Equipment and method for rapidly detecting concentration of suspended solid substance

Abstract

The invention relates to equipment and a method for rapidly detecting the concentration of suspended solid substances, and the equipment comprises a closed shooting box body, the inner wall of which is black; the light-emitting bottom plate is mounted at the bottom of the box body and is used for providing an upward uniform light source; the sample fixing module is a black entity device with a hollow middle part, and the hollow part of the sample fixing module is a beaker placing rack; the camera module is fixedly mounted at the top of the box body; the magnetic stirring module is arranged below the luminous bottom plate; the image analysis module is used for carrying out preprocessing and concentration classification analysis on the image acquired by the camera module; the image analysis module is arranged in the external computing device, and the external computing device is directly connected with the camera module and used for generating a concentration detection report according to the real-time classification result. On one hand, a traditional tedious laboratory analysis method can be replaced, and rapid and non-contact concentration detection is achieved; and on the other hand, the robustness and generalization ability of the model are improved by using the Inception-ResNet coupling neural network model, and the detection accuracy is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical fields of environmental monitoring, artificial intelligence, and deep learning applications, and particularly relates to a device and method for rapidly detecting the concentration of suspended solid substances. Background Art

[0002] Currently, the detection methods for the concentration of suspended solid substances (SS) usually adopt physicochemical analysis methods, such as the filter paper method or the optical turbidimeter method. Among them, the filter paper method requires filtering the water sample and then calculating the mass of the solid substance by drying and weighing. Although this method is directly simple to operate, the experimental steps are numerous, the efficiency is low, it is easy to introduce human errors, and real-time monitoring cannot be carried out; while the optical turbidimeter method indirectly measures the turbidity of the water sample based on the scattering or transmission characteristics of light. This method is relatively efficient but has limited accuracy and is extremely vulnerable to environmental interference.

[0003] In recent years, the analysis and detection of suspension images by machine learning models have achieved preliminary research in the field of environmental monitoring. For example, the concentration of the suspension is estimated by analyzing the color, brightness, or particle distribution of the sample image. However, there are still some difficult problems in using machine learning models for image analysis:

[0004] (1) Poor model robustness and insufficient generalization ability: The traditional machine learning model has a simple structure and a small amount of data for model training. Therefore, it is difficult to handle complex solution or suspension samples, and the practical performance of the model is poor;

[0005] (2) Interference from the background environment: The training process of the model is a self-learning black box process. Therefore, the non-standard shooting environment, external light, the material and shape of the sample container, and the background color will all interfere with the model's analysis of the image, resulting in a decrease in the accuracy of concentration detection.

[0006] To solve the above two key problems, first, we should construct a standardized background environment, including uniform light intensity, background color, sample container, etc. At the same time, with the booming development of deep learning in various fields such as medical treatment, finance, and intelligent transportation, using a deep learning model with a more complex network structure should be an effective method to solve the problems of poor model robustness and insufficient generalization ability. But it also poses new problems and challenges for us: The deep learning model is essentially a hungry model that needs to extract common features from a large amount of data to train the model. How to provide a large number of high-quality labeled data sets has become a key problem. Summary of the Invention

[0007] The object of the present invention is to provide a device and method for rapidly detecting the concentration of suspended solid substances. On the one hand, it solves the problems of low efficiency, large error, and inability to monitor in real time existing in the traditional methods for measuring the concentration of suspended solid substances (SS). On the other hand, it solves the problem of decreased detection accuracy caused by insufficient equipment standardization and poor model robustness when applying the image analysis model.

