A method and apparatus for measuring photocatalytic efficiency based on laser speckle and deep learning

Abstract

This invention relates to the field of computer vision processing technology, and more particularly to a method and apparatus for measuring photocatalytic efficiency based on laser speckle and deep learning. The photocatalytic efficiency measurement method includes: acquiring a speckle image of the test liquid; performing ROI cropping, Gaussian filtering, and histogram equalization on the speckle image; extracting speckle image features using the Inception-v3 model, and constructing a training concentration prediction model using a random forest regression algorithm; using the concentration prediction model to predict the concentration of the real-time acquired speckle image, and calculating the photocatalytic efficiency according to the degradation rate formula. This invention utilizes deep learning to automatically extract speckle image features and achieves non-contact detection through laser speckle, improving prediction accuracy. The automated sampling mechanism avoids the risk of direct exposure of experimental personnel to ultraviolet light, enabling continuous monitoring of the photocatalytic reaction process.

Description

Technical Field

[0001] This invention relates to the field of computer vision processing technology, and in particular to a method and apparatus for measuring photocatalytic efficiency based on laser speckle and deep learning. Background Technology

[0002] Photocatalysis, as a green and efficient means of environmental governance and energy conversion, has attracted much attention due to its wide application in pollutant degradation, energy production, and green synthesis. Semiconductor photocatalysis generates electron-hole pairs through photoexcitation, driving redox reactions to achieve processes such as organic pollutant degradation, water splitting for hydrogen production, and carbon dioxide reduction.

[0003] Photocatalytic efficiency is often characterized by degradation rate, which directly reflects the degree of removal of the target substance in the photocatalytic reaction. However, traditional absorption spectrometers have significant limitations in measuring degradation rates. In terms of equipment cost, high-end absorption spectrometers are expensive, with a single unit costing millions, and the cost of supporting equipment and daily maintenance further increases the operating cost. Regarding the physical characteristics of the equipment, traditional spectrometers are bulky, typically weighing over 50 kg, making them difficult to move and deploy, and each unit can only process one experimental sample at a time, resulting in low efficiency. More importantly, the measurement accuracy of traditional absorption spectrometers is limited by the shot noise limit, a fundamental physical limitation inherent in measurement methods based on classical light sources. This means that their sensitivity to low-concentration substances is insufficient, and they are easily affected by instrument noise and background interference, especially in the later stages of the photocatalytic reaction when the concentration of the target substance decreases, making accurate measurement difficult and affecting the assessment of degradation rate. These limitations severely restrict the practical application of degradation rate measurement based on absorption spectrometers, particularly for the accurate assessment of samples with low absorbance or low concentration.

[0004] Therefore, developing a low-cost, non-contact, and highly sensitive method for measuring photocatalytic efficiency is of significant scientific and practical value. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method and device for measuring photocatalytic efficiency based on laser speckle and deep learning, which realizes rapid and accurate prediction of solution concentration during photocatalytic degradation and effectively characterizes photocatalytic efficiency through degradation rate.

[0006] The following is a summary of this disclosure to provide a basic understanding of some aspects. This summary is not intended to identify key or important elements, nor is it intended to limit the implementation or any aspects of the claims. Furthermore, this summary provides a simplified overview of some aspects that can be described in more detail in other parts of this disclosure.

[0007] The present invention solves the above-mentioned technical problems through the following technical means:

[0008] In a first aspect, embodiments of this application provide a method for measuring photocatalytic efficiency based on laser speckle and deep learning, comprising the following steps:

[0009] Acquire speckle images of the photocatalytic test solution;

[0010] Perform ROI cropping, Gaussian filtering, and histogram equalization on speckle images;

[0011] The Inception-v3 model was used to extract speckle image features. Principal component analysis was introduced to reduce the dimensionality of the speckle image features. A concentration prediction model was constructed by combining the random forest regression algorithm. The concentration prediction model was then trained and validated.

[0012] A concentration prediction model was used to predict the concentration of speckle images acquired in real time, and the photocatalytic efficiency was calculated based on the degradation rate formula.

[0013] In conjunction with the first aspect, in some embodiments, the speckle image is an image acquired by placing the photocatalyzed test liquid in an optical measurement mechanism, the optical measurement mechanism including a laser, an optical modulation component and an imaging component, the optical modulation component including a variable aperture and a ground glass diffuser, the test liquid being located between the variable aperture and the ground glass diffuser; the imaging component including a 4f imaging element and a CCD camera.

[0014] In conjunction with the first aspect, in some embodiments, the laser emitter is provided with a beam expander. After the laser emitted by the laser is expanded and collimated by the beam expander, the resulting parallel beam has its beam diameter adjusted by a variable aperture. After irradiating the liquid to be tested, the beam forms a scattering speckle field after passing through a frosted glass scatterer, and the speckle image is received by the imaging component.

