Agricultural mulching film residue intelligent detection system and method

Abstract

The invention is suitable for the technical field of agricultural environment monitoring, and provides an intelligent detection system and method for agricultural mulching film residues, and the method comprises the steps: carrying out the multi-source data collection of an unmanned aerial vehicle and ground mobile equipment through a standardized interface, and sequentially carrying out the preprocessing of Gaussian filtering denoising, histogram equalization enhancement and perspective transformation correction on an image; based on a crop-terrain-soil scene parameter template, utilizing a deep learning model to automatically identify a mulching film residue area, and combining an area conversion coefficient and an area density database to accurately estimate the residual quantity per unit area; performing quality evaluation and feedback correction through three dimensions of data consistency, accuracy and integrity; data are stored in a distributed encryption mode, a result is visually displayed through a three-dimensional map, and trend analysis, report generation and self-defined threshold early warning are supported. According to the invention, full-process automation and multi-scene accurate adaptation of mulching film residue detection are realized, and the detection efficiency, precision and data security are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural environmental monitoring technology, specifically to an intelligent detection system and method for agricultural plastic film residue. Background Technology

[0002] With the acceleration of agricultural modernization, agricultural mulch film is widely used in agricultural production because it can effectively increase soil temperature, maintain soil moisture, and suppress weed growth. However, because the material of mulch film (such as polyethylene) is difficult to degrade naturally, a large amount of residual mulch film accumulates in the soil over a long period of time, which damages soil structure, hinders soil aeration and permeability, affects crop root growth, and leads to a decline in soil quality and reduced crop yields, becoming a significant environmental problem restricting the sustainable development of agriculture. Accurately detecting the amount of residual mulch film in the soil and understanding the distribution patterns and trends of residues are key prerequisites for solving the problem of mulch film residue pollution. Traditional mulch film detection mainly relies on manual sampling and laboratory analysis, which is not only time-consuming and labor-intensive with low detection efficiency, but also has a limited sampling range, making it difficult to comprehensively reflect the true situation of mulch film residue in large areas of farmland. Existing similar technologies include UAV remote sensing-based detection of residual plastic film, which consists of a UAV data acquisition unit, a simple image preprocessing unit, a human-assisted identification unit, and a data recording unit. The UAV acquires surface images, which are then preprocessed and manually marked to indicate areas of residual plastic film. The residual area is estimated and recorded based on the aerial photography parameters. Similar technologies also include ground-based manual detection technology, which determines the amount of residual film through on-site sampling and manual sorting and weighing in the laboratory.

[0003] Existing technologies have many shortcomings:

[0004] The detection efficiency is low, and it relies too much on manual intervention in the core identification and analysis process, making it impossible to achieve full-process automation.

[0005] The detection accuracy is poor, it is greatly affected by environmental and subjective human factors, and it lacks the ability to adapt to multiple scenarios and a quality assessment mechanism.

[0006] It has limited functionality, only capable of identification or area estimation, and cannot meet the needs of full-process monitoring and decision support.

[0007] It has poor adaptability, uses a single data collection method, and its algorithms are not optimized for different scenarios, making it difficult to cope with complex agricultural scenarios;

[0008] Insufficient data security and manageability; storage is not encrypted; there is no backup and recovery mechanism and no sound access control.

[0009] Therefore, in view of the above situation, there is an urgent need to provide an intelligent detection system and method for agricultural mulch film residue to overcome the shortcomings in current practical applications. Summary of the Invention

[0010] The purpose of this invention is to provide an intelligent detection system and method for agricultural mulch film residue, aiming to solve the problems mentioned in the background art.

[0011] This invention is implemented as follows: an intelligent detection system and method for agricultural plastic film residue, comprising the following steps:

[0012] Multi-source data acquisition and preprocessing: Image data is acquired by connecting to UAVs or ground mobile devices through standardized interfaces, the acquired raw images are preprocessed, and the preprocessed images are standardized.

[0013] Mulch film identification, residual amount estimation and quality assessment: Based on the user-selected scenario, the corresponding parameter template is loaded, and a deep learning model is used to identify the residual mulch film area; combined with the image pixel-actual area conversion coefficient and mulch film surface density database, the residual amount of mulch film per unit area is calculated; and the identification and estimation results are assessed from three dimensions: data consistency, accuracy and data integrity. If the assessment does not meet the standards, it is fed back to the identification stage for correction.

