes: the location coordinates (Boundingbox) of weeds and debris, the location coordinates of crop pests and diseases, the location coordinates of wild animals, etc.; the data annotation tool Labelme is used to perform location annotation, category annotation and pixel-level annotation on the soil surface image data; the location annotation uses a rectangular box to annotate the target area, and the coordinates of the four points of the rectangular box are stored in the annotation file; the category annotation, according to the annotation file storing the location coordinates, stores the category information in the corresponding annotation file at the same time; the pixel-level annotation uses different pixel values ​​to annotate the weeds, debris, pests and diseases, and wild animal areas, and the pixel value of the location area represents the category of the corresponding object; the annotated soil surface image data is obtained, and the annotated soil surface image data is integrated to obtain the farmland surface data set;

[0081] like Figure 2 As shown in the figure, the farmland soil surface is comprehensively detected, and the farmland surface dataset is trained and processed using the YOLACT++ model; the YOLACT++ model has the advantages of high accuracy and high speed, and can detect the target object quickly and in real time; the YOLACT++ model receives the RGB image in the input farmland surface dataset, uses the FPN network to extract features, obtains image features, and inputs the image features into two branches respectively; in the first branch, according to the image features, the prediction head is used for prediction and the NMS layer is used to generate rough target areas and categories to obtain prediction features; in the second branch, the image features are processed using the global mask layer to obtain image masks; in the detection, for weeds, debris, pests and diseases, and wild animals, the categories, location coordinates, and pixel-level labels need to be detected at the same time. For the drought level and the crop growth cycle status, only the image-level category needs to be identified; the prediction features obtained by the first branch and the image mask obtained by the second branch are feature fused to obtain fused features, and the fused features are successively cropped and thresholded to obtain the farmland surface detection results and soil drought classification results, completing the comprehensive detection of farmland surface and soil drought classification.

[0082] S3. Analyze the soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, and combine the soil drought classification results to obtain accurate drought level classification. According to Beidou satellite navigation data, obtain the current spatial and temporal positioning information of farmland soil. The specific steps are as follows:

[0083] S301, analyzing soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, combining soil drought classification results and soil moisture content, and using comparative analysis to obtain accurate classification of drought levels;

[0084] S302: According to Beidou satellite navigation data, the spatiotemporal positioning information of the current farmland soil is obtained by using spatiotemporal positioning analysis.

[0085] In this embodiment, the traditional soil analysis instrument has a relatively single function. The present invention uses a soil detector that can obtain and analyze soil data in real time; the soil detector goes deep into the soil, collects data, and analyzes multiple components at the same time; the soil comprehensive detector collects and analyzes data every three meters, which can ensure the integrity and balance of farmland samples; the soil environmental data is analyzed to obtain soil moisture content, heavy metal content, and pesticide residue content, and at the same time, combined with the soil drought classification results and soil moisture content, through comparative analysis, and setting different thresholds, a more accurate farmland soil drought level is obtained;

[0086] The threshold values ​​are: a drought score between 0 and 0.4 for mild drought, a drought score between 0.4 and 0.7 for moderate drought, and a drought score between 0.7 and 1.0 for severe drought.

[0087] S4. Integrate the farmland surface detection results, accurate classification of drought levels, heavy metal content, pesticide residue content, and the current spatial and temporal positioning information of farmland soil to obtain summary data and complete the comprehensive detection of farmland soil health. The specific steps are as follows:

[0088] S401, storing the farmland surface detection results, accurate classification of drought levels, heavy metal content, pesticide residue content, and spatiotemporal positioning information of the current farmland soil in a database management system, and displaying them through data visualization;

[0089] S402: Summarize the data stored in the database management system to obtain summary data, and use the data analysis system to generate agricultural management strategies to complete comprehensive detection of farmland soil health.