[0015] In conjunction with the first aspect, in some implementations, the step of using the Inception-v3 model to extract speckle image features, introducing principal component analysis to reduce the dimensionality of the speckle image features, constructing a concentration prediction model using a random forest regression algorithm, and training and validating the concentration prediction model includes:

[0016] A transfer learning strategy was adopted to extract speckle image features based on the ImageNet pre-trained Inception-v3 model;

[0017] Principal component analysis is introduced to reduce the dimensionality of speckle image features;

[0018] Concentration prediction model is constructed by combining random forest regression algorithm;

[0019] The performance of the concentration prediction model in training is evaluated based on the mean absolute error, mean square error, and coefficient of determination score.

[0020] The stability of the concentration prediction model was verified by five-fold cross-validation to obtain the final concentration prediction model.

[0021] In conjunction with the first aspect, in some implementation methods,

[0022] The photocatalytic efficiency is characterized by the degradation rate, and the degradation rate is calculated using the following formula:

[0023]

[0024] in, Indicates the initial concentration. This indicates that the reaction has proceeded to... The predicted concentration of the solution at a given time.

[0025] Secondly, embodiments of this application provide a photocatalytic efficiency measurement device based on laser speckle and deep learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the photocatalytic efficiency measurement method described in the first aspect.

[0026] Thirdly, embodiments of this application provide a photocatalytic efficiency measurement system based on laser speckle and deep learning, including: a photocatalytic reaction device for providing a photocatalytic degradation environment;

[0027] An automated sampling mechanism is installed on the photocatalytic reaction device to automatically sample the degradation liquid in the photocatalytic reaction device in real time.

[0028] An optical measurement mechanism is used to acquire speckle images of the degradation solution to be tested;

[0029] A computer device for processing speckle images, extracting features, and predicting the concentration of degradation solution to obtain photocatalytic efficiency; the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the photocatalytic efficiency measurement method described in the first aspect.

[0030] In conjunction with the third aspect, in some embodiments, the photocatalytic reaction device includes a light-shielding reaction chamber, and a quartz cold trap, a reaction flask, an ultraviolet strip lamp, and a magnetic stirrer installed inside the light-shielding reaction chamber, as well as a light source voltage regulator installed on the top outside the light-shielding reaction chamber; the reaction flask is used to hold the solution, and the magnetic stirrer is installed below the reaction flask to stir the solution inside the reaction flask; the light source voltage regulator drives the ultraviolet strip lamp light source to provide ultraviolet illumination; the quartz cold trap is circulated and cooled by an external refrigerator to control the temperature inside the reaction flask.

[0031] In conjunction with the third aspect, in some embodiments, the automated sampling mechanism includes a sampling cylinder, a piston rod, and a stepper motor. One end of the sampling cylinder is connected to the interior of the reaction flask, and a sampling tube is provided on the sampling cylinder outside the light-shielding reaction chamber. The output shaft of the stepper motor is connected to the piston rod, and the piston rod is slidably installed inside the sampling cylinder. The stepper motor controls the piston rod to perform piston movement inside the sampling cylinder.

[0032] In conjunction with the third aspect, in some embodiments, along the laser emission direction, the optical measurement mechanism includes a laser, a variable aperture, a sample stage, a frosted glass diffuser, a variable aperture, an imaging lens, a variable aperture, an imaging lens, and a CCD camera arranged sequentially, with the liquid to be tested placed on the sample stage.

[0033] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0034] This invention relates to a photocatalytic efficiency measurement system based on laser speckle and deep learning. Firstly, by designing the optical path based on the laser speckle principle, it achieves lower costs compared to traditional photocatalytic efficiency detection methods. Secondly, it utilizes deep learning to automatically extract speckle image features and achieves non-contact detection through laser speckle, effectively improving prediction accuracy. Thirdly, through an automated sampling mechanism, it not only avoids the risk of direct exposure of experimental personnel to ultraviolet light but also enables continuous monitoring of the photocatalytic reaction process. This invention offers high measurement accuracy, with a minimum detection limit reaching [missing information]. It is superior to traditional spectrometers, has a low cost, can perform one-click automatic measurement without professional operating skills, and can achieve real-time monitoring. In addition, it can be adapted to different systems by changing catalysts and pollutant types. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of a photocatalytic efficiency measurement system based on laser speckle and deep learning.

[0036] Figure 2 This is a flowchart of a photocatalytic efficiency measurement method based on laser speckle and deep learning;

[0037] Figure 3 This is a flowchart of deep learning processing;

[0038] Figure 4 This is a graph showing the results of the validation set.

