Method and device for measuring sediment concentration and sand grain grading in irrigation water by using AI image classification technology

Abstract

The invention discloses a method and a device for measuring sediment concentration and sand grain grading in irrigation water by utilizing an AI image classification technology. The method comprises the following steps: S1, shooting image data containing an identification mark through an underwater camera; s2, based on the image data, an AI image classification model is used for analysis, and the sediment concentration is obtained; s3, based on the sediment concentration and light sensation data of the light sensation sensor, a sand grain grading model is used for analysis, and sand grain grading data are obtained; and S4, storing and uploading the sediment concentration and the sand grain grading data to a cloud. The invention provides the sediment concentration and sand grain grading measurement method with low cost and high precision. The identification precision of an AI image classification model is improved by introducing an identification mark, and an expensive special sensor is replaced by a model calibrated by a traditional instrument. According to the method, real-time monitoring of sediment data in an irrigation water source and a pipe network can be achieved, and key data support is provided for a water and fertilizer integrated automatic regulation and control system.

Description

Technical Field

[0001] This invention relates to the field of water-saving irrigation, specifically to a method and apparatus for measuring sediment concentration and sand particle size distribution in irrigation water using AI image classification technology. Background Technology

[0002] In recent years, the development of integrated water and fertilizer technology has been rapid, and the automation, remote control, and real-time regulation systems of irrigation systems have also made significant progress. Integrated water and fertilizer regulation technology is gradually moving towards precision and micro-control. The concentration and gradation of sediment in micro-sprinkler irrigation water sources and pipe networks have always been important factors affecting pipe network blockage. However, the measurement of these indicators relies heavily on laboratory methods using dryers, laser sediment gradation analyzers, etc., which lack real-time accuracy. Some researchers have improved the efficiency of concentration data acquisition by using portable turbidimeters to measure sediment concentration on-site, but this measurement still cannot meet the requirements for real-time and long-term data monitoring, and therefore cannot provide real-time precision control based on changes in sediment concentration and gradation in irrigation water. Therefore, accurate and real-time monitoring of the long-term dynamic changes in sediment concentration and gradation, especially in integrated water and fertilizer precision irrigation systems, remains a challenge.

[0003] Advances in the field of computer science have led to methods for generating, storing, and computing digital images, and image classification technology is increasingly being applied to agricultural irrigation. Traditional image classification techniques utilize deep learning (convolutional neural network models); emerging image classification technologies are based on AI technology, or artificial intelligence, which automatically captures images after identifying the objects to be captured according to the image acquisition program.

[0004] The following problems also exist in AI image classification of irrigation water:

[0005] 1) Data on sediment concentration and sand particle size distribution rely on sampling and measurement through laboratory methods, which has a long measurement cycle and cannot be monitored in real time.

[0006] 2) Although turbidity sensors can acquire irrigation water turbidity in real time and convert it into concentration data, they are costly and have a limited measurement range. Summary of the Invention

[0007] In view of the above, the purpose of this invention is to provide a device for calculating sediment concentration and sand particle size distribution by using AI image classification sediment concentration and sand particle size distribution measurement model based on the light intensity, transmittance and refractive index α data of an underwater camera.

[0008] To achieve the above objectives, the present invention provides a method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology, comprising:

[0009] S1. Capture image data containing identification markers using an underwater camera;

[0010] S2. Based on the image data, analyze it using an AI image classification model to obtain the sediment concentration;

[0011] S3. Based on the sediment concentration and the light-sensing data from the light sensor, the sand particle size distribution model is used for analysis to obtain sand particle size distribution data.

[0012] S4. Store the sediment concentration and sand particle size distribution data and upload them to the cloud.

[0013] Preferably, the identification mark is a preset pattern located around the focal length of the underwater camera, and the blurriness of the image of the identification mark reflects the turbidity of the water.

[0014] Preferably, S2 includes:

[0015] A deep convolutional neural network is used to analyze the preprocessed image data to obtain the changing trend of ambiguity in the recognition marker region;

[0016] Based on the correlation between the changing trend of ambiguity and the light transmittance of water bodies, an AI image classification model for sediment concentration was constructed and trained. The trained AI image classification model for sediment concentration was then used to predict image data to obtain sediment concentration.

[0017] Preferably, the output of the sediment concentration AI image classification model is the corrected turbidity T′, and the sediment concentration C is calculated using a weighted superposition model. The weighted superposition model combines the corrected turbidity T′ and the equipment immersion depth H, and the calculation formula is as follows:

[0018]

[0019] Where, k T ω1 and ω2 represent the turbidity sensor correction coefficients; ω1 and ω2 represent the weighting coefficients, calculated using the entropy weighting method.