[0014] Data management, results display and early warning: The detection data is classified and stored in a distributed database according to the plot number and detection time and is encrypted; data backup and recovery functions are provided; 3D maps are built using WebGL technology for visualization, generating residual trend analysis charts and standardized reports; users can set early warning thresholds and trigger early warnings in real time when data exceeds the limit.

[0015] As a further aspect of the present invention: the preprocessing includes:

[0016] The acquired raw images were sequentially processed by Gaussian filtering for noise reduction, histogram equalization for enhancement, and perspective transformation for correction.

[0017] As a further aspect of the present invention: the data standardization includes:

[0018] The preprocessed image data is converted into a unified RGB color mode, preset pixel resolution, and encoding format;

[0019] The files are named and associated with metadata according to the rule of "plot number_collection time_collection device type".

[0020] As a further aspect of the present invention: the parameter template is pre-set based on a combination of crop type, terrain features and soil conditions, and the parameter template includes image preprocessing parameter adjustment values, recognition algorithm thresholds and estimation model correction coefficients;

[0021] Administrators can upload new scene parameters via an interface to update the template library.

[0022] As a further aspect of the present invention: the identification of residual plastic film areas using a deep learning model specifically involves:

[0023] Based on the ResNet-50 model, combined with the loaded scene parameter template, the feature extraction threshold and classification model parameters are dynamically adjusted to classify the extracted mulch film features and generate a mask image of the residual area.

[0024] As a further aspect of the present invention: the formula for calculating the residual amount of plastic film per unit area is:

[0025] ;

[0026] in, This is the actual residual area area calculated using pixel area and conversion factor. To match the surface density of the mulch film obtained from the mulch film surface density database, This represents the total area of ​​the land parcel.

[0027] As a further aspect of the present invention: the three dimensions of the quality assessment include:

[0028] Compare the identification results of drones and ground mobile devices on the same plot of land to assess data consistency;

[0029] Calculate the recognition accuracy and estimate the error to evaluate the accuracy;

[0030] Check that the metadata is complete to assess data integrity.

[0031] As a further aspect of the present invention: the data backup and recovery function supports automatic periodic backup and manual instant backup, and the backup files are in encrypted format;

[0032] During data recovery, data consistency is verified by calculating and comparing the MD5 value of the recovered data with that of the backup file.

[0033] As a further aspect of the present invention: the three-dimensional map uses green, yellow and red gradients to indicate different levels of residual plastic film, and supports users to rotate, zoom and pan the view through the interface.

[0034] The present invention also provides an intelligent detection system for agricultural mulch film residue, for performing the intelligent detection method for agricultural mulch film residue as described above.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] This invention constructs a fully automated technology chain, eliminating the need for manual intervention in core processes, significantly improving detection efficiency, reducing labor and time costs, and meeting the needs of large-scale farmland detection. Through multi-scenario adaptable parameter templates, the ResNet-50 model, and a three-dimensional quality assessment mechanism, it improves detection accuracy and ensures more reliable data. It includes 12 built-in scenario templates with expansion support, overcoming the limitations of traditional technologies and adapting to complex agricultural scenarios. Data is stored and categorized by dimension, combined with visualization, trend analysis, and standardized reports to deeply explore data value and support scientific decision-making. An encrypted storage, intelligent backup and recovery, and three-level access control architecture strengthen data security and management, ensuring data continuity and traceability. The automated operation process and intuitive interface lower the barrier to entry and facilitate widespread application. Detailed Implementation

[0037] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The present invention will be further explained below with reference to specific embodiments.