[0039] Figure 5 These are the spectral wavelengths and absorbance graphs of seven methylene blue solutions at different concentrations;

[0040] Figure 6 This is a comparison chart of the predicted values ​​from the deep learning model and the measured spectral values.

[0041] Figure 7 It is the photocatalytic efficiency curve;

[0042] The components include: a photocatalytic reaction device 1, a light source voltage regulator 11, a quartz cold trap 12, a reaction flask 13, an ultraviolet strip lamp 14, a magnetic stirrer 15, an automated sampling mechanism 2, a sampling tube 22, a piston rod 23, a stepper motor 21, an optical measurement mechanism 3, a laser 31, a variable aperture 32, an imaging lens 33, a CCD camera 34, a sample stage 35, a frosted glass diffuser 36, and a computer device 4. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0045] The formation mechanism of laser speckle originates from the high coherence of laser light. When a laser irradiates the surface of a rough scattering medium, the reflected light or the transmitted light after penetrating the scattering medium exhibits random propagation characteristics, thus generating random phase differences. When these transmitted and scattered lights carrying random phase information meet in the propagation path, constructive or destructive interference occurs, ultimately forming a speckle image on the image plane composed of randomly distributed bright and dark spots. Laser speckle technology exhibits unique advantages in the field of optical measurement due to its non-contact, non-invasive, and rapid imaging characteristics. Laser speckle images are formed through the interference of coherent light with the scattering medium, containing rich physical information. Combined with deep learning technology, high-precision analysis of complex systems can be achieved.

[0046] This application proposes a photocatalytic efficiency measurement method based on laser speckle and deep learning. By constructing a photocatalytic reaction device and an automated sampling mechanism, the degradation solution of the photocatalytic reaction is obtained. The speckle image of the degradation solution to be tested is acquired through the optical measurement mechanism 3. By combining the convolutional neural network (CNN) and random forest (RF) models, the solution concentration during the photocatalytic degradation process can be predicted quickly and accurately. Finally, the photocatalytic efficiency is effectively characterized by the degradation rate.

[0047] like Figure 1 As shown, the photocatalytic efficiency measurement system based on laser speckle and deep learning of this application includes a photocatalytic reaction device 1, an automated sampling mechanism 2, an optical measurement mechanism 3, and a computer device 4. The photocatalytic reaction device 1 provides the photocatalytic degradation environment. The photocatalytic reaction device 1 includes a light source voltage regulator 11 and a light-shielding reaction chamber, as well as a quartz cold trap 12, a reaction bottle 13, an ultraviolet strip lamp 14, and a magnetic stirrer 15 installed inside the light-shielding reaction chamber. The reaction bottle 13 is used to hold the solution to be photocatalytically reacted. The magnetic stirrer 15 is installed below the reaction bottle 13. The magnetic stirrer 15, by controlling the stirring speed (e.g., 1000 rpm), is used to achieve stirring of the solution in the reaction bottle 13, ensuring uniform dispersion of the catalyst in the solution and avoiding local concentration differences from affecting the measurement results. One end of the quartz cold trap 12 is located inside the reaction bottle 13, and the other end of the quartz cold trap 12 is circulated and cooled by an external refrigerator to control the temperature inside the reaction bottle 13 (e.g., controlling the reaction temperature at 25±1℃). A voltage regulator 11 is installed on the top of the light-shielding reaction chamber. An ultraviolet strip lamp 14 is inserted into a transparent quartz cold trap 12. The voltage regulator 11 acts as a control unit, driving the ultraviolet strip lamp 14 (e.g., a 500W mercury lamp light source) to provide stable ultraviolet illumination and achieve photocatalytic reaction.

[0048] An automated sampling mechanism 2 is installed on the photocatalytic reaction device 1 to automatically sample the degradation liquid inside the photocatalytic reaction device 1 in real time. Specifically, addressing the technical challenge of requiring interruption of ultraviolet light irradiation for sampling in traditional photocatalytic experiments, this application designs an automated sampling mechanism that allows sampling at any time without interrupting ultraviolet light irradiation. This automated sampling mechanism includes a sampling cylinder 22, a piston rod 23, and a stepper motor 21. One end of the sampling cylinder 22 is connected to the interior of the reaction flask 13, and a sampling tube is installed on the sampling cylinder 22 outside the light-shielding reaction chamber. The sampling tube can be a 1.5mm round tube. The output shaft of the stepper motor 21 is connected to the piston rod 23. The piston rod 23 is slidably installed inside the sampling tube 22. The stepper motor 21 controls the piston rod 23 to move like a piston inside the sampling tube 22 so that when sampling is needed, the degradation liquid in the reaction bottle 13 can be drawn to the sampling tube. The degradation liquid to be tested is collected at the sampling tube. After sampling, the stepper motor 21 is controlled to push the piston rod 23 close to the reaction bottle 13 and slide it to seal the reaction bottle 13, so as to prevent the degradation liquid from flowing out of the sampling tube again.

