CN118111980A - Rapid quantitative detection method based on smart phone and machine learning - Google Patents

Abstract

The invention provides a rapid quantitative detection method based on smart phones and machine learning, which comprises the following steps: preparing a sample, detecting by a color-changing sensor, shooting by a smart phone, analyzing by YOLOv algorithm, constructing a standard curve, and detecting an unknown sample. The method combines a visual color-changing sensor detection system and a smart phone WeChat applet, and utilizes the image acquisition function and a machine learning algorithm of the smart phone to perform data analysis and target identification, thereby realizing quick and accurate quantitative detection on site. The detection method can be used in various fields of food safety, medical diagnosis, environmental monitoring and the like, and has important application prospect and practical value.

Description

A rapid quantitative detection method based on smartphone and machine learning

Technical Field

The invention relates to a color change quantitative detection method, and in particular to a rapid quantitative detection method based on the dual auxiliary functions of a smart phone and machine learning.

Background technique

In the fields of food safety, medical diagnosis and environmental monitoring, rapid and efficient on-site detection methods are of great significance. Traditional detection methods often require expensive equipment and professional laboratory personnel, which limits their promotion in resource-scarce areas and convenient personal applications. At the same time, the accuracy of some non-quantitative detection methods also has certain limitations. Therefore, the development of a rapid quantitative detection method based on smartphones and machine learning has important application value. In traditional spectral change detection methods, expensive fluorescence spectrophotometers are often required to analyze the analytes. This method has high accuracy, but the detection process is complicated and requires professional laboratory personnel and expensive experimental equipment. Non-quantitative detection methods, such as pH test paper and qualitative test strips, although simple and easy to use, can only provide vague test results and cannot perform accurate quantitative analysis.

Among the existing patents, the patent number CN202210581500.3 proposes a dual-signal immunoassay method for bisphenol A integrated with a smartphone platform. Although this device combines quantitative detection with smartphones to a certain extent, the detection accuracy is greatly limited because the data pre-processing needs to be done manually, and the detection process is relatively long. The patent number CN202210364606.8 proposes a ratio fluorescence sensor based on a smartphone and its preparation method and application, but this method can only use existing color recognition software on the market instead of customized ones. Therefore, manual processing is required for post-processing and analysis of the detection data, which greatly limits the convenience and speed of the detection.

Summary of the invention

The purpose of the present invention is to provide a rapid quantitative detection method based on a smart phone and machine learning to address the problems existing in the prior art.

The present invention provides a rapid quantitative detection method based on a smartphone and machine learning, which mainly includes color change sensing detection, a smartphone WeChat applet application and a machine learning algorithm. This method combines the advantages of spectral change detection and non-quantitative detection, and uses the portability and ease of use of smartphones to achieve the goal of rapid on-site quantitative detection. The specific detection method includes the following steps:

Prepare the sample: load the corresponding visual color change sensing system into the sample area, and then add the sample to be detected into the sample area;

Color change sensor detection: shine a UV light source or a sunlight light source onto the sample and observe the color change of the sample;

Smartphone photography: Use the smartphone camera to take pictures of the color changes of the sample and upload the pictures to the WeChat applet on the phone (the name of the applet is: F Intelligent Fluorescence Analysis);

YOLOv3 algorithm analysis: The machine learning algorithm YOLOv3 is used in the WeChat applet to analyze the photos and identify the target area in the sample;

Constructing a standard curve: Compare the color parameters of the target area with samples of known concentration to establish a standard functional relationship between color and concentration, i.e., a standard curve;

Unknown sample detection: Take a photo of the unknown sample and upload it to the WeChat applet. Perform quantitative detection using the standard curve to obtain the concentration of the substance to be tested.

The visual color change sensing system can be a daylight colorimetric probe, a fluorescence colorimetric probe (including carbon dots), a phosphorescence colorimetric probe, etc.

The sample may be in the form of a test tube sample, a cuvette sample, a test paper sample or a microfluidic chip sample.

The method of the present invention has the following advantages:

Efficient and fast: Through smartphones and machine learning algorithms, on-site rapid quantitative detection is achieved, which greatly shortens the detection time and improves work efficiency.

Accurate and reliable: By constructing a standard curve, we can perform accurate quantitative detection of unknown samples, improving the accuracy and reliability of detection.

Portable and easy to use: This method uses a smartphone as an analysis device, which is portable and easy to use, and can be easily applied in various scenarios.

Wide application: This method can be used in fields such as food safety, medical diagnosis and environmental monitoring. It has wide application value and can meet the needs of rapid quantitative detection in different fields.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a schematic diagram of the detection method;

Figure 2 is a schematic diagram of the YOLOv3 machine learning algorithm;

FIG3 is a schematic diagram of the homepage of the WeChat applet designed independently by the smartphone in the present detection method;

In the figure: (1) take a photo, (2) select a photo, (3) start analysis, (4) query history records, (5) user.