[0039] The present invention provides an intelligent detection system and method for agricultural mulch film residue, wherein the intelligent detection method for agricultural mulch film residue includes:

[0040] Step 1: Multi-source data acquisition and preprocessing

[0041] 1.1 Multi-source data acquisition

[0042] Device Connection and Parameter Settings: Users select the data acquisition method (drone / ground mobile device) through the system interface. The system automatically establishes communication with the corresponding hardware through standardized interfaces (drone aerial photography interface: http: / / ip:port / agriculture_system / v1 / drone_aerial; ground mobile device acquisition interface: http: / / ip:port / agriculture_system / v1 / ground_device). If drone acquisition is selected, the user needs to input the flight altitude (unit: meters), shooting angle (unit: degrees), and shooting range (rectangular area coordinates) on the interface. If ground mobile device acquisition is selected, the user needs to select the device model and upload the preset acquisition route (coordinate point sequence).

[0043] Data Acquisition Triggering and Feedback: When the user clicks the "Start Acquisition" button, the system sends a control signal to the hardware, triggering the device to start image acquisition. During the acquisition process, the hardware transmits the raw image data (JPEG / PNG format) back to the system in real time as a binary stream. The system automatically stores the data temporarily in a temporary database and records metadata such as acquisition time and device identifier.

[0044] 1.2 Image Preprocessing

[0045] 1.2.1 Gaussian filtering for noise reduction: The system calls a preprocessing algorithm and loads a 3×3 Gaussian kernel ( (Value 1.0), based on the formula:

[0046]

[0047] in, Gaussian filter in coordinate The weight value or kernel function value at that location, The horizontal and vertical distances (in pixels) of the current calculation point relative to the center of the Gaussian kernel. Let be the standard deviation of the Gaussian distribution. This is the normalization coefficient.

[0048] The original image is filtered to remove Gaussian noise and salt-and-pepper noise while preserving the detailed features of the mulch film area.

[0049] 1.2.2 Histogram Equalization Enhancement: To address the insufficient contrast of the denoised image, the system performs histogram equalization on the grayscale image f(c,y) (grayscale level L), using the following transformation function:

[0050] ,(in For the original image grayscale value, The grayscale value is The number of pixels, The grayscale histogram is corrected (total number of pixels in the image) to make the grayscale values ​​more evenly distributed and enhance the color difference between the mulch film and the soil.

[0051] 1.2.3 Perspective Transformation Correction: Based on the intrinsic parameters (focal length, pixel size) and extrinsic parameters (shooting angle, position) of the acquisition device, calculate the perspective transformation matrix parameters. Based on the formula:

[0052]

[0053] (where (x,y) are the pixel coordinates of the original image, and (u / w,v / w) are the pixel coordinates of the corrected image), correcting the geometric distortion of the image caused by the shooting angle to ensure the accurate positioning of the mulch film area.

[0054] 1.3 Data Standardization

[0055] Unified format conversion: The system converts the pre-processed image data into a unified format of "RGB color mode + [X]×[Y] pixel resolution + [specific encoding format]" according to preset standards, eliminating the format differences of data collected by different devices.

[0056] File naming and metadata recording: Standardized image files are named according to the rule of "plot number_collection time_collection equipment type" (such as "001_20230915_drone"), and metadata such as plot number, collection time, equipment type, and crop type are recorded together.

[0057] Integrity verification: The system checks whether the image content is complete (no missing pixels) and whether the metadata is complete. If the verification is successful, the standardized data and metadata are output to the film identification and analysis module simultaneously. If the verification fails, it returns to the data acquisition stage and prompts the user to re-acquire the data.

[0058] Step 2: Identification of plastic film, estimation of residual amount and quality assessment

[0059] 2.1 Multi-scenario parameter adaptation

[0060] Scene template selection and loading: Users select the scene corresponding to the current farmland in the system interface (covering 3 types of crops: corn, wheat, and vegetables; 2 types of terrain: plains and mountains; 2 types of soil: sandy loam and clay, for a total of 12 combined scenes). The system automatically loads the corresponding scene template from the database. The template includes "image preprocessing parameter adjustment values" (such as edge detection enhancement parameters for mountain scenes), "recognition algorithm thresholds" (such as the RGB distinction threshold between mulch and soil in clay scenes), and "estimation model correction coefficients" (such as the crop shading area correction coefficient for vegetable fields).

[0061] Scene parameter update (administrator privileges): If it is necessary to adapt to new crops or new soil types, the administrator can upload new scene parameters through the "Multi-scene Adaptation Data Update Interface" (http: / / ip:port / agri-system / 1.0 / multi-scene-update). The system will automatically update the template library and synchronize it to the recognition and estimation stages.