[0049] The design of the aforementioned automated sampling mechanism not only avoids the risk of researchers being directly exposed to ultraviolet light, but also enables continuous monitoring of the photocatalytic reaction process.

[0050] The optical measurement mechanism 3 includes a laser 31, an optical modulation component, and an imaging component. The optical modulation component includes a variable aperture 32 and a ground glass diffuser 36, with the test liquid located between the variable aperture 32 and the ground glass diffuser 36. The imaging component includes a 4f imaging element and a CCD camera 34. The 4f imaging element includes a variable aperture 32, an imaging lens 33, another variable aperture 32, and an imaging lens 33 arranged sequentially. The focal length of both the objective lens and the imaging lens is 50 nm, and the system magnification is strictly maintained at 1:1 to ensure accurate reproduction of speckle features. Along the laser emission direction, the laser 31, variable aperture 32, sample stage 35, ground glass diffuser 36, variable aperture 32, imaging lens 33, variable aperture 32, imaging lens 33, and CCD camera 34 are arranged sequentially, with the test liquid placed on the sample stage 35. A beam expander is installed at the laser's emitting end. After the laser emitted by the laser is expanded and collimated by the beam expander, the resulting parallel beam is adjusted by a variable aperture to adjust the beam diameter. After irradiating the liquid to be tested, the beam is phase-modulated by a frosted glass scattering sheet to form a uniform speckle pattern, thereby enhancing the speckle signal. Finally, the speckle image is received by the imaging component.

[0051] In the specific implementation, a 532nm green laser is used as the laser source, with an output power of 300mW and an adjustable aperture of 1-12mm.

[0052] Computer device 4 is used to process speckle images, extract features, and predict the concentration of the degradation solution to obtain photocatalytic efficiency. Computer device 4 includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs actions such as... Figure 2 The steps of the photocatalytic efficiency measurement method are shown.

[0053] The aforementioned photocatalytic efficiency measurement system based on laser speckle and deep learning has several advantages. First, it designs the optical path based on the principle of laser speckle, resulting in lower costs compared to traditional photocatalytic efficiency detection methods. Second, it utilizes deep learning to automatically extract speckle image features and achieves non-contact detection through laser speckle, effectively improving prediction accuracy. Third, the automated sampling mechanism not only avoids the risk of researchers being directly exposed to ultraviolet light but also enables continuous monitoring of the photocatalytic reaction process.

[0054] like Figure 2 As shown, the photocatalytic efficiency measurement method based on laser speckle and deep learning in this application includes the following steps:

[0055] Step S100: Obtain a speckle image of the photocatalyst test solution.

[0056] Step S200: Perform Gaussian filtering, histogram equalization, and ROI cropping on the speckle image.

[0057] Step S300: Use the Inception-v3 model to extract speckle image features, introduce principal component analysis to reduce the dimensionality of speckle image features, combine random forest regression algorithm to construct a concentration prediction model, and train and validate the concentration prediction model.

[0058] In step S400, a concentration prediction model is used to predict the concentration of the speckle image acquired in real time, and the photocatalytic efficiency is calculated according to the degradation rate formula.

[0059] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0060] In sample preparation, methylene blue was selected as the target organic pollutant, and titanium dioxide was used as the semiconductor photocatalyst. Methylene blue mother liquor was serially diluted to prepare samples with concentrations ranging from [specific concentration range to be filled in]. to A series of solutions were used. This concentration range was chosen to cover the typical concentration range of photocatalysis experiments and to effectively test the model's resolution under low concentration conditions.

[0061] For data acquisition, speckle images were acquired using a 5-megapixel CCD camera. Thirty images were acquired for each concentration gradient, resulting in a total of 2190 speckle image samples. To ensure data representativeness and balance, multiple independent measurements were performed on samples of the same concentration during data acquisition to effectively reduce systematic bias.

[0062] according to Figure 1 The photocatalytic efficiency measurement system based on laser speckle and deep learning shown comprises a photocatalytic reaction device 1, an automated sampling mechanism 2, and an optical measurement mechanism 3. After assembly, the photocatalytic reaction device is started, the temperature inside the reaction flask is adjusted to 25°C, and the magnetic stirrer is turned on. The laser speckle optical path is calibrated to ensure the laser beam is collimated and uniformly irradiates the test liquid, and the CCD camera 34 is clearly focused. The photocatalytic efficiency measurement method based on laser speckle and deep learning can then be executed.

