A Deep Learning-Based Method and System for Measuring Median Particle Diameter of Suspended Solids in Reservoir Water Injection

Abstract

This invention discloses a method and system for measuring the median diameter of suspended particles in reservoir water injection based on deep learning. The two-chamber online monitoring device utilizes a window module to provide a good field of view for the camera module, enabling real-time optical imaging and acquisition at the micron level. Combined with an improved deep learning algorithm, the accuracy of classification and regression is enhanced, resulting in a better detection model and thus faster and more accurate calculation of the median particle size of suspended particles. By stratifying the liquid surface thickness and continuously sampling different layers at different flow velocities, and using an energy gradient function to filter the sampling results, the influence of interfering substances such as bubbles and oil droplets can be eliminated, making the collected target suspended particle samples more representative. The design of the turbulence module allows for flexible adjustment of the water injection flow pattern, creating multiple targeted comparative reference examples, which helps to further eliminate the influence of interfering substances such as bubbles and oil droplets, improving the accuracy of the results.

Description

Technical Field

[0001] This invention belongs to the technical field of effective monitoring of unconventional pollutants in oil and gas extraction, and particularly relates to a method and system for measuring the median diameter of suspended particles in reservoir water injection based on deep learning. Background Technology

[0002] With the advancement of science and technology, water injection has become a major technical means to maintain the efficient development of oilfields and an important method to improve the ultimate recovery rate of oilfields. The water quality indicators of the injected water are one of the important factors affecting the effectiveness of water injection development.

[0003] Generally, the indicators for water injection in oilfields are determined by reservoir permeability. When suspended particles are injected into the water, they are carried into the porous medium of the formation. They then move along with the injected water flow within the pore bends of the porous medium. Subsequently, these suspended particles are retained and precipitated in the pore skeleton, eventually being captured. This causes changes in the pore skeleton structure, reducing the porosity and permeability of the formation rocks. This phenomenon is usually called formation damage, and the degree of damage is measured by the permeability damage rate.

[0004] Currently, research on online monitoring of the median diameter of suspended particles is still in its early stages. The main method for detecting the median diameter of suspended particles remains manual sampling and laboratory testing. This involves placing manually obtained samples in a laboratory testing instrument and measuring the intensity of the scattered light through a laser diffraction beam. Larger particles scatter light at smaller angles, while smaller particles scatter light at larger angles. The scattered light intensity data is analyzed, and the particle size of the particles generating the scattered light is calculated using Mie scattering theory. Particle size is expressed as the diameter of a sphere of equal volume. Currently, the MASTERSIZER 3000 intelligent particle size analyzer is commonly used, capable of measuring particles ranging from 0.01 μm to 3.5 mm.

[0005] With the development of computer image recognition technology and related hardware, machine vision recognition systems have been applied in various fields. This system uses a camera to capture images of the target site, and a central processing unit processes and classifies the captured images to determine the target object. This technology is highly automated and flexible, possessing stronger judgment capabilities compared to other types of recognition technologies. Currently, machine vision recognition systems are widely used in industrial plants and for pollutant identification and monitoring. Invention patent CN112052634A discloses an analytical method and apparatus for the influence of particle migration during water injection in low-permeability oil reservoirs. This method can understand the specific processes of particle movement, deposition, and blockage with fluid, and their impact on the displacement process during water injection and extraction in oil reservoirs. However, this method lacks classification, effective collection, and screening of training samples, which affects the accuracy of the prediction results. Summary of the Invention

[0006] This invention addresses the problems existing in the prior art by providing a method and system for measuring the median diameter of suspended particles in reservoir water injection based on deep learning.

[0007] In a first aspect, the present invention provides a method for measuring the median diameter of suspended solids in reservoir water injection based on deep learning, comprising the following steps:

[0008] Step S1: Power on the measuring device system and connect the inlet and outlet pipes. Introduce the water flow into the measuring device through the conduit and adjust the water flow rate to the first speed.