[0062] 2.2 Identification of residual plastic film

[0063] 2.2.1 Feature Extraction: Based on the loaded scene parameters, the system extracts key features of the mulch film from the standardized image—texture features (such as the smooth texture pixel distribution of polyethylene mulch film) and color features (such as the RGB value range of white mulch film [230-255, 230-255, 230-255] and the RGB value range of black mulch film [0-20, 0-20, 0-20]), while excluding interfering features such as crop residues and soil impurities.

[0064] 2.2.2 Deep Learning Classification: A ResNet-50-based classification model is used, consisting of 49 convolutional layers and 1 fully connected layer, using the residual connection formula. ,(in Input for residual blocks, These are the weight parameters of the convolutional layer. The output is the nonlinear transformation output of the residual block. (For the final output of the residual block) This addresses the vanishing gradient problem in deep networks. The model classification threshold is adjusted based on scene parameters to classify the extracted features, mark the coordinates and contours of the residual film area, and generate a mask image of the residual area.

[0065] 2.3 Residual Amount Estimation

[0066] 2.3.1 Actual Area Conversion: The system reads the parameters of the acquisition device (such as the image pixels corresponding to the UAV's flight altitude - actual area conversion factor K), and combines them with the pixel area of ​​the residual plastic film area in the masked image. , through the formula S= ×K calculates the actual residual area. (Unit: square meters)

[0067] 2.3.2 Calculation of residual amount per unit area: Call the "Mulch Film Surface Density Database" (which stores the surface density of mulch films of different thicknesses). (Unit: g / m²), matching the areal density parameter of the currently detected mulch film, based on the formula:

[0068] ;(in Calculate the amount of residual plastic film per unit area (total area of ​​the plot, unit: square meters). (Unit: g / m²), supports automatic conversion to common units of measurement such as "kg / hectare".

[0069] 2.4 Quality Assessment and Feedback Correction

[0070] Three-dimensional evaluation: The system verifies the results from three dimensions: ① Data consistency: Compare the recognition results of drones and ground mobile devices on the same plot of land. If the area deviation exceeds 5%, it is marked as abnormal; ② Accuracy standard: Calculate the recognition accuracy (number of correctly recognized plastic film pixels / actual number of plastic film pixels) and the estimation error (|calculated residual amount - manually sampled and verified residual amount| / manually sampled and verified residual amount). If the accuracy is lower than 90% or the error exceeds 8%, it is judged as unqualified; ③ Data integrity: Check whether it contains all metadata such as plot identification, detection time, and equipment parameters. If any item is missing, it is judged as incomplete.

[0071] Feedback and correction: If the evaluation fails to meet the standard, the system will automatically feed back to the plastic film residue identification unit, reload the scene parameters (such as adjusting the RGB distinction threshold and model classification weight), and repeat the "feature extraction-deep learning classification" process; if the evaluation meets the standard, the identification results, residue data and metadata will be output to the data management and storage module.

[0072] Step 3: Data Management, Results Display, and Early Warning

[0073] 3.1 Data Classification, Storage, and Security Management

[0074] Categorized Storage: The system adopts a two-tier architecture of "distributed database (MySQL cluster) + local backup". A primary directory is established based on "plot number", and each primary directory has secondary directories based on "detection time (year-month-day)". Detection data (image data, identification results, and residual data) are categorized and stored in the corresponding directories. The database table structure contains 10 core fields (plot identifier, timestamp, residual plastic film, detection equipment identifier, crop type, terrain features, soil conditions, image data path, data source identifier, and detection personnel identifier), supporting multi-dimensional queries.

[0075] Data encryption and access control: The database data is encrypted using the AES-256 algorithm (key length 256 bits, encryption process includes round key addition, byte substitution, row shifting, and column mixing operations). At the same time, three levels of user permissions are set (administrator: full data access and modification; inspector: only access to data of the plots under their responsibility; ordinary user: only view public statistical data). Users can obtain the corresponding data only after their permissions are verified through the "data query interface" (http: / / ip:port / agri-system / data-query).