[0063] In step S100, the concentration range is configured as follows: to For each concentration of methylene blue series solutions, 30 speckle images were acquired, resulting in a total of 2190 image samples, which were used to establish a standard dataset. The speckle images were acquired using a 5-megapixel CCD camera, with a 532nm green laser as the laser source.

[0064] In step S200, in order to improve the quality of the speckle image and standardize the input to adapt it to the subsequent machine learning model, all input speckle images are sequentially subjected to ROI cropping, Gaussian filtering, histogram equalization and preprocessing.

[0065] In the specific implementation, firstly, all input speckle images are cropped using Region of Interest (ROI) to ensure that each image contains important speckle regions. Next, the cropped speckle images are processed with a Gaussian filter using a kernel size of 5 and a standard deviation of 1.0 to remove noise and smooth the images, making it easier for the deep learning model to capture meaningful patterns. Then, the brightness and contrast of the images are optimized, and histogram equalization is used to enhance image details, making the speckle patterns clearer. All processed images are finally resized to a uniform size of 299×299 and converted to RGB format to ensure consistency in color channels for each image. Finally, the images are normalized, compressing pixel values ​​to the range [0, 1].

[0066] The preprocessing steps described above effectively smooth out subtle noise in the original image. Gaussian filtering removes noise and softens the image texture while preserving its main structural information. Subsequently, histogram equalization significantly enhances the overall contrast of the image by expanding the pixel intensity range. In practice, the preprocessed image is visually sharper, and the main structural and texture information is more prominent due to the improved contrast, providing better contrast input data for subsequent analysis or model training.

[0067] like Figure 3 As shown, in step S300, high-dimensional features are extracted using a deep convolutional neural network, followed by robust regression prediction using a random forest. This hybrid architecture of the deep learning model can capture complex speckle texture features in images while avoiding the risk of overfitting in small sample sizes, and maintaining the interpretability and robustness of traditional deep learning models in regression tasks. Specifically, firstly, a transfer learning strategy is used to extract speckle image features based on the ImageNet pre-trained Inception-v3 model; then, principal component analysis is introduced to reduce the dimensionality of the speckle image features, and a concentration prediction model is constructed by combining it with the random forest regression algorithm; then, the performance of the trained concentration prediction model is evaluated based on the mean absolute error, mean squared error, and coefficient of determination score; finally, five-fold cross-validation is used to verify the stability of the concentration prediction model, obtaining the final concentration prediction model.

[0068] In the specific implementation, the pre-trained Inception-v3 model was used as the backbone network in the feature extraction stage. This model has learned rich general image features on ImageNet and can be directly applied to speckle image analysis through transfer learning. Considering the special characteristics of speckle images, this embodiment did not completely unfreeze all parameters, but adopted a layered fine-tuning strategy: freezing about 70% of the shallow convolutional parameters and training only the high-level feature layers, so that the model can better adapt to the statistical regularity of speckle texture while retaining the general feature expression. Based on Inception-v3, a feature enhancement layer was further designed, including global average pooling and multiple fully connected layers, with Dropout after each layer to reduce the risk of overfitting. In addition, a path network for learning first-order statistical properties was built to enhance the model's sensitivity to concentration changes through feature concatenation. Finally, the high-dimensional features output by the deep network contain both local texture patterns and statistical regularities, providing a solid foundation for subsequent deep learning regression.

[0069] Because the feature dimension generated by combining Inception-v3 with a custom network is extremely high, directly inputting it into a regression model would impose a computational burden and potentially introduce redundant information. Therefore, this embodiment introduces Principal Component Analysis (PCA) for dimensionality reduction. PCA retains the top 100 principal components with the largest variance through linear projection. These principal components can cover most of the key information while compressing the feature space to a processable size. Through visualization analysis of the cumulative variance explained, it is verified that 100-dimensional features are sufficient to retain the main information, improving training efficiency and reducing the risk of overfitting in subsequent models.

[0070] In the regression phase, this embodiment selected the Random Forest model. This model consists of multiple decision trees and improves overall prediction performance through ensemble learning. In this embodiment, the Random Forest consists of 200 decision trees with no limit on tree depth, uses bootstrapping, and leverages multi-core parallelism to accelerate training efficiency. Since speckle concentration prediction tasks often involve noise and uneven sample distribution, RF exhibits significant advantages in such scenarios: firstly, it reduces the high variance of individual decision trees through the ensemble mechanism, thereby improving model stability; secondly, its non-parametric properties allow it to flexibly fit complex nonlinear relationships, enhancing the model's adaptability to different concentration ranges.