[0008] The present invention provides a device for rapidly detecting the concentration of suspended solid substances, comprising:

[0009] A closed shooting box with a black inner wall, used to provide a completely black background environment;

[0010] A light-emitting bottom plate, installed at the bottom of the box, used to provide a uniform upward light source, so that the light passes through the suspension of suspended solid substances from bottom to top, to enhance the contrast of the image;

[0011] A sample fixing module, which is a black solid device with a hollow in the middle. The hollow part is a beaker placement rack, used to stabilize the sample beaker;

[0012] A camera module, fixedly installed on the top of the box, used to vertically downward shoot the image of the suspension of suspended solid substances in the beaker;

[0013] A magnetic stirring module, arranged below the light-emitting bottom plate, used to stir the suspended solid substances during the shooting process to make them evenly distributed;

[0014] An image analysis module, used to preprocess and classify and analyze the images collected by the camera module to obtain the classification result of the concentration of suspended solid substances;

[0015] An external computing device, with the image analysis module built-in and directly connected to the camera module, used to generate a concentration detection report through the real-time classification result, and support the storage and comparative analysis of historical data.

[0016] Further, the light-emitting bottom plate includes an LED light source with adjustable light intensity and color temperature, and the surface is covered with a frosted diffuser plate, used to ensure the uniformity of light and reduce the light reflection at the edge of the beaker.

[0017] Further, the shooting module is a high-resolution industrial camera, equipped with a fixed focal length lens, and its aperture is set to a constant value to avoid the influence of light intensity fluctuation on the shooting quality.

[0018] Further, the top of the closed shooting box is also provided with an anti-reflection coating transparent glass, used to protect the camera module and avoid interference from external stray light.

[0019] Further, the magnetic stirring module has a constant-speed stirring function to ensure that the suspended solid substances remain evenly dispersed during the shooting.

[0020] Further, the shape of the hollowed-out part perfectly fits the outer shape of a 1000 ml beaker.

[0021] Further, the image analysis module includes:

[0022] A preprocessing unit for graying, denoising, contrast enhancement, and edge feature extraction of the acquired images;

[0023] A classification unit for feature extraction of the preprocessed images based on the Inception-ResNet deep learning model and finally outputting the classification result of the concentration of suspended solid substances (SS).

[0024] Further, the preprocessing unit is specifically used for:

[0025] S1, Graying: Convert the original color image into a grayscale image to reduce the computational complexity;

[0026] S2, Denoising: Use the Gaussian filtering algorithm to remove image noise;

[0027] S3, Image contrast enhancement: Use the method based on histogram equalization to enhance the contrast of the acquired images;

[0028] S4, Edge feature extraction: Use the Canny edge detection algorithm to extract the edge features inside the SS suspension.

[0029] Further, the classification unit is specifically used for:

[0030] S1, Feature extraction using the pre-trained model of the Inception-ResNet neural network, where the pre-trained model includes multiple convolutional modules and residual connections, and the input image size is 299×299 pixels;

[0031] S2, The final classification layer is a Softmax fully connected layer, which is divided into 20 categories in total, and outputs the SS suspension concentration level of 50 - 1000 mg SS / L.

[0032] The present invention also provides a method for quickly detecting the concentration of suspended solid substances by applying the device, including the following steps:

[0033] Step 1, Equipment installation and inspection:

[0034] Ensure that the enclosed shooting box is correctly assembled to avoid the influence of external light and internal light reflection on shooting; check whether the LED light source of the light-emitting bottom plate is normal and whether the surface frosted diffuser plate is properly covered to ensure the uniformity of illumination; ensure that the magnetic stirring module operates normally; check whether the camera port of the camera module is contaminated to avoid visual field obstacles and ensure that the shooting visual field completely covers the beaker.

[0035] Step 2, Equipment initialization:

[0036] Turn on the power and start the light-emitting bottom plate; turn on the magnetic stirring module and set the stirring speed to a moderate constant value; start the camera module, test the image acquisition function, and confirm that the image is clear and distortion-free.