[0020] Preferably, S3 includes:

[0021] Based on the sediment concentration and light intensity data obtained by the light sensor, the initial particle size data of the sand particles are calculated through the sand particle size distribution model.

[0022] The final sand grain size distribution data is obtained by iterative calculation based on the initial particle size data and continuously updated light intensity data.

[0023] Preferably, the iterative calculation process of the sand grain gradation model includes:

[0024] Light intensity data is continuously acquired and updated at preset time intervals;

[0025] Based on the current sediment concentration and the updated light intensity data, the sand particle size distribution parameters are iteratively calculated using a formula that includes attenuation relationships.

[0026] After completing a predetermined number of iterations, the average of all calculated results is taken to obtain the final sand grain size distribution data.

[0027] Preferably, S4 includes:

[0028] Store sediment concentration and sand particle size distribution data in a local database;

[0029] Data from the local database is regularly uploaded to the cloud platform to enable remote storage, sharing, and monitoring of the data.

[0030] The present invention also provides a device for measuring the sediment concentration and sand particle size distribution in irrigation water using AI image classification technology. The device is used to implement the above method and includes: a camera, a fixed bracket, a light sensor, and an identification mark.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] This invention improves the recognition accuracy of AI image classification models by introducing identification markers and replaces expensive dedicated sensors with models calibrated using traditional instruments. This invention enables real-time monitoring of sediment data in irrigation water sources and pipe networks, providing crucial data support for integrated water and fertilizer automatic control systems. Attached Figure Description

[0033] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of an AI image classification device according to an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of the AI ​​image classification device according to an embodiment of the present invention;

[0037] Figure 4 This is an AA cross-sectional view of the AI ​​image classification device according to an embodiment of the present invention;

[0038] Figure 5 This is a detailed view of the AI ​​image classification device according to an embodiment of the present invention.

[0039] Explanation of reference numerals in the attached figures:

[0040] 1. Underwater camera; 2. Light sensor; 3. Identification marker; 4. Mounting bracket. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 The diagram shown is a schematic representation of the method flow in this embodiment, and the steps include:

[0044] S1. Capture image data containing identification markers using an underwater camera.

[0045] More than 200 images containing identification markers were captured using an underwater camera on the support structure. The images underwent preprocessing, including:

[0046] In this embodiment, the identification mark is a black concentric circle pattern sprayed on the front bracket of the underwater camera. The spacing between the rings of the identification mark captured in the image is 1mm, 2mm and 3mm respectively. As the sediment concentration increases, the gap between the 1mm rings becomes blurred first, then the gap between the 2mm rings becomes blurred, and finally the gap between the 3mm rings becomes blurred, which enhances the accuracy of image classification. The accuracy is improved by more than 10% compared with the image classification model without identification marks.

[0047] Feature extraction data acquisition: Collect image data containing identification markers at different depths and locations, including identification marker images with varying degrees of blur.

[0048] To ensure accurate data collection, the identification marker is placed near the focal length of the underwater camera, i.e., 15cm away from the underwater camera, to ensure that the blurriness of the camera is only related to the turbidity of the water and not to the focal length.

[0049] Data annotation: The collected image data is annotated using Lambeling software, and the characteristic parameters of each identification mark are annotated, including equipment depth, sediment concentration, and equipment location.

[0050] Data processing: The processed identification marker images underwent data processing, including data cleaning, deduplication, and format conversion. Data augmentation methods were used to expand the training dataset. The images were rotated by 90°, 180°, and 270°, then horizontally and vertically flipped, and cropped into 244*244 square images with the identification marker centered.

[0051] S2. Based on image data, analyze using an AI image classification model to obtain sediment concentration.

[0052] The above images were analyzed using a deep convolutional neural network to obtain the marked regions. The real-time changes in the blurriness of the marked regions were analyzed using dynamic tracking technology, and an AI image classification model for sediment concentration based on the ResNet series was constructed.

[0053] The ResNet series of sediment concentration AI image classification models are transformed into water body sediment concentration AI models by recognizing markers and associating them with the light transmittance of images captured by underwater cameras.

[0054] The training steps for the sediment concentration AI model are as follows: create 100 folders, each containing a sediment concentration image file named [0.01, 5], with a gradient of 0.05, representing sediment concentrations from 0.01 g / L to 5.00 g / L, while ensuring that the paths correspond to the category labels.

[0055] The training set of the model is augmented with random cropping, flipping, rotation and other enhancement strategies. The validation set and test set are only resized and normalized. The image data is classified into test set, validation set and prediction set in a ratio of 8:1:1 and batch loading is encapsulated using DataLoader.