Figure 4 is a schematic diagram of the main function page of the WeChat applet;

In the figure: (6) fitting the standard curve, (7) customizing the standard curve, and (8) calculating and predicting unknown samples.

Figure 5 is a logic diagram for using WeChat Mini Programs;

In the figure: (9) home page, (10) main function page, (11) YOLOv3 definition interface, (12) parameter selection interface, (13) standard curve definition interface, (14) custom standard curve interface, (15) prediction calculation input interface, (16) standard curve selection interface, (17) prediction result interface.

FIG6 is a mercury ion standard sample of Example 1;

FIG. 7 is a mercury ion test sample of Example 1;

FIG8 is a glutathione standard sample of Example 2;

FIG. 9 is a glutathione test sample of Example 2.

Detailed ways

The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

A rapid quantitative detection method based on smartphone and machine learning. The detection process is shown in Figure 1. The specific steps include:

Prepare samples: Since this method relies on the correspondence between color parameters and concentrations to achieve quantitative detection, it is necessary to select a sensor probe with color-changing behavior, that is, the sensor probe is required to change the color of the sample when the respondent is added, and a sensor probe with obvious color change is preferred. Then, the sensor probe is loaded on the sample area to complete the sample preparation. The sample can be in the form of a test tube, a cuvette, a test paper, or a microfluidic chip.

Color change sensor detection: shine a UV light source or a sunlight light source onto the sample and observe the color change of the sample;

Smartphone photography analysis: Use the camera of a smartphone to photograph the color changes of the sample. When analyzing, first search for "F Intelligent Fluorescence Analysis" on the WeChat applet page. After opening it, enter the sample photo on the homepage, as shown in Figure 3. It is required that there are at least two samples in the picture. The input picture can be taken in real time (1 in Figure 3) or stored in a file (2 in Figure 3). After entering the picture, click "analysis" (3 in Figure 3) to enter the next page, as shown in Figure 4, and then click "fitting" (6 in Figure 4) to enter the fitting process. On the fitting page (11 in Figure 5), enter the concentrations corresponding to the samples with known concentrations in sequence, separated by commas, in units of μM. Then name this fitting scheme, which must not be repeated with the stored name. Click "confirm" on this page to confirm and enter the next page.

YOLOv3 algorithm analysis: The WeChat applet is equipped with the YOLOv3 machine learning algorithm to accurately identify and capture photos of reaction areas in different observation fields. The principle is shown in Figure 2. The specific algorithm code can be found at: https://github.com/ultralytics/yolov3. After the code operation, the picture input on the homepage can accurately identify and separate the target colorimetric area and give the six chromaticity values of R, G, B, H, S, and V of the area. All three values are between 0 and 255. After this step, the color of the discolored area of the picture is accurately quantified.

Constructing a standard curve: After the WeChat applet quantifies the color, it can construct a standard curve by combining the concentration of each sample input by the user. Since different color sensors change color in different ways, the applet is equipped with a variety of color parameter representation methods to fit the function. In the fitting curve, the parameter is the ordinate and the concentration is the abscissa. The specific parameters include: R, G, B, R/G, R/B, G/B, (R+G)/B, (R+B)/G, (B+G)R, H, S, V, H/S, H/V, S/V, (H+S)/V, (H+V)/S, (S+V)H. Considering the demand for different fitting algorithms, the applet is also equipped with a variety of algorithms to select the appropriate fitting algorithm, including: Least Squares, Ridge, Lasso, Lasso Lars, Bayesian Ridge, SVM. All of the above parameters can be selected by the user on the parameter selection page (12 in Figure 5). After the selection is completed, click "confirm" on this page to obtain the fitting curve image and function equation (13 in Figure 5) according to the parameters, and store the fitting data locally in real time. The fitting data is named after the fitting scheme entered by the user this time. In addition, the WeChat applet also provides custom fitting. When the user does not have a picture of the standard sample, the custom function can be used to manually enter the functional correspondence between the concentration and the color parameters. The specific method is to click "self-defined" (7 in Figure 4) to enter the custom interface (14 in Figure 5), enter the functional relationship and the selected parameters and name the function scheme, and click "confirm" on this page to store.

Detection of unknown samples: Take a photo of the sample of unknown concentration and input it into the applet, click "calculate" (8 in Figure 4) to enter the concentration prediction calculation process (15 in Figure 5), input the number of samples, click "confirm" on this page to enter the function selection page (16 in Figure 5), select the function and parameters stored through the fitting page or custom page, then click "confirm" on this page to read the sample concentration on the result page (17 in Figure 5). The multiple numbers in the result represent the concentration of each sample in the input image in turn, in μM.