[0037] Step 3, Sample treatment and fixation:

[0038] Pour the sample suspension of the suspended solid substance to be measured into the beaker, ensuring the transparency and cleanliness of the beaker to avoid affecting the optical imaging quality; place the beaker containing the sample into the sample fixation module, ensuring that the hollow part of the beaker and the sample fixation module fit completely to prevent the sample from shaking or displacing; place the sample fixation module at the central position of the magnetic stirring module to ensure that the action of the stirrer is evenly distributed in the sample.

[0039] Step 4, Image acquisition:

[0040] Start the light-emitting bottom plate and the magnetic stirring module. After stirring for 10 - 15 seconds, ensure that the suspended particles in the sample are evenly distributed; start the camera module and vertically shoot the sample image. After shooting, save the image to the specified directory of the computing device, ensuring that the file name of each shooting is unique; repeat the acquisition of the sample image at different stirring times for multiple subsequent image analyses.

[0041] Step 5, Image preprocessing:

[0042] Input the acquired image into a pre-set program and perform grayscale conversion, denoising, contrast enhancement, and edge feature extraction operations in sequence.

[0043] Step 6, Image classification model analysis:

[0044] Adjust the preprocessed image to the standard size of 299×299 pixels and input it into the Inception-ResNet model; use the convolutional layer and residual connections of the model to extract the texture features, density distribution, and optical properties of the particles in the image; use the fully connected layer to classify the concentration of the SS suspension sample and output the SS concentration level of the sample. The preset measurement range of the model is 0 - 1000mgSS / L, and a concentration difference of 50mgSS / L is regarded as one level, with a total of 20 SS concentration levels.

[0045] Step 7: Output and store results:

[0046] The test results are displayed in real time through the computing equipment, including sample number, test time, stirring speed, SS concentration level information; the image and analysis results of each test are stored in the local database; the test report is automatically generated, including images, concentration analysis results, and experimental parameters;

[0047] Step 8, Cleaning and Maintenance:

[0048] After the test, remove the beaker and fixed sample module, clean and dry them to avoid sample residue affecting subsequent tests; perform equipment maintenance regularly, including replacing the light-emitting base plate, checking the cleanliness of the camera module lens and the stirring effect of the magnetic stirring module, to ensure stable equipment performance.

[0049] By means of the above scheme, the device and method for quickly detecting the concentration of suspended solid matter have the following technical effects:

[0050] 1) The present invention eliminates the interference of external light, reflection and background debris on image acquisition by designing a completely closed shooting box and a black inner wall of the box.

[0051] 2) The present invention not only optimizes the shooting light environment, but also ensures the consistency of the sample position and avoids the deviation caused by the displacement of the beaker position through the combination of the luminous bottom plate and the black fixed module with a hollow middle portion.

[0052] 3) The present invention uses a built-in magnetic stirring module to ensure that the solid particles in the suspension are always evenly distributed during the shooting process through continuous stirring, avoiding the problem of uneven concentration distribution caused by particle sedimentation or aggregation in static samples, thereby improving the repeatability and stability of the detection.

[0053] 4) The present invention uses a shooting module to shoot the sample vertically downward, avoiding the pollution or sample loss that may be introduced in traditional methods such as the traditional filter paper method. At the same time, it does not require any additional chemical reagents or consumables, is simple to operate, and is green and environmentally friendly.

[0054] 5) The present invention adopts the Inception-ResNet coupled convolutional neural network model, which can automatically extract complex features (such as particle distribution, concentration pattern, etc.) in SS suspension images after training with a large number of annotated data sets, and realize fast and accurate concentration classification. At the same time, the model preserves the underlying features in the image through the direct connection edges of the residual network, making the model have excellent robustness and generalization ability.

[0055] 6) The present invention integrates a light-emitting bottom plate, a sample fixing module, a stirring module, a camera module, and an intelligent image analysis system into one, forming a set of efficient, portable, and intelligent detection equipment, which can be widely applied to water quality detection, sewage treatment plants, environmental protection departments, and laboratory scientific research activities, and is of great significance for the rapid detection and dynamic monitoring of the concentration of suspended solid substances (SS).