[0056] The model construction method uses transmittance as the original turbidity T, and the turbidity measured by the turbidity sensor is T0. std After correction by the turbidity sensor, the turbidity T' yields a correction relationship that satisfies the linear correction formula:

[0057]

[0058] Where, k T The correction coefficient for the turbidity sensor is obtained by linear regression fitting of the standard turbidity meter and the original turbidity data of the device.

[0059]

[0060] Among them, T std This indicates the standard turbidity value measured by the turbidity sensor.

[0061] The sediment concentration C is calculated using a weighted superposition model, combining the corrected turbidity T' and the equipment immersion depth H, as shown in the following formula:

[0062]

[0063] Where, ω 1、 ω2 represents the weighting coefficient, which is calculated using the entropy weighting method.

[0064] The model is then used to validate the validation set data. The above steps are repeated until the relative error is ≤5%. The data is then used to make predictions on the prediction set to achieve the prediction of sediment concentration.

[0065] S3. Based on the sediment concentration and the light-sensing data from the light sensor, the sand particle size distribution model is used for analysis to obtain sand particle size distribution data.

[0066] Based on the predicted sediment concentration information and combined with the light sensing data from the light sensor, the maximum, median and minimum particle sizes of the sand are calculated using the sand particle size distribution model, thus obtaining the sand particle size distribution data.

[0067] The sand grain gradation model is based on switching the underwater camera to infrared mode, assuming the range of light intensity variation under clear water conditions is as follows: Concentration density coefficient γ, with This serves as an auxiliary reference for estimating the range of changes in light intensity.

[0068]

[0069] Where e represents the Euler number, and the relationship between light intensity and concentration follows an exponential function.

[0070] The light sensor continuously measures the light intensity range for 8 seconds and obtains the minimum, maximum, and median light intensity I. min I max I med The maximum, minimum, and median grain sizes of sand are respectively d max d min d med After correction by a laser sand particle size distribution analyzer, the correction coefficients k1, k2, and k3 are as follows:

[0071]

[0072] Where b1, b2, and b3 represent the bias vectors of the relationship between light intensity and particle size, used to eliminate zero-point error, and their values ​​are the intercepts of the fitting function when the light intensity is 0.

[0073] The sand grain size distribution model needs to be corrected after the sediment concentration is determined to avoid overfitting. The light intensity data for the sand grain size distribution model needs to be continuously updated, with the measured light intensity data updated every 8 seconds and measured continuously for 5 minutes. The average light intensity after the nth update is then calculated. The impact coefficient of the nth update is The sediment concentration C exhibits a negative exponential decay relationship, where M represents the number of samples collected within 8 seconds.

[0074]

[0075]

[0076] Where M represents the number of data collections within 8 seconds; i represents the i-th data collection; x represents the relationship between light intensity and particle size distribution; α x The concentration attenuation coefficient, representing the particle size distribution, is obtained by fitting measured light intensity particle size data under different sediment concentrations.

[0077] The concentration attenuation coefficient for different particle sizes is obtained by fitting measured light intensity particle size data under different sediment concentrations.

[0078] After iterating the sand grain gradation model for 5 minutes, the average of 37 calculated values ​​was taken to obtain the final sand grain gradation data. :

[0079]

[0080] Where n represents the number of model iterations.

[0081] S4. Store the sediment concentration and sand particle size distribution data and upload them to the cloud.

[0082] Using a MySQL database as a data buffer, the stored data is periodically uploaded to the cloud (in this example, it is a cloud-based platform) to monitor the sediment concentration and grain size distribution data in irrigation water sources and irrigation networks in real time.

[0083] MySQL database: As a data storage module, it is deployed on the aforementioned image classification sediment concentration AI model and sand grain gradation model, and uploads the data obtained from the model to the Youren Cloud platform.

[0084] Some cloud computing platforms, acting as network storage and transmission modules, can share data on the network, providing data support for control systems.

[0085] Example 2

[0086] Reference Figure 2-5 This embodiment also provides a device for implementing the method of measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology as described in Embodiment 1. The device includes: an underwater camera 1, a fixed bracket 2 welded to the underwater camera 1, a light sensor 3 welded to the fixed bracket 2, and an identification mark 4 drawn on the fixed bracket 2. Its working principle is as follows:

[0087] (1) Submerge the entire device in the irrigation water to be measured, turn on the white light mode of the underwater camera 1, and take pictures of the identification mark 4 through the camera.

[0088] (2) The distances between the captured identification rings are 1mm, 2mm and 3mm respectively. As the sediment concentration increases, the gap between the 1mm rings becomes blurred first, then the gap between the 2mm rings becomes blurred, and finally the gap between the 3mm rings becomes blurred. This allows for a more accurate image of the concentration change, and the sediment concentration data can be determined by the sediment concentration AI model.

