Rapid detection method for transparent extracellular polymeric particles in water body based on deep learning

Abstract

The invention relates to the technical field of water environment detection, and particularly discloses a deep learning-based rapid detection method for transparent extracellular polymeric particles in a water body, and the method comprises the following steps: S1, obtaining N sample images, and augmenting the sample images to obtain M * N first target images, enhancing the first target image to obtain a second target image, and dividing the second target image into a training set and a test set according to a preset proportion; and S2, establishing a concentration prediction model based on a deep learning model, training and verifying the concentration prediction model based on the training set and the test set, and predicting the TEP concentration based on the verified concentration prediction model. According to the invention, TEP concentration measurement can be rapidly and conveniently carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of water environment detection, and in particular to a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning. Background Art

[0002] Transparent extracellular polymeric particles (TEP) are a type of transparent particles mainly composed of acidic polysaccharides that can be stained with Alcian blue. Due to their high carbon content and high viscosity, they can adsorb and aggregate various organisms, and therefore have attracted the attention of more and more scholars.

[0003] The quantitative determination method of TEP includes microscopy and colorimetry. During the microscopy, the filter membrane collected after staining needs to be stained with Alcian blue, and then the film is processed and observed under a microscope to obtain the image. Finally, the counting result, that is, the abundance content of TEP, is obtained by processing with specific image software. Since the obtained image needs to undergo a series of denoising processes, etc., before a reasonable count can be performed, the operation steps are cumbersome, and the subsequent image data processing takes a lot of time; although the colorimetric method is simpler, it needs to be soaked and extracted with high-concentration concentrated sulfuric acid, and then the absorbance is measured using a colorimetric dish. In this process, if the operation is improper, concentrated sulfuric acid will be splashed, and the use of concentrated sulfuric acid as a waste reagent can cause soil and water acidification, which is extremely unfriendly to the environment. Therefore, there is an urgent need for a method to quickly and easily obtain TEP in a water environment. Summary of the invention

[0004] The purpose of the present invention is to provide a rapid detection method for transparent extracellular polymeric particles in water based on deep learning to solve the above-mentioned technical problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A method for rapid detection of transparent extracellular aggregate particles in water based on deep learning, comprising the following steps: S1: Obtain N sample images, where N is a preset number, augment the sample images to obtain M*N first target images, where M represents a preset constant, enhance the first target images to obtain second target images, and divide the second target images into a training set and a test set according to a preset ratio; S2: establishing a concentration prediction model based on the deep learning model, training and verifying the concentration prediction model based on the training set and the test set, and predicting the TEP concentration based on the verified concentration prediction model; The process of training and verifying the concentration prediction model specifically includes: Inputting the training set into the concentration prediction model to obtain a training set prediction output, obtaining a training set loss value through a loss function, and determining a gradient of a parameter of the concentration prediction model based on the training set loss value; Inputting the test set into the concentration prediction model to obtain the test set prediction output, obtaining the training set loss value through the loss function, and performing gradient descent; Repeat the above steps until the number of gradient descents reaches a preset value and / or the concentration prediction model converges.

[0006] As a further solution of the present invention: in the step S1, the process of augmenting the sample image specifically includes: Rotate the sample image, and the angle interval of the sample image rotation is [-12°, +12°]; The sample image is scaled, and the scaling ratio range of the sample image is [-15%, +15%].

[0007] As a further solution of the present invention: in the step S1, the process of enhancing the first target image specifically includes: Flipping the first sample image horizontally, with a probability of 50%; The first sample image is vertically flipped, and the probability of vertical flipping is 50%; Cropping the first sample image, with a probability of 50% and a cropping ratio interval of [0.7, 1]; The first sample image is rotated, the rotation angle interval is [-30°, +30°], and the rotation probability is 50%.

[0008] As a further solution of the present invention: in the step S1, the ratio of the training set to the test set is 3:1.

[0009] As a further solution of the present invention: in the step S2, during the training and verification of the concentration prediction model, cross entropy is used as the loss function.