[0009] Step S2: The water flow through the inner window box of the measuring device is stratified, and reservoir water injection images corresponding to different liquid level heights are collected;

[0010] Step S3: Determine the sharpness of each layer of images acquired layer by layer, discard images that do not meet the sharpness standard, and save the clear images that meet the set conditions in the AI ​​edge computing module of the measuring device;

[0011] Step S4: Analyze the images stored in the AI ​​edge computing module using an improved deep learning-based detection algorithm to identify target suspended particles;

[0012] Step S5: Use the deep learning-based HED algorithm to perform edge detection on the detected target suspended particles, fit the edges to a circle, and then calculate the diameter;

[0013] Step S6: Statistically analyze the distribution of target suspended particles within different diameter ranges;

[0014] Step S7: Adjust the water injection flow rate to the second speed, and repeat steps S2-S6;

[0015] Step S8: Transmit the statistical results of the target suspended particle diameter distribution to the shore via the network.

[0016] Preferably, step S2 specifically includes:

[0017] Step S21: Divide the liquid surface into n layers according to the thickness of the liquid surface in the viewing box, where n≥3;

[0018] Step S22: Move the high-definition industrial camera to the position corresponding to the i-th liquid layer using the focusing module, where 0 < i ≤ n;

[0019] Step S23: Adjust the parameters of the high-definition industrial camera and the high-brightness light source to meet the imaging requirements;

[0020] Step S24: Collect m images continuously, where m ≥ 1000.

[0021] Preferably, step S3 specifically includes:

[0022] Step S31: Traverse the n layers (n*m images) in steps S21-S24. For the j-th image P(i,j) in the i-th layer, calculate its sharpness F value using the energy gradient function:

[0023]

[0024] Where x and y represent the x-th row and y-th column of image P(i,j), f(x,y) represents the gray value of the x-th row and y-th column of image P(i,j), and 0 < j ≤ m;

[0025] Step S32: Discard images with a sharpness F value less than the set threshold, and treat the remaining images as sharp images that meet the set conditions, and save them in the AI ​​edge computing module of the measuring device.

[0026] Preferably, step S4 specifically includes:

[0027] Step S41: Crop the clear images that meet the set conditions, and crop each image into 25 sub-images of 5×5 pixels;

[0028] Step S42: Label the cropped sub-images to identify the target suspended particles, and use the labeled sub-images as the target detection dataset;

[0029] Step S43: Train the object detection dataset using the improved YOLOv8 network to obtain the detection model;

[0030] Step S44: Deploy the trained detection model on the server and use TensorRT to accelerate inference and achieve real-time inference.

[0031] Preferably, step S6 specifically includes:

[0032] Step S61: Count the number of target suspended particles whose diameter values ​​fall within the intervals [0,0.5], [0.5-5], [5-10], [10-15], [15-25], [25-50], [50-100], [100-200], [200-300], [300-400] respectively;

[0033] Step S62: Calculate the D50 value of the diameter of all target suspended particles.

[0034] Preferably, step S7 specifically includes:

[0035] Step S71: Before entering the viewing box, the water flow is disturbed to change its flow rate and the distribution pattern of impurities;

[0036] Step S72: Repeat steps S2-S6, and then weight the D50 values ​​of all statistically obtained target suspended particle diameters.

[0037] Preferably, step S8 specifically includes:

[0038] Step S81: Transmit the detection results, such as the D50 value of the target suspended particles diameter, to the offshore platform network control center via serial port and / or wired / wireless LAN;

[0039] Step S82: Transmit the detection results, such as the D50 value of the target suspended particles diameter, from the offshore platform network control center to the shore via satellite communication methods such as Haiweitong.

[0040] In a second aspect, the present invention provides a deep learning-based system for measuring the median diameter of suspended solids in reservoir water injection. The system includes a measuring device and a network control center for an offshore platform. The measuring device specifically includes: a shell, a flow rate adjustment module, a viewing window module, an image acquisition module, a focusing module, an algorithm processing module, a communication module, and a power supply module.

[0041] The flow rate regulating device includes a magnetic pump and a driver, the driver being used to provide driving force to the magnetic pump; the magnetic pump being used to regulate the water injection flow rate;

[0042] The viewing window module is composed of two pieces of transparent glass stacked opposite each other in a vertical direction, used to introduce water flow from the magnetic pump;

[0043] The image acquisition module includes a high-spot light source, a high-magnification microscope lens, and a high-definition industrial camera;

[0044] The focusing module includes a focusing component and a stepper motor;

[0045] The algorithm processing module includes an embedded AI edge computing module, which is used to process the acquired images through visual algorithms;

[0046] The offshore platform network control center is used to receive image processing results from the communication module and transmit the image processing results to the shore via satellite communication.