[0089] (3) Switch the underwater camera 1 to infrared light mode and read the light sensing data of the light sensor 3.

[0090] (4) By adjusting the infrared light intensity of the underwater camera 1 through the light sensor 3’s light sensor data change trend, a wider range of light sensor data with more obvious change trend is obtained. Then, through the sand grain distribution model, more accurate data on the maximum, minimum and median particle size of sand are obtained, and then the sand grain distribution of irrigation water is obtained.

[0091] (5) Change the depth and position of the device of the present invention, and repeat the process (1)-(4) until the overall sediment concentration and sand particle size distribution of the irrigation water are obtained.

[0092] Example of sand grain gradation model calculation: Assuming calibration parameters k=[2.5,3.2,2.8], α=[0.12,0.09,0.1], b=[1.2,1.5,1.3], clear water light intensity ΔI0=200 lux, sediment concentration measured by AI model at a certain moment C=2.0g / L, and measured light intensity I=[85,123,104] lux, then:

[0093] ΔI(C) = 37 lux

[0094] ω = 1 + 3.8 / 15 ≈ 1.253

[0095] k x =(e -0.12×2 e -0.09×2 e -0.10×2 ≈ (0.787, 0.835, 0.819)

[0096]

[0097] in, This represents the sand grain gradation data, corresponding to the maximum, minimum, and median grain sizes, respectively, d. max d min d med .

[0098] Based on the example above, the sand particle size distribution at that moment is estimated to be approximately 183.09 μm for the smallest particle size, approximately 243.75 μm for the largest particle size, and approximately 221.86 μm for the median particle size.

[0099] Finally, it should be noted that, in the description of this application, unless otherwise specified and limited, the terms "installation", "connection" and "linkage" should be interpreted broadly, and can refer to mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0100] Secondly, the accompanying drawings of the embodiments disclosed in this invention only involve structures related to the embodiments disclosed in this invention. Other structures can refer to general designs. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0101] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology, characterized in that, include: S1. Capture image data containing identification markers using an underwater camera; S2. Based on the image data, analyze it using an AI image classification model to obtain the sediment concentration; S3. Based on the sediment concentration and the light-sensing data from the light sensor, the sand particle size distribution model is used for analysis to obtain sand particle size distribution data. S4. Store the sediment concentration and sand particle size distribution data and upload them to the cloud.

2. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 1, characterized in that, The identification mark is a preset pattern located around the focal length of the underwater camera, and the blurriness of the image of the identification mark reflects the turbidity of the water.

3. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 1, characterized in that, S2 includes: A deep convolutional neural network is used to analyze the preprocessed image data to obtain the changing trend of ambiguity in the recognition marker region; Based on the correlation between the changing trend of ambiguity and the light transmittance of water bodies, an AI image classification model for sediment concentration was constructed and trained. The trained AI image classification model for sediment concentration was then used to predict image data to obtain sediment concentration.

4. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 3, characterized in that, The output of the AI ​​image classification model for sediment concentration is the corrected turbidity T′. The sediment concentration C is calculated using a weighted superposition model, which combines the corrected turbidity T′ and the equipment immersion depth H. The calculation formula is as follows: Where, k T ω1 and ω2 represent the turbidity sensor correction coefficients; ω1 and ω2 represent the weighting coefficients, calculated using the entropy weighting method.

5. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 1, characterized in that, S3 includes: Based on the sediment concentration and light intensity data obtained by the light sensor, the initial particle size data of the sand particles are calculated through the sand particle size distribution model. The final sand grain size distribution data is obtained by iterative calculation based on the initial particle size data and continuously updated light intensity data.

6. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 5, characterized in that, The iterative calculation process of the sand grain gradation model includes: Light intensity data is continuously acquired and updated at preset time intervals; Based on the current sediment concentration and the updated light intensity data, the sand particle size distribution parameters are iteratively calculated using a formula that includes attenuation relationships. After completing a predetermined number of iterations, the average of all calculated results is taken to obtain the final sand grain size distribution data.

7. The method for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology according to claim 1, characterized in that, S4 includes: Store sediment concentration and sand particle size distribution data in a local database; Data from the local database is regularly uploaded to the cloud platform to enable remote storage, sharing, and monitoring of the data.

8. A device for measuring sediment concentration and grain size distribution in irrigation water using AI image classification technology, the device being used to implement the method described in any one of claims 1-7, characterized in that... include: Camera, mounting bracket, light sensor, and identification sign.

Images