[0071] like Figure 4As shown, performance evaluation of the validation set is a crucial means of checking for overfitting and determining the model's generalization ability during model training. This embodiment divides the standard dataset into training and validation sets, using the validation set to evaluate the model's performance at each stage of training. First, several performance evaluation metrics are used during training, including mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) score. For regression tasks, the R² score reflects the model's fit to the data, while MAE and MSE help assess the magnitude of the prediction error. By tracking the training process on the validation set, a slight overfitting phenomenon was observed in the early stages of training; that is, the error on the validation set began to increase, while the error on the training set continued to decrease. To address this issue, an early stopping strategy was adopted during training. Once the performance on the validation set no longer improved, training was terminated early, thus avoiding overfitting. Figure 4 In the figure, (a) the model loss varies with training epochs; (b) the mean absolute error varies with training epochs; and (c) all prediction results on the validation set.

[0072] During model tuning, a five-fold cross-validation method was used to assess the stability of the concentration prediction model. Hierarchical cross-validation ensured that the training and validation data in each tradeoff had similar distributions, avoiding the impact of imbalanced dataset partitioning on model performance. After tuning, the model's performance on the validation set significantly improved, with both MAE and MSE metrics meeting expectations, and an R² score of 0.96, indicating that the model can accurately predict concentration changes. Figure 4 As shown.

[0073] To assess the accuracy of the predicted concentration, seven methylene blue solutions of different concentrations were used as samples. The absorbance of the solution samples was measured using a UV-Vis absorption spectrometer, and the absorption spectrum curves of the seven methylene blue solutions of different concentrations are shown below. Figure 5 As shown in Table 1, the specific concentration values ​​calculated according to Lambert-Beer's Law are as follows:

[0074] Table 1. Absorbance and Spectral Calibration Concentration Data

[0075]

[0076] Using the seven methylene blue solution samples of different concentrations mentioned above as a benchmark, the deep learning model of this application was used to predict the sample concentrations, and the results were compared and analyzed with those obtained by traditional spectrometers. The errors obtained by the two methods were calculated respectively, and the specific results are shown in Table 2 and... Figure 6 As shown:

[0077] Table 2. Predicted Sample Concentration Results and Error Analysis by Deep Learning Model

[0078]

[0079] Table 2 and Figure 6 The data show that the results obtained by the two methods are highly consistent. The root mean square error (RMSE) between the proposed method and the spectrometer measurement results is... The Pearson correlation coefficient was as high as 0.9993 (significant). The results indicate a highly significant linear correlation between the two. The relative errors between the concentration determined by the proposed method and the concentration determined by the spectrometer are controlled at low levels, with a maximum relative error of 7.2% and a minimum of 0.35%. These results fully demonstrate the effectiveness of the prediction method. to It exhibits high accuracy and reliability within a wide concentration range and can be effectively applied to concentration monitoring in photocatalytic processes.

[0080] In step S400, after completing the concentration prediction, this embodiment uses the concentration data obtained by the photocatalytic efficiency measurement method based on laser speckle and deep learning to calculate and plot the curve of photocatalytic efficiency changing with reaction time. The photocatalytic efficiency is characterized by the degradation rate, which is defined as follows:

[0081]

[0082] in, Indicates the initial concentration. This indicates that the reaction has proceeded to... The formula directly reflects the degree of degradation of the methylene blue solution during the photocatalytic reaction. Solution samples are collected at different time points using an automated sampling mechanism. Laser speckle imaging and a deep learning model are used to directly predict the solution concentration, which is then substituted into the degradation rate formula to calculate the degradation rate. Finally, a complete photocatalytic efficiency curve is obtained, as shown below. Figure 7 As shown.

[0083] Uncertainty analysis was performed on the photocatalytic efficiency measurement system based on laser speckle and deep learning in this application. The uncertainty was assessed and calculated using the Monte Carlo method at a confidence level of P=68.3%, with concentration as the metric. For example, the results of six parallel measurements obtained in the experiment are as follows: , , , , , .

[0084] The measurement error of the measurement system in this application mainly originates from model uncertainty, instrument uncertainty, and combined uncertainty. Model uncertainty primarily stems from random fluctuations in experimental data, and is assessed through statistical analysis of the average value of multiple repeated measurements on the validation set. Its calculation uses the experimental standard deviation formula:

[0085]

[0086] in, To measure the number of times, This is a single measurement value. The average value is calculated. Substituting the data from the six measurements above, the model uncertainty is obtained as follows:

[0087]

[0088] Instrument uncertainty primarily assesses the systematic errors introduced by the inherent precision limitations of the experimental instruments. In this experiment, the uncertainty mainly considers the uncertainties introduced by weighing with a balance and measuring volume with a graduated cylinder when preparing standard solutions. Its relative uncertainty is calculated using the following formula:

[0089]

[0090] in, The uncertainty of the measured mass, The uncertainty in measuring volume.