[0056] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly and to be implemented in accordance with the content of the description, the following describes in detail with reference to the preferred embodiments of the present invention and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic structural diagram of a device and method for detecting the concentration of suspended solid substances in an embodiment of the present invention;

[0058] Figure 2 It is a specific flowchart of the image preprocessing unit of the present invention;

[0059] Figure 3 It is the specific process of the Inception-ResNet coupled neural network model selected by the image classification unit of the present invention;

[0060] Figure 4 It is the structure of the Inception module of the present invention;

[0061] Figure 5 It is the structure of the residual unit of the present invention.

[0062] Description of the reference numerals in the drawings:

[0063] 1. Closed shooting box;

[0064] 21. LED light source with adjustable light intensity and color temperature; 22. Frosted diffuser plate;

[0065] 3. Sample fixing module; 31. Beaker placement rack

[0066] 4. Camera module; 41. Fixed focal length lens; 42. Anti-reflection coating transparent glass;

[0067] 5. Magnetic stirring module;

[0068] 6. Image analysis module; 61. Preprocessing unit; 62. Classification unit;

[0069] 7. External computing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0070] The following will further describe in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0071] As Figure 1 shown, a device for detecting the concentration of suspended solid substances (SS) includes a closed shooting box 1, a light-emitting bottom plate, a sample fixing module 3, a camera module 4, a magnetic stirring module 5, an image analysis module 6, and an external computing device 7. The closed shooting box 1 is used to isolate the interference of external light and is combined with the sample fixing module 3 with a hollow middle part to provide a completely black and interference-free background image; the light-emitting bottom plate provides a light source for shooting and enhances the difference between pictures of SS suspension solutions with different concentrations through different light transmittance; the sample fixing module 3 is used to fix the SS suspension solution sample; the camera module 4 is installed on the top of the shooting box to shoot the cross-sectional image of the SS suspension solution from top to bottom; the magnetic stirring module 5 provides an adjustable stirring speed for the suspension solution.

[0072] The suspended solid substance (SS) concentration detection device of this embodiment, through the combined action of the shooting box 1, the light-emitting bottom plate, the sample fixing module 3, the camera module 4, and the magnetic stirring module 5, provides an SS suspension solution image with clear image, no light and background interference for the subsequent image analysis module 6.

[0073] The specific process of the preprocessing unit 61 in the image analysis module is as Figure 2 shown. Grayscale conversion is the process of converting a color image into a grayscale image, and the three-channel values of the pixel points in the original image are mapped to grayscale values through specific rules. The grayscale conversion method selected in this embodiment is the weighted average method, and the characteristic of this method is that it can more truly reflect the original brightness characteristics of the picture. The grayscale value is Gray, and R, G, and B represent the red, green, and blue channels respectively. The specific formula is as follows (1):

[0074] Gray = 0.299·R + 0.587·G + 0.114·B (1)

[0075] The noise reduction method selected in this embodiment is the Gaussian filtering method. Its basic principle is to use the weight values generated by the Gaussian function to perform weighted averaging on the image pixel points, thereby smoothing the image to weaken the noise. Compared with the traditional mean filtering method, the weight value of the center position of the selected area in this method is the highest, and the weight value is lower closer to the edge, so that the edge information of the image can be better retained and the image blurring effect can be reduced. (x, y) represents the offset of the pixel in the filtering core relative to the center point, and σ is the standard deviation. Its weight value calculation formula is as follows (2):

[0076]

[0077] In this embodiment, the selected image enhancement algorithm is the histogram equalization method. This method uses a histogram to represent the frequency distribution of an image and can redistribute the gray-level distribution of the image through non-linear mapping processing, thereby increasing the difference between different gray levels in the image to enhance the visual effect of the image. For images with a concentrated histogram distribution or low contrast, the histogram equalization method has good processing effects, but this method is not suitable for images with a wide histogram distribution range.