[0010] As a further solution of the present invention: in the step S2, the process of establishing the concentration prediction model specifically includes: The deep learning model selects the convolutional neural network LeNet, and the convolutional neural network LeNet includes 3 convolutional layers, 2 pooling layers, 1 fully connected layer, and 1 output layer; The convolution layer extracts the features of the second target image through a filter and introduces nonlinearity through an activation function; The pooling layer reduces the dimension of the features of the second target image to obtain a one-dimensional vector; The fully connected layer classifies the one-dimensional vector by weight calculation; The output layer is used to output the prediction results.

[0011] As a further solution of the present invention: in the step S1, the process of obtaining the sample image specifically includes: In an acidic medium, an acidic Alcian blue solution is prepared using Alcian blue as a dye, and N xanthan gum solutions of preset concentrations are prepared based on the acidic Alcian blue solution. The xanthan gum solution is filtered onto a filter membrane with a pore size of 0.4 microns, and after a preset time interval, pure water is dripped onto the filter membrane, and the water is dried to obtain a standard sample; Under preset environmental conditions, an image of the standard sample is collected at a preset position and used as a sample image.

[0012] As a further solution of the present invention: in the process of obtaining the sample image, the acidic medium is a glacial acetic acid solution, the concentration of the acidic Alcian blue solution is 0.02%, pH=2.5, and the material of the filter membrane includes a nylon filter membrane and a polycarbonate membrane.

[0013] Beneficial effects of the present invention: Compared with the prior art, this solution does not require the use of a microscope for detection, but can complete image acquisition through the camera function, eliminating the reliance on professional equipment and microscope inspection, making the operation simpler and faster; in addition, compared with the colorimetric method, this method does not require the use of dangerous chemical reagents such as concentrated sulfuric acid, avoiding environmental pollution, reducing safety hazards in experimental operations, and making the entire detection process more green and environmentally friendly. Since no expensive equipment and reagents are required, this solution can significantly reduce costs, thereby having a higher economic efficiency; By introducing a deep learning model, the solution also realizes automated detection and can complete quantitative analysis of TEP concentration in a short time, greatly improving the speed and efficiency of detection. After model training, simple image acquisition can quickly complete the analysis without complex experimental operations, shortening the detection time and reducing the error of manual operation. In addition, this technology based on image processing and deep learning also has good scalability and can provide technical support for real-time online monitoring. It is suitable for TEP concentration detection in multiple scenarios, such as industry, environmental monitoring and other fields, and has broad application prospects. Through the augmentation of a large number of images and model training, the scheme also improves the accuracy and robustness of the prediction model, and can provide more stable and accurate quantitative analysis results. Overall, the scheme realizes the rapid detection of TEP concentration by simplifying equipment requirements, reducing environmental pollution, and improving detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings.

[0015] Figure 1 It is a schematic flow chart of a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning, comprising the following steps: S1: Obtain N sample images, where N is a preset number, augment the sample images to obtain M*N first target images, where M represents a preset constant, enhance the first target images to obtain second target images, and divide the second target images into a training set and a test set according to a preset ratio; S2: establishing a concentration prediction model based on the deep learning model, training and verifying the concentration prediction model based on the training set and the test set, and predicting the TEP concentration based on the verified concentration prediction model; The process of training and verifying the concentration prediction model specifically includes: Inputting the training set into the concentration prediction model to obtain a training set prediction output, obtaining a training set loss value through a loss function, and determining a gradient of a parameter of the concentration prediction model based on the training set loss value; Inputting the test set into the concentration prediction model to obtain the test set prediction output, obtaining the training set loss value through the loss function, and performing gradient descent; Repeat the above steps until the number of gradient descents reaches a preset value and / or the concentration prediction model converges.

[0018] It should be noted that image acquisition can be completed through the camera function, eliminating the need for professional equipment and microscope inspection, making the operation simpler and faster; in addition, compared with the colorimetric method, this method does not require the use of dangerous chemical reagents such as concentrated sulfuric acid, avoiding environmental pollution, reducing safety hazards in experimental operations, and making the entire detection process more environmentally friendly. Since expensive equipment and reagents are not required, this solution can significantly reduce costs and thus has a higher economic efficiency; By introducing a deep learning model, the solution also realizes automated detection and can complete quantitative analysis of TEP concentration in a short time, greatly improving the speed and efficiency of detection. After model training, simple image acquisition can quickly complete the analysis without complex experimental operations, shortening the detection time and reducing the error of manual operation. In addition, this technology based on image processing and deep learning also has good scalability and can provide technical support for real-time online monitoring. It is suitable for TEP concentration detection in multiple scenarios, such as industry, environmental monitoring and other fields, and has broad application prospects. Through the augmentation of a large number of images and model training, the scheme also improves the accuracy and robustness of the prediction model, and can provide more stable and accurate quantitative analysis results. Overall, the scheme realizes the rapid detection of TEP concentration by simplifying equipment requirements, reducing environmental pollution, and improving detection efficiency and accuracy.