[0047] Preferably, the direction from the inlet to the outlet of the housing is defined as the first direction, and the direction from the bottom surface to the top surface of the housing is defined as the second direction. A partition assembly parallel to the first plane is provided on the first plane formed by the first direction and the second direction. The partition assembly divides the housing into a first chamber and a second chamber. The flow rate adjustment module, algorithm processing module, communication module, and power supply module are located in the first chamber, and the window module, image acquisition module, and focusing module are located in the second chamber.

[0048] Preferably, the measurement system further includes a turbulence module located between the magnetic pump and the window module, used to turbulently guide the water flow from the magnetic pump into the window module.

[0049] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0050] In a fourth aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of any of the methods described above.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1) The two-chamber online monitoring device of the present invention can provide a good field of view for the camera module using the window module, and realize real-time optical imaging and acquisition of micron-level suspended particles of water injection in oil reservoirs while ensuring the stable and efficient operation of each module. With the improved deep learning algorithm, the accuracy of classification and regression can be improved, a better detection model can be obtained, and the median particle size of suspended particles can be calculated more quickly and accurately.

[0053] 2) By dividing the liquid surface thickness into layers and continuously sampling different layers at different flow rates, and using the energy gradient function to filter the sampling results, the influence of interfering objects such as bubbles and oil droplets can be eliminated, making the collected target suspended matter samples more representative and improving the accuracy of the results.

[0054] 3) The design of the turbulence module can flexibly adjust the water flow pattern and form multiple comparative reference examples in a targeted manner, which helps to further eliminate the influence of interference such as bubbles and oil droplets and improve the reliability of the acquired images. Attached Figure Description

[0055] Figure 1 Flowchart of the deep learning-based method for measuring the median diameter of suspended particles in reservoir water injection;

[0056] Figure 2 First-view side view of the reservoir water injection suspended particle diameter median measurement device of the present invention;

[0057] Figure 3 A second-view side view of the reservoir water injection suspended particle diameter median measurement device of the present invention;

[0058] Figure 4 The structural diagram of the turbulence module of the present invention.

[0059] Explanation of reference numerals in the attached drawings: 1. Housing; 2. Magnetic pump; 3. Driver; 4. Algorithm processing module; 5. Communication module; 6. Power supply module; 7. Fluctuation module; 8. Fluctuation plate; 9. Viewing window module; 10. High-brightness light source; 11. High-definition industrial camera; 12. Focusing module; 13. Partition assembly. Detailed Implementation

[0060] The techniques described below can be modified in various ways and have multiple embodiments, which are described in detail below with reference to the accompanying drawings. However, this does not mean that the techniques described below are limited to the specific embodiments. It should be understood that the present invention includes all similar modifications, equivalents, and substitutions without departing from the spirit and scope of the techniques described below.

[0061] Example 1

[0062] like Figure 1 As shown, this invention provides a method for measuring the median diameter of suspended particles in reservoir water injection based on deep learning, comprising the following steps:

[0063] Step S1: Power on the measuring device system and connect the inlet and outlet pipes. Introduce the water flow into the measuring device through the conduit and adjust the water flow rate to the first speed.

[0064] Step S2: The water flow through the inner window box of the measuring device is stratified, and reservoir water injection images corresponding to different liquid level heights are collected;

[0065] Step S3: Determine the sharpness of each layer of images acquired layer by layer, discard images that do not meet the sharpness standard, and save the clear images that meet the set conditions in the AI ​​edge computing module of the measuring device;

[0066] Step S4: Analyze the images stored in the AI ​​edge computing module using an improved deep learning-based detection algorithm to identify target suspended particles;

[0067] Step S5: Use the deep learning-based HED algorithm to perform edge detection on the detected target suspended particles, fit the edges to a circle, and then calculate the diameter;

[0068] Step S6: Statistically analyze the distribution of target suspended particles within different diameter ranges;

[0069] Step S7: Adjust the water injection flow rate to the second speed, and repeat steps S2-S6;

[0070] Step S8: Transmit the statistical results of the target suspended particle diameter distribution to the shore via the network.