[0078] In this embodiment, the selected feature extraction method is Canny edge feature extraction. This method can detect significant edges in an image and remove the black background in the collected SS suspension image, greatly reducing the computational cost of the Inception-ResNet model in the subsequent image classification unit 62.

[0079] In this embodiment, the neural network model selected by the image classification unit 62 is the coupled Inception-ResNet network model. The specific modules of this network are as Figure 3 shown. Among them, the Inception network uses convolutional kernels of different sizes at the same time, enabling the model to obtain feature information of different dimensions while effectively reducing the computational complexity and avoiding the occurrence of the gradient vanishing problem. The Inception module structure is as Figure 4 shown. The ResNet network improves the information propagation efficiency by adding direct connections (also known as residual connections) to the non-linear convolutional layers. Its basic principle is to split the objective function into two parts: the identity function x and the residual function h(x)-x. Use the non-linear unit f(x;θ) to approximate the residual function h(x)-x, that is, use f(x;θ)+x to approximate h(x). For a computer, it is easier to approximate the residual function with a non-linear function. At the same time, because there is a residual part x, the occurrence of the gradient vanishing problem is avoided. Therefore, a neural network using a residual network module can be stacked deeper, greatly improving the performance of the neural network model. The residual unit structure is as Figure 5 shown.

[0080] The specific implementation method of this embodiment is as follows:

[0081] S1, Equipment installation and inspection: Ensure that the enclosed shooting box is correctly assembled to avoid the influence of external light and internal light reflection on shooting; Check whether the LED light source of the light-emitting bottom plate is normal and whether the surface frosted diffuser plate is properly covered to ensure the uniformity of illumination; Ensure that the magnetic stirring module can operate normally; Check whether the camera port of the camera module is contaminated to avoid visual obstacles and ensure that the shooting field of view completely covers a 1000 ml beaker.

[0082] S2, Equipment Initialization: Turn on the power supply and start the luminous bottom plate; Activate the magnetic stirring module and set the stirring speed to a moderate constant value (recommended rotation speed is 500 rpm); Start the camera module, test the image acquisition function, and confirm that the image is clear and distortion-free.

[0083] S3, Sample Treatment and Fixation: Pour the sample suspension of the suspended solid substance (SS) to be measured into a 1000 ml beaker, ensuring the transparency and cleanliness of the beaker to avoid affecting the optical imaging quality; Place the beaker containing the sample into the sample fixation module, ensuring that the hollow part of the beaker is fully fitted to prevent the sample from shaking or displacing; The sample fixation module should be placed at the central position of the magnetic stirring module to ensure that the action of the stirrer is evenly distributed in the sample.

[0084] S4, Image Acquisition: Start the luminous bottom plate and the magnetic stirring module. After stirring for 10 - 15 seconds, ensure that the suspended particles in the sample are evenly distributed; Start the camera module and vertically shoot the sample image downward. After shooting, save the image to the specified directory of the computing device, ensuring that the file name for each shot is unique (such as sample number + timestamp); To verify the stability of the results, the sample images can be repeatedly acquired at different stirring times for multiple subsequent image analyses.

[0085] S5, Image Preprocessing: Input the acquired image into a pre-set program and perform operations such as grayscale conversion, denoising, contrast enhancement, and edge feature extraction in sequence.

[0086] S6, Image Classification Model Analysis: Adjust the preprocessed image to the standard size of 299×299 pixels and input it into the Inception-ResNet model; Use the convolutional layer and residual connections of the model to extract the texture features, density distribution, and optical properties of the particles in the image; Use the fully connected layer to classify the concentration of the SS suspension sample and output the SS concentration level of the sample. The preset measurement range of the model installed on this device is 0 - 1000 mg SS / L, and a concentration difference of 50 mg SS / L is regarded as one level, with a total of 20 SS concentration levels.