[0076] 3.2 Data Backup and Recovery

[0077] Backup operations: The system supports "automatic backup + manual backup":

[0078] Automatic backups are performed according to the user-defined schedule (daily / weekly / monthly). The database data is compressed into an encrypted backup file (.dat format) and stored on the local backup server through the "Data Backup Interface" (http: / / ip:port / agriculture_system / 1.0 / data_backup). Users can trigger an immediate backup and generate a backup log by clicking the "Manual Backup" button in the system backend.

[0079] Recovery Operation: When data is lost (e.g., database corruption), users can select a backup file through the "Data Recovery Interface" (http: / / ip:port / agriculture_system / 1.0 / data_restore). After the system performs the recovery operation, it automatically calculates and compares the MD5 value (128-bit hash value) of the recovered data with that of the backup file. If they match, the recovery is considered successful, and the database is updated synchronously. If they do not match, the system will prompt "Backup file is corrupted" and guide the user to select another backup file.

[0080] 3.3 Visualization of Results

[0081] 3D Visualization Map: The system is based on WebGL technology (rendered in HTML5 Canvas via OpenGL ES 2.0 API), integrating plot terrain data (elevation, slope) and residual plastic film data (residual amount, location) to construct a 3D map model. Users can perform rotation, zoom, and pan operations to adjust the map view through the "view operation interface" (http: / / ip:port / agro_system / v1 / view_operation). When the mouse hovers over any area, details such as the residual amount and detection time at that location are automatically displayed, and the residual amount level is indicated by a three-color gradient (green: 0-20g / m³). 2 Slight residue; yellow: 20-50g / m³ 2 Moderate residue; Red: ≥50g / m³ 2 , heavy residue).

[0082] Residual Trend Analysis: Users select the target plot and time range (e.g., January 2023 - December 2023). The system obtains residual data at multiple time points through the "Historical Data Query Interface" (http: / / ip:port / agriculture-system / v1 / historical-data). A linear regression algorithm (regression equation (y=ax+b), where y is the residual amount, x is the time variable, a is the rate of change coefficient, and b is the intercept, with the values ​​of a and b calculated using the least squares method) is used to generate a residual trend line chart. This supports trend comparison between different years for the same plot and between different plots during the same period, assisting in the evaluation of the effectiveness of plastic film removal measures.

[0083] 3.4 Test Report Generation and Early Warning Prompts

[0084] Standardized report generation: The system has a built-in report template (containing four sections: "Basic Information of the Land Parcel," "Summary of Monitoring Data," "Residual Trend Charts," and "Recommended Measures"). Monitoring data and chart data are obtained through the "Monitoring Report Data Interface" (http: / / ip:port / agriculture_system / report_data) and automatically populated into the template. Users can select customization options such as "Whether to add 3D map screenshots" and "Whether to include historical comparison data" on the interface. After clicking the "Export" button, the system generates a PDF report, which can be saved to a specified path or downloaded directly.

[0085] Real-time alerts: Users can set residual amount alert thresholds (e.g., 50g / ㎡) and notification methods (system pop-up, SMS, email) for target plots through the "Alert Threshold Setting Interface" (http: / / ip:port / agri_system / warning / set). The system monitors new detection data in real time. If the residual amount exceeds the threshold, it automatically triggers the "Alert Notification Interface" (http: / / ip:port / agri_system / warning / notify) to send a notification message (including plot identifier, excess amount, and suggested cleanup measures) to the designated user. Simultaneously, it stores alert records (threshold setting time, excess time, and processing result), allowing users to query historical alerts and trace the causes of excesses.

[0086] This invention also provides an intelligent detection system for agricultural mulch film residue, used to perform the intelligent detection method for agricultural mulch film residue as described above.

[0087] In summary, the present invention has the following advantages:

[0088] A fully automated mulch film detection technology system: This system constructs a fully automated technology chain from "multi-source data acquisition → image preprocessing → mulch film identification → residual amount estimation → data storage → result output," eliminating the need for manual intervention in core processes. Standardized interfaces enable automatic communication between hardware and software. The image preprocessing unit automatically performs noise reduction, enhancement, and correction operations, while the mulch film identification and analysis module automatically completes area identification and quantitative estimation, solving the problems of traditional detection methods that rely on manual labor and are inefficient.