[0071] Specifically, step S2 includes:

[0072] Step S21: Divide the liquid surface into n layers according to the thickness of the liquid surface in the viewing box, where n≥3;

[0073] Step S22: Move the high-definition industrial camera to the position corresponding to the i-th liquid layer using the focusing module, where 0 < i ≤ n;

[0074] Step S23: Adjust the parameters of the high-definition industrial camera and the high-brightness light source to meet the imaging requirements;

[0075] Step S24: Collect m images continuously, where m ≥ 1000.

[0076] Because the viewing box has a certain thickness (about 1 mm), and the depth of field of a high-powered microscope lens is generally in the tens of micrometers range, the position of the suspended object in the viewing box is random. If stratified sampling is not performed, the collected suspended object sample will lack representativeness.

[0077] Specifically, step S3 includes:

[0078] Step S31: Traverse the n layers (n*m images) in steps S21-S24. For the j-th image P(i,j) in the i-th layer, calculate its sharpness F value using the energy gradient function:

[0079]

[0080] Where x and y represent the x-th row and y-th column of image P(i,j), f(x,y) represents the gray value of the x-th row and y-th column of image P(i,j), and 0 < j ≤ m;

[0081] Step S32: Discard images with a sharpness F value less than the set threshold, and treat the remaining images as sharp images that meet the set conditions, and save them in the AI ​​edge computing module of the measuring device.

[0082] The threshold is generally set to 5000 based on experience, but it can be adjusted flexibly according to the actual situation.

[0083] Specifically, step S4 includes:

[0084] Step S41: Crop the clear images that meet the set conditions, and crop each image into 25 sub-images of 5×5 pixels;

[0085] Step S42: Label the cropped sub-images to identify the target suspended particles, and use the labeled sub-images as the target detection dataset;

[0086] Step S43: Train the object detection dataset using the improved YOLOv8 network to obtain the detection model;

[0087] Step S44: Deploy the trained detection model on the server and use TensorRT to accelerate inference and achieve real-time inference.

[0088] The YOLOv8 network architecture mainly consists of three parts: Backbone, Neck, and Detect Head. The Backbone is largely based on the CSPDarkNet structure, but uses the C2f (CSPLayer_2Conv) module instead of the C3 module. The C2f module optimizes CSPDarkNet, reducing computational cost while improving convergence speed and performance. The input image first undergoes a series of convolutional and pooling layers for feature extraction, forming feature maps of different scales. The Neck part adopts a two-stream FPN (Feature Pyramid Network) structure similar to PANet (PathAggregation Network), which integrates feature information at different scales, improving the model's ability to detect objects at different scales. The Detect Head uses a decoupled head design similar to YOLOX, decoupling classification and regression tasks, which are handled by separate branches.

[0089] To improve the network's detection performance, this invention inserts a CoT module into the Neck region. The CoT module first performs context encoding on the input features using 3×3 convolutions to obtain a static context representation. Then, the encoded features are concatenated with the query and fed into the Transformer's self-attention mechanism. Through this mechanism, the model learns the dynamic dependencies between features. Finally, the features output by the self-attention mechanism are fused with the original features to obtain an enhanced feature representation. The detection head uses these enhanced features from the Transformer attention mechanism as input, improving the accuracy of classification and regression.

[0090] For step S5, HED is a classic deep learning-based edge detection algorithm. It uses a fully convolutional network to extract features at different levels and fuses features of different scales through skip connections to obtain more accurate edge detection results. This multi-scale feature fusion method enables HED to capture edge information at different scales in the image. Simultaneously, it can be trained end-to-end, learning edge features directly from the image without the need for manually designed feature extractors. This allows HED to automatically learn edge features suitable for specific tasks, improving detection accuracy and robustness. Then, the ODEC method (overlapping 2D circles) is used to fit the circle. The core idea of ​​this method is to overlap a series of 2D circles inside the particle shape contour. Starting from a node on the contour, a circle is drawn along the inner normal direction of that node, continuously expanding the circle until its contour touches the particle contour. This process is repeated for other nodes, ensuring that all circles overlap inside the particle and cover the entire particle shape as much as possible. Finally, based on the outer contour of the overlapping 2D circles, an optimal circle is fitted using the algorithm.