Example 1

In the field of food safety, smartphones and machine learning-assisted quantitative detection methods are used for rapid detection. Taking the detection of mercury ion residues in agricultural products as an example, a color-changing sensor detection system based on carbon dots is constructed. The color changes on the surface of agricultural products are captured by the camera of a smartphone, and machine learning algorithms are used for analysis and quantitative detection.

Specific steps are as follows:

Prepare samples: Place the agricultural product sample to be tested in a cuvette loaded with a carbon dot-based mercury ion fluorescent colorimetric probe.

Color change sensor detection: shine a UV light source onto the sample and observe the color change of the sample. The presence of the analyte will cause the color change.

Smartphone photography: Use the smartphone camera to capture the color change of the sample (Figure 6) and upload the photo to the WeChat applet.

YOLOv3 algorithm analysis: Use machine learning algorithms (such as YOLOv3) in the WeChat applet to analyze photos and identify target areas in the samples.

Construct a standard curve: Compare the color parameters of the identification area with samples of known concentration to establish a standard functional relationship between color and concentration, i.e., a standard curve.

Unknown sample detection: The cuvettes containing two unknown samples were photographed and uploaded to the WeChat applet (Figure 7). Quantitative detection was performed using the standard curve to obtain the concentration of mercury ions in the samples.

Through this method, we can quickly and accurately detect mercury ion residues in agricultural products on site, avoiding the need for complex experimental procedures and expensive experimental equipment in traditional methods, and improving the convenience and feasibility of detection.

Example 2

In the field of medical diagnosis, we can use smartphones and machine learning-assisted quantitative detection methods for rapid testing. Taking glutathione analysis as an example, we can design a smartphone-based glutathione carbon dot visualization detection system to quantitatively determine glutathione in samples through color-changing sensors and machine learning algorithms.

Specific steps are as follows:

Prepare samples: Collect a patient's body fluid sample that may contain glutathione and place it in a cuvette loaded with a carbon-dot-based glutathione fluorescent colorimetric probe.

Color-changing sensor detection: react the sample with the carbon dot fluorescence sensor and observe the color change under UV light. The presence of different components will cause the color to change.

Smartphone photography: Use the smartphone camera to capture the color changes of the urine sample (Figure 8) and upload the photo to the WeChat applet.

YOLOv3 algorithm analysis: Use machine learning algorithms (such as YOLOv3) in the WeChat applet to analyze the photos and identify the target areas in the sample, that is, the areas with color changes.

Construct a standard curve: Compare the color parameters of the identification area with samples of known concentration to establish a standard functional relationship between color and concentration, i.e., the standard curve.

Unknown sample detection: The cuvettes with two unknown samples were photographed and uploaded to the WeChat applet (Figure 9). Quantitative detection was performed using the standard curve to obtain the concentrations of glutathione components in the samples.

Through this method, we can quickly and accurately measure the concentration of various components in the sample at the medical diagnosis site, providing doctors with timely diagnostic basis. At the same time, it avoids the need for complex experimental procedures and expensive experimental equipment in traditional methods, and improves the convenience and feasibility of detection.

Claims (4)

1. A rapid quantitative detection method based on smart phones and machine learning, characterized in that it includes the following steps: (1) Sample preparation: Load the visual color change sensing system in the sample area, and then add the sample to be detected into the sample area; (2) Color change sensor detection: shine a UV light source or a sunlight light source onto the sample and observe the color change of the sample; (3) Smartphone photography analysis: Use the smartphone camera to take pictures of the color changes of the sample and upload the pictures to the WeChat applet on the phone; (4) YOLOv3 algorithm analysis: The machine learning algorithm YOLOv3 is used in the WeChat applet to analyze the photos and identify the target area in the sample; (5) Constructing a standard curve: Compare the color parameters of the target area with samples of known concentration to establish a standard functional relationship between color and concentration, i.e., a standard curve; (6) Unknown sample detection: Take a photo of the unknown sample and upload it to the WeChat applet. Perform quantitative detection using the standard curve to obtain the concentration of the substance to be tested. 2. The quantitative detection method according to claim 1, characterized in that the sample is in the form of a test tube sample, a cuvette sample, a test paper sample or a microfluidic chip sample. 3. The quantitative detection method as described in claim 1 is characterized in that the WeChat applet is the F intelligent fluorescence analysis applet. 4. The quantitative detection method according to claim 1, characterized in that the visual color change sensing system is a daylight colorimetric probe, a fluorescence colorimetric probe or a phosphorescence colorimetric probe.