[0091] The following derivation process is to determine the initial concentration. The solution was diluted to the target concentration. For example.

[0092] The weighing uncertainty is calculated using a balance with a permissible error range of 0.0005 g. For a uniformly distributed error, the standard uncertainty is calculated using the following formula:

[0093]

[0094] this The value reflects the absolute uncertainty introduced by a single weighing of the balance. This is relevant when preparing the initial standard solution. At that time, weighed = 0.04g of methylene blue solid. The relative uncertainty introduced by weighing is as follows:

[0095]

[0096] This relative uncertainty will propagate to the final diluted solution concentration. Therefore, the effect of weighing on the final concentration... absolute uncertainty as follows:

[0097]

[0098] this The value reflects the uncertainty introduced by the balance weighing into the final solution concentration.

[0099] Uncertainty in volumetric measurements during dilution, to account for the initial concentration. The solution was diluted to the target concentration. For example, to achieve this dilution ratio ( Using a graduated cylinder with a precision of 1 mL, take out... and fixed to volume Standard uncertainty of a single volume measurement ( The minimum reading half-range of the graduated cylinder is 0.5 mL. The standard uncertainty of a single volume measurement is as follows:

[0100]

[0101] this The value applies to each volume measurement performed using this 1mL precision graduated cylinder.

[0102] For sampling volume ( The relative standard uncertainty of the measurement:

[0103]

[0104] For constant volume ( ):

[0105]

[0106] Concentration after dilution ( The mathematical model for the propagation of relative standard uncertainty and the dilution process is as follows:

[0107]

[0108] The formula for the propagation of the relative uncertainty of the diluted concentration is as follows:

[0109]

[0110] Substituting the above relative uncertainty values, the calculation is as follows:

[0111]

[0112] The relative standard uncertainty introduced by this dilution step is approximately 0.0042. Based on the target concentration... Concentration after dilution ( The absolute standard uncertainty of ) )as follows:

[0113]

[0114] Instrument uncertainty Combining all instrument-related error sources, the final instrument uncertainty is:

[0115]

[0116] Combined uncertainty ( ) is the model uncertainty ( ) and instrument uncertainty ( The combination of ) is calculated using the method of finding the square and the root, as follows:

[0117]

[0118] The combined uncertainty was calculated. The final quantitative analysis results are expressed as the measurement average and its combined standard uncertainty: ,in, To measure the average value, the text in parentheses... The combined standard uncertainty is given. This result indicates that, at a confidence level of P=68.3%, the true value falls within the range of... Within the range.

[0119] Detailed experimental data and uncertainty analysis results for this embodiment are shown in Table 3:

[0120] Table 3. Experimental Data and Uncertainty Analysis Results

[0121]

[0122] Another embodiment of this application provides a photocatalytic efficiency measurement device based on laser speckle and deep learning, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a photocatalytic efficiency measurement method program based on laser speckle and deep learning. When the processor executes the computer program, it implements the steps in the various embodiments of the photocatalytic efficiency measurement method based on laser speckle and deep learning described above. Figure 2 The steps.

[0123] For example, the above-mentioned computer program can be divided into one or more modules / units, which are stored in memory and executed by a processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions. These instruction segments describe the execution process of the computer program in a photocatalytic efficiency measurement device based on laser speckle and deep learning. For example, the computer program can be divided into an image acquisition module, an image processing module, a model building and training module, and a prediction calculation module. The specific functions of each module are as follows:

[0124] The image acquisition module acquires speckle images of the test solution for photocatalysis. The image processing module performs ROI cropping, Gaussian filtering, and histogram equalization on the speckle images. The model building and training module uses the Inception-v3 model to extract features from the speckle images and combines them with a random forest regression algorithm to build and train a concentration prediction model. The prediction calculation module uses the concentration prediction model to predict the concentration of the real-time acquired speckle images and calculates the photocatalytic efficiency based on the degradation rate formula.

[0125] A photocatalytic efficiency measurement device based on laser speckle and deep learning can be a desktop computer, laptop, PDA, or cloud server. This device may include, but is not limited to, processors and memory; for example, it may also include output devices, network access devices, and buses. The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the photocatalytic efficiency measurement device, connecting all parts of the device via various interfaces and lines. The memory can be used to store computer programs and / or modules. The processor implements various functions of the photocatalytic efficiency measurement device by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory.

[0126] If the integrated module / unit of the photocatalytic efficiency measurement device based on laser speckle and deep learning is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-described embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various embodiments of the above-described photocatalytic efficiency measurement method based on laser speckle and deep learning.

[0127] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention. Technical aspects, shapes, and structures not described in detail in this invention are all well-known technologies.