[0091] Specifically, step S6 includes:

[0092] Step S61: Count the number of target suspended particles whose diameter values ​​fall within the intervals [0,0.5], [0.5-5], [5-10], [10-15], [15-25], [25-50], [50-100], [100-200], [200-300], [300-400] respectively;

[0093] Step S62: Calculate the D50 value of the diameter of all target suspended particles.

[0094] D50 refers to the particle size at which the cumulative distribution percentage reaches 50%. D50 is also known as median diameter or median particle size. Assuming a sample has a D50 of 5 μm, it means that in the particles composing this sample, particles larger than 5 μm account for 50%, and particles smaller than 5 μm also account for 50%. D50 is a typical indicator for evaluating suspended particle size and is often used to represent the average particle size of suspended matter.

[0095] Specifically, step S7 includes:

[0096] Step S71: Before entering the viewing box, the water flow is disturbed to change its flow rate and the distribution pattern of impurities;

[0097] Step S72: Repeat steps S2-S6, and then weight the D50 values ​​of all statistically obtained target suspended particle diameters.

[0098] The morphology and distribution of impurities in the injected water flow will vary under different flow velocities and conditions. Furthermore, the camera's image acquisition location will also affect the imaging. Therefore, by employing a layered approach—acquiring multiple images at different locations and flow velocities, processing and analyzing them separately, and then fusing and comparing them—the accuracy of identifying suspended target matter can be maximized.

[0099] When repeating step S32 for the second time, the threshold used for comparison with the sharpness F value can be different from the threshold used when performing step S32 for the first time.

[0100] When repeating step S41 for the second time, each image can be cropped into 36 sub-images of 6×6, which is different from the cropping size of the image when performing step S41 for the first time.

[0101] When weighting the D50 values ​​obtained in the first and second statistical analyses, their respective weights can be different.

[0102] Specifically, step S8 includes:

[0103] Step S81: Transmit the detection results, such as the D50 value of the target suspended particles diameter, to the offshore platform network control center via serial port and / or wired / wireless LAN;

[0104] Step S82: Transmit the detection results, such as the D50 value of the target suspended particles diameter, from the offshore platform network control center to the shore via satellite communication methods such as Haiweitong.

[0105] Example 2

[0106] like Figure 2-3 As shown, the present invention also provides a deep learning-based system for measuring the median diameter of suspended solids in reservoir water injection. The system includes a measuring device and an offshore platform network control center. The measuring device specifically includes: a housing 1, a flow rate adjustment module, a viewing window module 8, an image acquisition module, a focusing module 11, an algorithm processing module 4, a communication module 5, and a power supply module 6.

[0107] A first direction is defined as the direction from the inlet to the outlet of the housing 1, and a second direction is defined as the direction from the bottom surface to the top surface of the housing 1. A partition assembly 12 parallel to the first plane is provided on a first plane formed by the first direction and the second direction. The partition assembly 12 divides the housing 1 into a first chamber and a second chamber. The flow rate adjustment module, the algorithm processing module 4, the communication module 5, and the power supply module 6 are located in the first chamber, and the window module 8, the image acquisition module, and the focusing module 11 are located in the second chamber.

[0108] The inlet of the housing 1 is connected to a water inlet pipe, and the outlet is connected to a water outlet pipe.

[0109] The flow rate regulating device includes a magnetic pump 2 and a driver 3. The driver 3 is used to provide driving force to the magnetic pump 2. The magnetic pump 2 is used to regulate the water injection flow rate in response to user input, or to regulate the water injection flow rate based on a pre-set program, or to automatically regulate the water injection flow rate according to parameters such as the current water injection flow rate, water pressure, and water quality.

[0110] The viewing window module 8 is a viewing box formed by stacking two pieces of transparent glass opposite each other in an orientation perpendicular to the first plane and perpendicular to the first direction, used to introduce water flow from the magnetic pump; the interlayer spacing between the two pieces of transparent glass in the viewing window module 8 is h, and h < 1 mm.