[0089] Accurate detection mechanism adaptable to multiple scenarios: For the diversity of agricultural scenarios (different crop types, terrain features, soil conditions), it has built-in exclusive detection parameter templates. Through multi-scenario adaptation units, it provides dynamic adaptation parameters for mulch film identification and residue estimation, ensuring stable detection results in different scenarios and breaking through the limitation of traditional technologies that can only adapt to a single simple scenario.

[0090] Secure and controllable data management architecture: Adopting a three-tier data management architecture of "distributed classified storage + encrypted backup + access control": The detection data is classified and stored according to the "plot-time" dimension for easy and quick query; the data is encrypted with AES-256 algorithm, and is automatically backed up regularly and provided with a fast recovery function; Three-level user permissions are set, and different roles can only access data within their access scope to ensure the security, integrity and traceability of agricultural production data.

[0091] An intuitive and efficient results output and early warning system: The system features four main output functions: 3D visualization, residue trend analysis, standardized report generation, and real-time early warning. It utilizes WebGL technology to construct a 3D map that intuitively presents the spatial distribution of residues, generates residue change curves based on historical data to aid trend judgment, automatically generates PDF reports containing key indicators and charts, and supports user-defined early warning thresholds. When thresholds are exceeded, real-time alerts are sent via system pop-ups, SMS messages, and other methods, providing direct support for agricultural decision-making.

[0092] Innovative multi-source data acquisition and preprocessing technologies:

[0093] Multi-source data collaborative acquisition technology:

[0094] The technical solution protects the integration of drone aerial photography and ground mobile equipment, and achieves automatic data acquisition and transmission between the two types of equipment through standardized interfaces (drone aerial photography interface: http: / / ip:port / agriculture_system / v1 / drone_aerial; ground mobile equipment acquisition interface: http: / / ip:port / agriculture_system / v1 / ground_device), thus solving the problem that it is difficult to balance coverage and accuracy with a single acquisition method.

[0095] Integrated image preprocessing technology: This technology protects the "integrated preprocessing flow of sequentially performing Gaussian filtering for noise reduction, histogram equalization enhancement, and perspective transformation correction," as well as the technical means of "dynamically adjusting the correction algorithm parameters according to the acquisition device parameters (such as the shooting angle)," ensuring the output of high-quality image data and laying the foundation for subsequent recognition.

[0096] Multi-scenario adaptable plastic film identification and residual amount estimation technology:

[0097] Contextualized plastic film identification technology: This technology protects the technical solution of "based on the ResNet-50 deep learning model, combined with built-in crop-terrain-soil scene parameter templates (such as corn-mountain-sand loam template, wheat-plain-clay template), dynamically adjusting feature extraction thresholds and classification model parameters to achieve automatic identification of plastic film residue areas in multiple scenarios". The core lies in the collaborative adaptation of scene parameters and deep learning algorithms.

[0098] Precise residual amount estimation technology: This technology protects the technical means of "calculating the residual amount of mulch film per unit area (unit: g / ㎡) based on the 'area-weight conversion model', combined with the image pixel-actual area conversion coefficient and the mulch film surface density database, and supports switching between multiple measurement units". The key lies in the correlation application of the conversion coefficient and the surface density database, which realizes the breakthrough from "qualitative identification" to "quantitative estimation".

[0099] The full-process quality assessment and feedback correction mechanism protects the technical solution of "assessing the quality of the test results from three dimensions: data consistency (comparing results from different acquisition methods), accuracy (identification accuracy and estimation error), and data integrity (verifying whether metadata is missing). If the results do not meet the standards, feedback is sent to the identification unit to reload the adaptation parameters for processing." This mechanism ensures the reliability of the test results and is different from the traditional technology design that lacks result verification.

[0100] Secure and controllable data management and storage technologies:

[0101] Categorized storage and encryption technology: The technical solution of "building a distributed database storage structure according to the dimension of 'plot number-detection time', encrypting data with the AES-256 algorithm, and naming data according to the rule of 'plot number_collection time_collection device type'" ensures orderly management and secure storage of data.