[0019] In another preferred embodiment of the present invention, in step S1, the process of augmenting the sample image specifically includes: Rotate the sample image, and the angle interval of the sample image rotation is [-12°, +12°]; The sample image is scaled, and the scaling ratio range of the sample image is [-15%, +15%].

[0020] It is worth noting that, since the sample images are limited, more sample images, namely the first target images, are obtained by data enhancement.

[0021] In another preferred embodiment of the present invention, in step S1, the process of enhancing the first target image specifically includes: Flipping the first sample image horizontally, with a probability of 50%; The first sample image is vertically flipped, and the probability of vertical flipping is 50%; Cropping the first sample image, with a probability of 50% and a cropping ratio interval of [0.7, 1]; The first sample image is rotated, the rotation angle interval is [-30°, +30°], and the rotation probability is 50%.

[0022] It is understandable that the purpose of enhancing the first target image is to further increase the diversity of the data set and improve the generalization ability of the model, thereby improving the performance of the concentration prediction model in practical applications. Data enhancement technology generates new samples by artificially performing various transformations on the original image, effectively reducing the risk of model overfitting and making the model more robust when facing different scenes or shooting conditions.

[0023] In another preferred embodiment of the present invention, in step S1, the ratio of the training set to the test set is 3:1.

[0024] In another preferred embodiment of the present invention, in step S2, during the training and verification of the concentration prediction model, cross entropy is used as the loss function.

[0025] It should be noted that the cross entropy compares the true label with the probability distribution of the model output. When the probability of the model output is closer to the true label, the cross entropy value is smaller. When the gap between the prediction and the true distribution is large, the cross entropy value is larger. Therefore, by minimizing the cross entropy loss, the model's prediction will gradually approach the true value distribution.

[0026] In another preferred embodiment of the present invention, in step S2, the process of establishing the concentration prediction model specifically includes: The deep learning model selects the convolutional neural network LeNet, and the convolutional neural network LeNet includes 3 convolutional layers, 2 pooling layers, 1 fully connected layer, and 1 output layer; The convolution layer extracts the features of the second target image through a filter and introduces nonlinearity through an activation function; The pooling layer reduces the dimension of the features of the second target image to obtain a one-dimensional vector; The fully connected layer classifies the one-dimensional vector by weight calculation; The output layer is used to output the prediction results.

[0027] In another preferred embodiment of the present invention, the process of acquiring the sample image specifically includes: In an acidic medium, an acidic Alcian blue solution is prepared using Alcian blue as a dye, and N xanthan gum solutions of preset concentrations are prepared based on the acidic Alcian blue solution. The xanthan gum solution is filtered onto a filter membrane with a pore size of 0.4 microns, and after a preset time interval, pure water is dripped onto the filter membrane, and the water is dried to obtain a standard sample; Under preset environmental conditions, an image of the standard sample is collected at a preset position and used as a sample image.

[0028] In another preferred embodiment of the present invention, in the process of acquiring the sample image, the acidic medium is a glacial acetic acid solution, the concentration of the acidic Alcian blue solution is 0.02%, pH=2.5, and the material of the filter membrane includes a nylon filter membrane and a polycarbonate membrane.

[0029] It is worth noting that the specific operation process is as follows: take 0.1 ± 0.0001 g of xanthan gum dried at 110oC for 2 hours, dissolve it in 100 ml of ultrapure water, and then stir and shake to prepare a 1 g / L stock solution; then, dilute it according to a certain ratio to obtain standard solutions with concentrations of 2 mg / L, 4 mg / L, 6 mg / L, 8 mg / L and 10 mg / L, as well as a blank group; A 0.02% by mass volume acidic Alcian blue solution and a 0.06% by mass volume acetic acid solution were prepared, and then the prepared solution was further adjusted to a pH of 2.5 using acetic acid; Use a pipette to draw up 5 ml of the standard solution and collect the sample by filtration using a 0.4 μm pore size polycarbonate membrane under low pressure (150 mmHg) vacuum; After staining the collected filter membrane with the staining solution for 2-5 seconds, the excess staining solution is then removed with pure water; The collected filter membranes were dried with absorbent paper to remove excess water, and placed at a fixed position and height to take photos without direct light.