Claims

1. A photocatalytic efficiency measurement method based on laser speckle and deep learning, characterized in that, Includes the following steps: Acquire speckle images of the photocatalyst test liquid, wherein the speckle images are images acquired by placing the photocatalyst test liquid in an optical measurement mechanism; Perform ROI cropping, Gaussian filtering, and histogram equalization on speckle images; The Inception-v3 model was used to extract speckle image features. Principal component analysis was introduced to reduce the dimensionality of the speckle image features. A concentration prediction model was constructed by combining the random forest regression algorithm. The concentration prediction model was then trained and validated. A concentration prediction model was used to predict the concentration of speckle images acquired in real time, and the photocatalytic efficiency was calculated based on the degradation rate formula. The process involves using the Inception-v3 model to extract speckle image features, introducing principal component analysis to reduce the dimensionality of the speckle image features, and combining this with a random forest regression algorithm to construct a concentration prediction model. The concentration prediction model is then trained and validated, including: A transfer learning strategy was adopted to extract speckle image features based on the ImageNet pre-trained Inception-v3 model; Principal component analysis is introduced to reduce the dimensionality of speckle image features; Concentration prediction model is constructed by combining random forest regression algorithm; The performance of the concentration prediction model in training is evaluated based on the mean absolute error, mean square error, and coefficient of determination score. The stability of the concentration prediction model was verified by five-fold cross-validation to obtain the final concentration prediction model.

2. The photocatalytic efficiency measurement method according to claim 1, wherein, The optical measurement mechanism includes a laser, an optical modulation component, and an imaging component. The optical modulation component includes a variable aperture and a ground glass diffuser, with the liquid to be measured located between the variable aperture and the ground glass diffuser. The imaging component includes a 4f imaging element and a CCD camera.

3. The photocatalytic efficiency measurement method according to claim 2, wherein, The laser emitter is equipped with a beam expander. After the laser emitted by the laser is expanded and collimated by the beam expander, the resulting parallel beam is adjusted by a variable aperture to adjust the beam diameter. After irradiating the liquid to be tested, the beam is scattered by a frosted glass scattering sheet to form a speckle field, and the speckle image is received by the imaging component.

4. The photocatalytic efficiency measurement method according to claim 1, wherein, The photocatalytic efficiency is characterized by the degradation rate, and the degradation rate is calculated using the following formula: ； in, Indicates the initial concentration. This indicates that the reaction has proceeded to... The predicted concentration of the solution at a given time.

5. A photocatalytic efficiency measurement device based on laser speckle and deep learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the photocatalytic efficiency measurement method as described in any one of claims 1-4.

6. A photocatalytic efficiency measurement system based on laser speckle and deep learning, characterized in that, include: A photocatalytic reaction device used to provide an environment for photocatalytic degradation; An automated sampling mechanism is installed on the photocatalytic reaction device to automatically sample the degradation liquid in the photocatalytic reaction device in real time. An optical measurement mechanism is used to acquire speckle images of the degradation solution to be tested; A computer device for processing speckle images, extracting features, and predicting the concentration of degradation solution to obtain photocatalytic efficiency; the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the photocatalytic efficiency measurement method as described in any one of claims 1-4.

7. The photocatalytic efficiency measurement system according to claim 6, characterized in that, The photocatalytic reaction device includes a light-shielding reaction chamber, a quartz cold trap, a reaction flask, an ultraviolet strip lamp, and a magnetic stirrer installed inside the light-shielding reaction chamber, and a light source voltage regulator installed on the top outside the light-shielding reaction chamber. The reaction flask is used to hold the solution, and the magnetic stirrer is installed below the reaction flask to stir the solution inside the reaction flask. The light source voltage regulator drives the ultraviolet strip lamp to provide ultraviolet illumination. The quartz cold trap is circulated and cooled by an external refrigerator to control the temperature inside the reaction flask.

8. The photocatalytic efficiency measurement system according to claim 7, characterized in that, The automated sampling mechanism includes a sampling cylinder, a piston rod, and a stepper motor. One end of the sampling cylinder is connected to the interior of the reaction flask, and a sampling tube is provided on the sampling cylinder outside the light-shielding reaction chamber. The output shaft of the stepper motor is connected to the piston rod, and the piston rod is slidably installed inside the sampling cylinder. The stepper motor controls the piston rod to perform piston movement inside the sampling cylinder.

9. The photocatalytic efficiency measurement system according to claim 6, characterized in that, Along the laser emission direction, the optical measurement mechanism includes a laser, a variable aperture, a sample stage, a frosted glass diffuser, a variable aperture, an imaging lens, a variable aperture, an imaging lens, and a CCD camera arranged in sequence, with the liquid to be tested placed on the sample stage.

Images