[0090] In this embodiment, different test results are classified and stored in a database management system, and displayed by means of data visualization, such as a graph or a table;

[0091] Summarize the data stored in the database management system to obtain summary data. Based on the summarized test results, the data analysis system can automatically summarize and generate agricultural management strategies, such as allocating resources, selecting appropriate crop types, planning fertilization and irrigation plans, etc., to help farm managers make accurate decisions;

[0092] The present invention can conduct comprehensive and comprehensive detection of farmland soil, so the data are complex and diverse, and the data volume is large. In order to facilitate classification management, these data are classified and summarized; these data are summarized into the background management system, and the relevant management departments can conduct supervision and early warning based on them.

[0093] Example 2

[0094] In this embodiment, Figure 3 As shown, the present invention provides a farmland soil health comprehensive detection device based on deep learning, which is a farmland soil health comprehensive detection robot based on deep learning, including: a high-definition camera 1, an inclinometer 2, a motor controller 3, an automatic adjustment telescopic rod 4, a rangefinder 5, an environmental detector 6, a digital leveling device 7, a POS stable gimbal 8, a laser scanner 9, a control processing device 10, a processing terminal 11 and a Beidou positioning terminal 12;

[0095] The automatically adjustable telescopic rod 4 is located at the bottom of the farmland soil health comprehensive detection device and is equipped with wheels;

[0096] The POS stabilizing platform 8 is above the automatically adjustable telescopic rod 4;

[0097] The high-definition camera 1, inclinometer 2, motor controller 3, rangefinder 5, environmental detector 6, digital leveling device 7, laser scanner 9, control processing device 10, processing terminal 11 and Beidou positioning terminal 12 are installed on the POS stabilized gimbal 8;

[0098] The high-definition camera 1 is located at the left rear of the POS stabilized platform 8; the inclinometer 2 is located at the upper left corner of the POS stabilized platform 8;

[0099] The motor controller 3 is located on the left side of the high-definition camera 1; the rangefinder 5 is located in front of the POS stabilized platform 8; the environmental detector 6 is located in front of the right side of the POS stabilized platform 8; the digital leveling device 7 is located on the right side of the environmental detector 6 and the upper right corner of the POS stabilized platform 8;

[0100] The laser scanner 9 is located at the right rear of the POS stabilized platform 8; the control processing device 10 is located in the middle of the POS stabilized platform 8; the processing terminal 11 is located at the rear of the control processing device 10; the Beidou positioning terminal 12 is located at the right side of the processing terminal 11;

[0101] The control processing device 10 is respectively connected to the high-definition camera 1, the inclinometer 2, the motor controller 3, the rangefinder 5, the environment detector 6, the digital leveling device 7, the laser scanner 9, the processing terminal 11 and the Beidou positioning terminal 12.

[0102] In this embodiment, the high-definition camera 1 includes an array camera and an infrared camera, which are used to collect high-definition images and infrared images;

[0103] The inclinometer 2 is used to help the farmland soil health comprehensive detection device to make adaptive adjustments according to the terrain, ensuring that the device remains stable when operating on a slope or uneven area;

[0104] The motor controller 3 is used to power the farmland soil health comprehensive detection device;

[0105] The automatically adjustable telescopic rod 4 is used to automatically adjust the distance between the wheels to ensure that the farmland soil health comprehensive detection device remains stable on uneven ground or slopes;

[0106] The distance meter 5 is used to measure the distance between the farmland soil health comprehensive detection device and the target object (such as the farmland boundary, crop row, obstacle, etc.), provide accurate navigation information, and help the farmland soil health comprehensive detection device to accurately drive in complex terrain;

[0107] The environmental detector 6 is a soil detector, connected to the drilling device, and enters the soil through the drilling device, and is used to collect and analyze data every three meters, collect data, and analyze multiple components at the same time;

[0108] The digital leveling device 7 is used to adjust the center of gravity or posture of the farmland soil health comprehensive detection device to reduce the risk of rollover and ensure safe operation;

[0109] The laser scanner 9 is used to collect three-dimensional point cloud data;

[0110] The control processing device 10 is used to receive the collected data and transmit it to the processing terminal 11;

[0111] The processing terminal 11 includes a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the processing terminal executes the comprehensive detection method of farmland soil health based on deep learning;

[0112] The Beidou positioning terminal 12 is used to collect positioning and navigation data of the farmland soil health comprehensive detection device.