[0111] The image acquisition module includes a high-spot light source 9, a high-magnification microscope lens, and a high-definition industrial camera 10; the high-spot light source 9 is located on the side of the viewing window module 8 facing the exit, and the high-magnification microscope lens and the high-definition industrial camera 10 are arranged sequentially on the side of the viewing window module 8 facing the inlet.

[0112] The focusing module 11 includes a focusing component and a stepper motor. The focusing component includes a slider and a slide rail. The slider is provided with a connector for fixing the high-definition industrial camera. The stepper motor is used to control the movement of the slider.

[0113] The algorithm processing module 4 includes an embedded AI edge computing module, which is used to process the acquired images through visual algorithms; the processing of the acquired images through visual algorithms includes image cropping, target recognition, edge detection, shape fitting, and image fusion.

[0114] The offshore platform network control center is used to receive image processing results from the communication module and transmit the image processing results to the shore via satellite communication.

[0115] It also includes a turbulence module 7, located between the magnetic pump 2 and the window module 8, which is used to turbulent the water flow from the magnetic pump 2 and guide it into the window module 8.

[0116] like Figure 4As shown, the flow-disrupting module 7 is connected to the outlet of the magnetic pump 2 at one end and to the inlet of the viewing box 8 at the other end, forming a nearly rectangular closed fluid passage. On the upper and lower sides or left and right sides of the passage, flow-disrupting plates 71, one end fixed and the other rotatable, are staggered. The rotatable end of the flow-disrupting plate 71 faces the side of the flow-disrupting module 7 connected to the outlet of the magnetic pump 2, and the rotation angle α is adjustable, specifically within the range of 0°≤α≤45°. By adjusting the different rotation angles of the different flow-disrupting plates 71, the water flow can be effectively disturbed, changing the state of bubbles, oil droplets, and suspended matter in the water flow, thus facilitating the effective identification and extraction of target suspended matter through subsequent image acquisition and analysis. When the flow-disrupting module 7 is not in use, the flow-disrupting plate 71 is in its initial position, α=0°.

[0117] During step S71, the water flow is disturbed by rotating a portion of the baffle plate 71 to a certain angle. The specific baffle plate 71 to be rotated and the rotation angle value can be flexibly adjusted according to actual needs.

[0118] The positions of the spoiler 71, the high-definition industrial camera 10, and the flow rate of the water injection can be combined and varied in various ways according to actual needs, and are not limited to the above situations.

[0119] Example 3

[0120] The present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0121] Example 4

[0122] The present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of any of the methods described above.

[0123] Although the present invention has been described in detail above with general descriptions and specific embodiments, some modifications or improvements can be made to it. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Other changes and modifications made by those skilled in the art without departing from the spirit and scope of the present invention are still included within the scope of protection of the present invention.

Claims

1. A method for measuring the median diameter of suspended solids in reservoir water injection based on deep learning, characterized in that, Includes the following steps: Step S1: Power on the measuring device system and connect the inlet and outlet pipes. Introduce the water flow into the measuring device through the conduit and adjust the water flow rate to the first speed. Step S2: The water flow through the inner window box of the measuring device is stratified, and reservoir water injection images corresponding to different liquid level heights are collected; Step S3: Determine the sharpness of each layer of images acquired layer by layer, discard images that do not meet the sharpness standard, and save the clear images that meet the set conditions in the AI ​​edge computing module of the measuring device; Step S4: Analyze the images stored in the AI ​​edge computing module using an improved deep learning-based detection algorithm to identify target suspended particles; Step S4 specifically includes: Step S41: Crop the clear images that meet the set conditions, and crop each image into 25 sub-images of 5×5 pixels; Step S42: Label the cropped sub-images to identify the target suspended particles, and use the labeled sub-images as the target detection dataset; Step S43: Train the object detection dataset using the improved YOLOv8 network to obtain the detection model, wherein a CoT module is inserted in the Neck part; Step S44: Deploy the trained detection model on the server and use TensorRT to accelerate inference and achieve real-time inference; Step S5: Use the deep learning-based HED algorithm to perform edge detection on the detected target suspended particles, fit the edges to a circle, and then calculate the diameter; Step S6: Statistically analyze the distribution of target suspended particles within different diameter ranges; Step S7: Adjust the water injection flow rate to the second speed, and repeat steps S2-S6; Step S7 specifically includes: Step S71: Before entering the viewing box, the water flow is disturbed by a baffle to change its flow rate and the distribution pattern of impurities. Step S72: Repeat steps S2-S6, and then weight the D50 values ​​of all statistically obtained target suspended particle diameters. Step S8: Transmit the statistical results of the target suspended particle diameter distribution to the shore via the network.