[0030] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A rapid detection method for transparent extracellular aggregate particles in water based on deep learning, characterized in that: The following steps are involved: S1: Obtain N sample images, where N is a preset number, augment the sample images to obtain M*N first target images, where M represents a preset constant, enhance the first target images to obtain second target images, and divide the second target images into a training set and a test set according to a preset ratio; S2: establishing a concentration prediction model based on the deep learning model, training and verifying the concentration prediction model based on the training set and the test set, and predicting the TEP concentration based on the verified concentration prediction model; The process of training and verifying the concentration prediction model specifically includes: Inputting the training set into the concentration prediction model to obtain a training set prediction output, obtaining a training set loss value through a loss function, and determining a gradient of a parameter of the concentration prediction model based on the training set loss value; Inputting the test set into the concentration prediction model to obtain the test set prediction output, obtaining the training set loss value through the loss function, and performing gradient descent; Repeat the above steps until the number of gradient descents reaches a preset value and / or the concentration prediction model converges.

2. According to a method for rapid detection of transparent extracellular aggregate particles in water based on deep learning in claim 1, it is characterized in that: In the step S1, the process of augmenting the sample image specifically includes: Rotate the sample image, and the angle interval of the sample image rotation is [-12°, +12°]; The sample image is scaled, and the scaling ratio range of the sample image is [-15%, +15%].

3. According to a method for rapid detection of transparent extracellular aggregate particles in water based on deep learning in claim 1, it is characterized in that: In the step S1, the process of enhancing the first target image specifically includes: Flipping the first sample image horizontally, with a probability of 50%; The first sample image is vertically flipped, and the probability of vertical flipping is 50%; Cropping the first sample image, with a probability of 50% and a cropping ratio interval of [0.7, 1]; The first sample image is rotated, the rotation angle interval is [-30°, +30°], and the rotation probability is 50%.

4. According to a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning in claim 1, it is characterized in that: In step S1, the ratio of the training set to the test set is 3:

1.

5. According to a method for rapid detection of transparent extracellular aggregate particles in water based on deep learning in claim 1, it is characterized in that: In the step S2, cross entropy is used as the loss function during the training and verification of the concentration prediction model.

6. According to a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning in claim 1, it is characterized in that: In step S2, the process of establishing the concentration prediction model specifically includes: The deep learning model selects the convolutional neural network LeNet, and the convolutional neural network LeNet includes 3 convolutional layers, 2 pooling layers, 1 fully connected layer, and 1 output layer; The convolution layer extracts the features of the second target image through a filter and introduces nonlinearity through an activation function; The pooling layer reduces the dimension of the features of the second target image to obtain a one-dimensional vector; The fully connected layer classifies the one-dimensional vector by weight calculation; The output layer is used to output the prediction results.

7. According to a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning in claim 1, it is characterized in that: In the step S1, the process of obtaining the sample image specifically includes: In an acidic medium, an acidic Alcian blue solution is prepared using Alcian blue as a dye, and N xanthan gum solutions of preset concentrations are prepared based on the acidic Alcian blue solution. The xanthan gum solution is filtered onto a filter membrane with a pore size of 0.4 microns, and after a preset time interval, pure water is dripped onto the filter membrane, and the water is dried to obtain a standard sample; Under preset environmental conditions, an image of the standard sample is collected at a preset position and used as a sample image.

8. According to claim 7, a method for rapid detection of transparent extracellular polymeric particles in water based on deep learning, characterized in that: In the process of acquiring the sample image, the acidic medium is a glacial acetic acid solution, the concentration of the acidic Alcian blue solution is 0.02%, the pH value is 2.5, and the materials of the filter membrane include a nylon filter membrane and a polycarbonate membrane.

Images