[0113] In this embodiment, the comprehensive detection device for farmland soil health collects data through the various sensor devices it is equipped with, and inputs the collected data into a processing terminal for data preprocessing to obtain soil surface image data, soil environmental data, and Beidou satellite navigation data; secondly, in the processing terminal, the soil surface image data is input into the existing YOLACT++ model for comprehensive farmland surface detection and soil drought classification; the soil environmental data is analyzed to output soil moisture content, heavy metal content, and pesticide residue information, wherein the soil moisture content is further combined with the soil drought classification to accurately classify the drought level; the Beidou satellite navigation data is subjected to spatiotemporal positioning analysis, and finally the obtained data are summarized to help managers conduct constructive analysis.

[0114] In Example 1, a unified comprehensive detection method for farmland soil health is implemented, which can simultaneously detect soil moisture content, heavy metal content, pesticide residue content, weeds, debris, wild animals, pests and diseases, and crop growth status, etc.; in addition, the present invention combines traditional detection methods with deep learning methods to further improve the detection accuracy; in Example 2, the present invention proposes a new type of comprehensive automatic detection device for farmland soil health, which can automatically collect soil data and analyze it in real time, saving a lot of time costs, without the need for manual intervention, and with higher efficiency and better detection effects.

Claims

1. A comprehensive detection method for farmland soil health based on deep learning, characterized in that: The following steps are involved: S1. Use a data acquisition device to collect data on the current farmland soil to obtain collected data, and pre-process the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data; S2. Label the soil surface image data to obtain the farmland surface data set, and use the YOLACT++ model to process the farmland surface data set to obtain the farmland surface detection results and soil drought classification results; S3. Analyze soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, and combine soil drought classification results to obtain accurate drought grade classification. According to Beidou satellite navigation data, obtain the current spatial and temporal positioning information of farmland soil; S4. Integrate the farmland surface detection results, precise classification of drought levels, heavy metal content, pesticide residue content, and the current spatiotemporal positioning information of farmland soil to obtain summary data and complete the comprehensive detection of farmland soil health.

2. The method for comprehensive detection of farmland soil health based on deep learning according to claim 1, characterized in that: The S1 comprises the following steps: S101. Use a soil detector to obtain soil sample data, use a high-speed linear array camera to obtain a high-definition color system image of the soil surface, use a laser scanner to obtain three-dimensional point cloud data, use an infrared camera to obtain an infrared image, and use a Beidou positioning terminal to obtain the current farmland soil positioning and navigation data; S102, integrating soil sample data, soil surface high-definition color system images, three-dimensional point cloud data, infrared images, and current farmland soil positioning and navigation data to obtain collected data; S103, performing data cleaning, data integration, data transformation and data reduction processing on the collected data to obtain soil surface image data, soil environment data and Beidou satellite navigation data.

3. The method for comprehensive detection of farmland soil health based on deep learning according to claim 1, characterized in that: The S2 comprises the following steps: S201, according to the soil surface image data, the data annotation contents are divided into: image-level annotations including weed categories, debris categories, wild animal categories, pest and disease categories, crop growth cycle status categories, and drought level categories, and instance-level annotations including position coordinates of weeds and debris, crop pest and disease location coordinates, and wild animal location coordinates; S202, using an image annotation tool to perform position annotation, category annotation, and pixel-level annotation on image-level annotation and instance-level annotation to obtain annotated soil surface image data, and integrating the annotated soil surface image data to obtain a farmland surface data set; S203, inputting the farmland surface dataset into the YOLACT++ model, and extracting features using a feature pyramid network to obtain image features; S204, inputting the image features into the prediction head, and obtaining the prediction features through non-maximum suppression; inputting the image features into the global mask to generate the image mask; performing feature fusion on the prediction features and the image mask to obtain the fusion features; S205, the fused features are clipped and thresholded in sequence to obtain farmland surface detection results and soil drought classification results.