[0087] S7, Result Output and Storage: The detection results are displayed in real-time through the computing device, including information such as sample number, detection time, stirring speed, SS concentration level, etc.; Store the images and analysis results of each detection in the local database to support subsequent result traceability and comparative analysis; Automatically generate a detection report, including images, concentration analysis results, experimental parameters, etc.

[0088] S8, Cleaning and Maintenance: After the detection is completed, remove the beaker and the fixed sample module, clean and dry them to avoid sample residue affecting subsequent detections; perform equipment maintenance regularly, such as replacing the light-emitting bottom plate, checking the cleanliness of the camera module lens, and the stirring effect of the magnetic stirring module, to ensure stable equipment performance.

[0089] The present invention can provide a standardized detection environment through an integrated device, and at the same time use an Inception-ResNet coupled neural network to identify and analyze the concentration level of the collected suspended solid (SS) suspension sample images. On the one hand, it can replace the traditional cumbersome laboratory analysis method to achieve fast and non-contact concentration detection; on the other hand, using the Inception-ResNet coupled neural network model improves the robustness and generalization ability of the model, ensuring the accuracy of the detection.

[0090] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. An apparatus for rapidly detecting the concentration of suspended solid substances, characterized in that, Comprising: A closed shooting box (1) with a black inner wall, used to provide a completely black background environment; A light-emitting bottom plate, installed at the bottom of the box, used to provide an upward uniform light source, so that the light passes through the suspension of suspended solid substances from bottom to top to enhance the contrast of the image; A sample fixing module (3), which is a black solid device with a hollow in the middle. The hollow part is a beaker placement rack (31), used to stabilize the sample beaker; A camera module (4), fixedly installed on the top of the box, used to vertically downwardly shoot the image of the suspension of suspended solid substances in the beaker; A magnetic stirring module (5), arranged below the light-emitting bottom plate, used to stir the suspended solid substances during shooting to make them evenly distributed; An image analysis module (6), used to preprocess and classify and analyze the concentration of the images collected by the camera module (4) to obtain the classification result of the concentration of suspended solid substances; An external computing device (7), with the image analysis module (6) built-in and directly connected to the camera module (4), used to generate a concentration detection report through the real-time classification result and support the storage and comparative analysis of historical data.

2. The device for rapidly detecting the concentration of suspended solid substances according to claim 1, wherein The light-emitting bottom plate includes an LED light source (21) with adjustable light intensity and color temperature, and the surface is covered with a frosted diffuser plate (22), used to ensure the uniformity of light and reduce the light reflection at the edge of the beaker.

3. An apparatus for rapidly detecting the concentration of suspended solid substances according to claim 1, characterized in that, The shooting module (4) is a high-resolution industrial camera, equipped with a fixed focal length lens (41), and its aperture is set to a constant value to avoid the influence of light intensity fluctuation on the shooting quality.

4. An apparatus for rapidly detecting the concentration of suspended solid substances according to claim 3, characterized in that, The top of the closed shooting box (1) is also provided with an anti-reflection coating transparent glass (42), used to protect the camera module and avoid interference from external stray light at the same time.

5. An apparatus for rapidly detecting the concentration of suspended solid substances according to claim 1, characterized in that, The magnetic stirring module (5) has a constant speed stirring function to ensure that the suspended solid substances always maintain a uniformly dispersed state during shooting.

6. An apparatus for rapidly detecting the concentration of suspended solid substances according to claim 1, characterized in that, The shape of the hollow part completely fits the outer shape of a 1000 ml beaker.