[0102] Intelligent backup and recovery technology: This technology protects data by supporting automatic / manual backups, using encrypted backup files (.dat), and automatically verifying data consistency during recovery (by comparing MD5 values), thus solving the problems of easy data loss and difficult recovery associated with traditional technologies.

[0103] Intuitive results output and real-time early warning technology:

[0104] 3D visualization display technology: The protection solution is based on WebGL technology, integrates terrain data and residual data to build a 3D map, uses a three-color gradient (green / yellow / red) to identify the residual amount level, and supports view rotation, zoom, and hover to view details, so as to realize the intuitive presentation of residual distribution.

[0105] Customizable early warning technology: This technology protects users by allowing them to customize early warning thresholds and notification methods (pop-up / SMS / email) for land parcels via the early warning threshold setting interface (http: / / ip:port / agri_system / warning / set). The system monitors data in real time and automatically triggers the early warning notification interface (http: / / ip:port / agri_system / warning / notify) to send information when limits are exceeded. The key lies in the combination of customizable thresholds and multiple notification methods to achieve timely control of residual exceedances.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent detection of mulch film residue on agricultural land, characterized in that, The method comprises the following steps: Multi-source data acquisition and preprocessing: connect unmanned aerial vehicles or ground mobile devices through standardized interfaces for image data acquisition, preprocess the acquired raw images, and standardize the data of the preprocessed images; Mulch identification, residue estimation and quality evaluation: load the corresponding parameter template based on the user's selected scene, identify the mulch residue area using a deep learning model; calculate the unit area mulch residue based on the image pixel-actual area conversion coefficient and the mulch surface density database; and evaluate the identification and estimation results from three dimensions of data consistency, accuracy and data integrity, and if the evaluation is not up to standard, feedback to the identification link for correction processing; Data management, result display and early warning: store the detection data classified by plot number and detection time to a distributed database and encrypt it; provide data backup and recovery functions; build a three-dimensional map through WebGL technology for visual display, generate residue trend analysis charts and standardized reports; support users to set warning thresholds and trigger real-time warnings when data exceeds the thresholds. 2.The method according to claim 1, wherein, The preprocessing comprises: The acquired raw images are sequentially subjected to Gaussian filter denoising, histogram equalization enhancement and perspective transformation correction. 3.The method according to claim 1, wherein, The data standardization comprises: The preprocessed image data is converted to a unified RGB color mode, a preset pixel resolution and an encoding format; And the file is named and metadata associated according to the rule "plot number_collection time_collection device type". 4.The method according to claim 1, wherein, The parameter template is pre-set based on the combination of crop type, terrain feature and soil condition, and the parameter template includes image preprocessing parameter adjustment value, identification algorithm threshold and estimation model correction coefficient; The administrator can upload new scene parameters through the interface to update the template library.

5. The method of claim 1, wherein, The deep learning model is used to identify the mulch residue area, specifically: Based on the ResNet-50 model, combined with the loaded scene parameter template, the feature extraction threshold and classification model parameters are dynamically adjusted, the extracted mulch features are classified, and a residual area mask image is generated. 6.The method of claim 1, wherein, The formula for calculating the unit area mulch residue is: ; wherein, is an actual residual area calculated from the pixel area and the conversion factor, is a mulch surface density matched from a mulch surface density database, is the total area of the plot.

7. The mulch film residue intelligent detection method according to claim 1, characterized in that, The three dimensions of the quality evaluation include: Compare the identification results of unmanned aerial vehicles and ground mobile devices in the same plot to evaluate data consistency; Calculate the identification accuracy and estimation error to evaluate accuracy; Check if the metadata is complete to evaluate data integrity. 8.The method according to claim 1, wherein, The data backup and recovery function supports automatic periodic backup and manual immediate backup, and the backup file is in encrypted format; When data is recovered, the MD5 values of the recovered data and the backup file are calculated and compared to verify data consistency. 9.The method of claim 1, wherein, The three-dimensional map uses green, yellow and red three-color gradient to identify different mulch residue levels, and supports user's rotation, scaling and translation operations through the interface.

10. An intelligent detection system for mulch film residue on agricultural land, characterized by, A method for intelligent detection of agricultural mulch residue is provided.

Images