4. The method for comprehensive detection of farmland soil health based on deep learning according to claim 1, characterized in that: The S3 comprises the following steps: S301, analyzing soil environmental data to obtain soil moisture content, heavy metal content and pesticide residue content, combining soil drought classification results and soil moisture content, and using comparative analysis to obtain accurate classification of drought levels; S302: According to Beidou satellite navigation data, the spatiotemporal positioning information of the current farmland soil is obtained by using spatiotemporal positioning analysis.

5. The method for comprehensive detection of farmland soil health based on deep learning according to claim 1, characterized in that: The S4 step is specifically as follows: S401, storing the farmland surface detection results, accurate classification of drought levels, heavy metal content, pesticide residue content, and spatiotemporal positioning information of the current farmland soil in a database management system, and displaying them through data visualization; S402: Summarize the data stored in the database management system to obtain summary data, and use the data analysis system to generate agricultural management strategies to complete comprehensive detection of farmland soil health.

6. A comprehensive farmland soil health detection device based on deep learning, characterized in that: include: A high-definition camera (1), an inclinometer (2), a motor controller (3), an automatically adjustable telescopic rod (4), a rangefinder (5), an environmental detector (6), a digital leveling device (7), a stabilizing pan / tilt of a positioning and orientation system (8), a laser scanner (9), a control processing device (10), a processing terminal (11), and a Beidou positioning terminal (12); The automatically adjustable telescopic rod (4) is located at the bottom of the farmland soil health comprehensive detection device and is equipped with wheels; The positioning and orientation system stabilizing platform (8) is located above the automatically adjustable telescopic rod (4); The high-definition camera (1), inclinometer (2), motor controller (3), rangefinder (5), environmental detector (6), digital leveling device (7), laser scanner (9), control processing device (10), processing terminal (11) and Beidou positioning terminal (12) are installed on a stabilizing gimbal (8) of the positioning and orientation system; The high-definition camera (1) is located at the left rear of the positioning and orientation system stable platform (8); the inclinometer (2) is located at the upper left corner of the positioning and orientation system stable platform (8); The motor controller (3) is located on the left side of the high-definition camera (1); the rangefinder (5) is located directly in front of the positioning and orientation system stable platform (8); the environmental detector (6) is located in front of the right side of the positioning and orientation system stable platform (8); the digital leveling device (7) is located on the right side of the environmental detector (6) and at the upper right corner of the positioning and orientation system stable platform (8); The laser scanner (9) is located at the right rear of the positioning and orientation system stable platform (8); the control processing device (10) is located in the middle of the positioning and orientation system stable platform (8); the processing terminal (11) is located at the rear of the control processing device (10); and the Beidou positioning terminal (12) is located at the right side of the processing terminal (11); The control processing device (10) is respectively connected to the high-definition camera (1), the inclinometer (2), the motor controller (3), the rangefinder (5), the environment detector (6), the digital leveling device (7), the laser scanner (9), the processing terminal (11) and the Beidou positioning terminal (12).

7. The deep learning-based comprehensive detection device for farmland soil health according to claim 6 is characterized in that: The environmental detector (6) is a soil detector connected to the drilling device, and enters the soil through the drilling device, and is used to collect and analyze data every three meters, collect data, and analyze multiple components at the same time; The laser scanner (9) is used to collect three-dimensional point cloud data; The high-definition camera (1) comprises an array camera and an infrared camera, and is used to collect high-definition images and infrared images; The Beidou positioning terminal (12) is used to collect positioning and navigation data of the farmland soil health comprehensive detection device; The control processing device (10) is used to receive the collected data and transmit it to the processing terminal (11); The processing terminal (11) comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the processing terminal executes the comprehensive detection method for farmland soil health based on deep learning; The motor controller (3) is used to supply power to the farmland soil health comprehensive detection device.

Images