7. An apparatus for rapidly detecting the concentration of suspended solid substances according to claim 1, characterized in that, The image analysis module includes: A preprocessing unit (61), used to grayscale, denoise, enhance the contrast and extract edge features of the collected images; A classification unit (62), used to extract features from the preprocessed images based on the Inception-ResNet deep learning model, and finally output the classification result of the concentration of suspended solid substances (SS); 8. A device and method for quickly detecting the concentration of suspended solid substances according to claim 7, characterized in that, The preprocessing unit (61) is specifically used for: S1, Grayscaling: Convert the original color image into a grayscale image to reduce the computational complexity; S2, Denoising: Use the Gaussian filtering algorithm to remove image noise; S3, Image contrast enhancement: Use the method based on histogram equalization to enhance the contrast of the collected images; S4, Edge feature extraction: Use the Canny edge detection algorithm to extract the edge features inside the SS suspension; 9. The device for rapidly detecting the concentration of suspended solid substances according to claim 8, wherein, The classification unit (62) is specifically used for: S1, Use the pre-trained model of the Inception-ResNet neural network for feature extraction. The pre-trained model includes multiple convolutional modules and residual connections. Among them, the input image size is 299×299 pixels; S2. The final classification layer is a Softmax fully connected layer, which is divided into 20 categories in total, and outputs the SS suspension concentration level of 50 - 1000 mgSS / L.

10. A method for rapidly detecting the concentration of suspended solid substances by applying the device according to any one of claims 1-9, characterized in that, It includes the following steps: Step 1, equipment installation and inspection: Ensure that the enclosed shooting box is correctly assembled to avoid the influence of external light and internal light reflection on shooting; check whether the LED light source of the light-emitting bottom plate is normal and whether the surface frosted diffuser plate is normally covered to ensure the uniformity of illumination; ensure that the magnetic stirring module operates normally; check whether the camera port of the camera module is contaminated to avoid visual field obstacles and ensure that the shooting visual field completely covers the beaker. Step 2, equipment initialization: Turn on the power supply and start the light-emitting bottom plate; turn on the magnetic stirring module and set the stirring speed to a moderate constant value; start the camera module and test the image acquisition function to confirm that the image is clear and distortion-free. Step 3, sample processing and fixation: Pour the suspension sample of the suspended solid substance to be measured into the beaker, ensuring the transparency and cleanliness of the beaker to avoid affecting the optical imaging quality; place the beaker containing the sample into the sample fixation module, ensuring that the hollow part of the beaker and the sample fixation module are completely fitted to prevent the sample from shaking or displacing; place the sample fixation module at the central position of the magnetic stirring module to ensure that the action of the stirrer is evenly distributed in the sample. Step 4, image acquisition: Start the light-emitting bottom plate and the magnetic stirring module. After stirring for 10 - 15 seconds, ensure that the suspended particles in the sample are evenly distributed; start the camera module and vertically shoot the sample image. After shooting, save the image to the specified directory of the computing device, ensuring that the file name of each shooting is unique; repeat the acquisition of the sample image at different stirring times for multiple subsequent image analyses. Step 5, image preprocessing: Input the collected images into a pre-set program, and successively perform grayscale conversion, denoising, contrast enhancement, and edge feature extraction operations. Step 6, image classification model analysis: Adjust the preprocessed image to the standard size of 299×299 pixels and input it into the Inception-ResNet model; use the convolutional layer and residual connection of the model to extract the texture features, density distribution, and optical properties of the particles in the image; use the fully connected layer to classify the concentration of the SS suspension sample and output the SS concentration level of the sample. The preset measurement range of the model is 0 - 1000 mgSS / L, and a concentration difference of 50 mgSS / L is regarded as one level, with a total of 20 SS concentration levels. Step 7, result output and storage: The detection results are displayed in real time through the computing device, including sample number, detection time, stirring speed, and SS concentration level information; store the images and analysis results of each detection in the local database; automatically generate a detection report, including images, concentration analysis results, and experimental parameters. Step 8, cleaning and maintenance: After the detection, remove the beaker and the sample fixation module, clean and dry them to avoid sample residue affecting subsequent detections; perform regular equipment maintenance, including replacing the light-emitting bottom plate, checking the cleanliness of the camera module lens, and the stirring effect of the magnetic stirring module to ensure the stable performance of the equipment.

Images