2. The measurement method according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Divide the liquid surface into n layers according to the thickness of the liquid surface in the viewing box, where n≥3; Step S22: Move the high-definition industrial camera to the position corresponding to the i-th liquid layer using the focusing module, where 0 < i ≤ n; Step S23: Adjust the parameters of the high-definition industrial camera and the high-brightness light source to meet the imaging requirements; Step S24: Collect m images continuously, where m ≥ 1000.

3. The measurement method according to claim 2, characterized in that, Step S3 specifically includes: Step S31: Traverse the n layers (n*m images) in steps S21-S24. For the j-th image P(i,j) in the i-th layer, calculate its sharpness F value using the energy gradient function: Where x and y represent the x-th row and y-th column of image P(i,j), f(x,y) represents the gray value of the x-th row and y-th column of image P(i,j), and 0 < j ≤ m; Step S32: Discard images with a sharpness F value less than the set threshold, and treat the remaining images as sharp images that meet the set conditions, and save them in the AI ​​edge computing module of the measuring device.

4. The measurement method according to claim 3, characterized in that, Step S6 specifically includes: Step S61: Calculate the range for each particle diameter value. The number of target suspended particles within [0,0.5],[0.5-5],[5-10],[10-15],[15-25],[25-50],[50-100],[100-200],[200-300],[300-400]; Step S62: Calculate the D50 value of the diameter of all target suspended particles.

5. The measurement method according to claim 4, characterized in that, Step S8 specifically includes: Step S81: Transmit the detection results, such as the D50 value of the target suspended particles diameter, to the offshore platform network control center via serial port and / or wired / wireless LAN. Step S82: Transmit the detection results, such as the D50 value of the target suspended particles diameter, from the offshore platform network control center to the shore via satellite communication methods such as Haiweitong.

6. A measurement system for implementing the deep learning-based method for measuring the median diameter of suspended solids in reservoir water injection as described in any one of claims 1-5, characterized in that, The system includes a measurement device and an offshore platform network control center. The measurement device specifically includes: a shell, a flow rate adjustment module, a viewing window module, an image acquisition module, a focusing module, an algorithm processing module, a communication module, and a power supply module. The flow rate regulating device includes a magnetic pump and a driver, the driver being used to provide driving force to the magnetic pump, and the magnetic pump being used to regulate the water injection flow rate; The viewing window module is composed of two pieces of transparent glass stacked opposite each other in a vertical direction, used to introduce water flow from the magnetic pump; The image acquisition module includes a high-spot light source, a high-magnification microscope lens, and a high-definition industrial camera; The focusing module includes a focusing component and a stepper motor; The algorithm processing module includes an embedded AI edge computing module, which is used to process the acquired images through visual algorithms; The offshore platform network control center is used to receive image processing results from the communication module and transmit the image processing results to the shore via satellite communication.

7. The measurement system according to claim 6, characterized in that, A first direction is defined as the direction from the inlet to the outlet of the housing, and a second direction is defined as the direction from the bottom surface to the top surface of the housing. A partition assembly parallel to the first plane is provided on a first plane formed by the first direction and the second direction. The partition assembly divides the housing into a first chamber and a second chamber. The flow rate adjustment module, algorithm processing module, communication module, and power supply module are located in the first chamber, and the window module, image acquisition module, and focusing module are located in the second chamber.

8. The measurement system according to claim 7, characterized in that, It also includes a turbulence module, located between the magnetic pump and the window module, which is used to turbulent the water flow from the magnetic pump and guide it into the window module.

Images