A method and system for identifying a detection result of a detection cartridge

Abstract

The application provides a method and system for identifying detection results of a detection box, and the method comprises the following steps: using a target detection method to detect a reagent box picture, detecting regions where reference lines and detection lines are located, and confidence of each detection type of the regions, and taking a type with the maximum confidence as a detection type of the reagent box; and calculating specific values of the reference lines and the detection lines in the reagent box picture according to the confidence of the detection type; the target detection method is a BP neural network or a convolutional neural network. Advantages of the application are as follows: the application can perform detection without the cooperation of a background card; the application uses detection confidence to calculate color values of existing reference lines and detection lines, so that the color values are obtained through a reasonable calculation parameter between experience values and detection values, the values of the reference lines and the detection lines are more accurate and objective, and the detection results have a strong reference significance.

Description

Technical Field

[0001] The present invention belongs to the technical field of image detection, and specifically relates to a method and system for identifying and detecting the detection results of a detection box. Background Art

[0002] The kit is the main carrier of the detection reagents in the current market. For example, kits such as the new coronavirus antigen detection kit and pregnancy test stick, and the reagent strip is fixed in the kit to provide the only detection result identifier for detection.

[0003] The COVID-19 Neutralizing Ab Test product of the heyer company is a kit for new coronavirus antigen detection. Different from other products on the market, this product is connected and interacted with the cloud through a mobile phone APP to manage the antigen detection results of users.

[0004] There are two lines on the kit, namely the C line and the T line. Among them, in the detection result, if the C line shows red, it means that this reagent strip can be used normally, and if it does not show color, it means that this reagent strip cannot be used normally; the depth of the color of the T line represents the diversity of the detection results, such as weak positive, medium, strong positive and other types. However, in the actual detection result test process, the value of the color of the C line only represents whether the reagent strip can be used normally, and there is no special specific meaning at the present stage. However, the value of the color of the T line represents the quantification of specific types such as weak positive, medium, and strong positive. There are many types of COVID-19 viruses, such as the Delta virus and the Omicron virus. For these two viruses alone, their toxicity is different. The Delta virus has strong toxicity, and the Omicron virus has weak toxicity. The specific meaning of the value is that the thresholds of weak positive, medium, strong positive and other types of the virus with strong toxicity are higher than those of the virus with weak toxicity. For example, the quantification value of the Delta virus is weak positive (0 < x < 59), medium (60 < x < 149), strong positive (150 < x < 249), and the quantification value of the Omicron virus is weak positive (0 < x < 39), medium (40 < x < 119), strong positive (120 < x < 249). The same color on the reagent strip represents different infection degrees for two different viruses, especially it is easy to produce misjudgment near the threshold of the type.

[0005] The technical solution of the heyer company is that first, there must be a background card. There are four positioning two-dimensional codes in the upper left, upper right, lower left, and lower right of the background card. Through these four two-dimensional codes, position calibration is performed. There is a standard sample picture of the detection box in the middle black frame. The size of the produced detection box is the same as that of the black frame. Only when the detection box coincides with the position of the black frame can the next step of detection be carried out;

[0006] The specific method is as follows:

[0007] (1) Place the reagent strip to be identified in the background card with positioning QR codes in the four corners, and align the position of the reagent strip with the standard sample diagram of the detection box on the background card;

[0008] (2) Use a mobile phone to take a picture of the reagent strip to be identified and load the picture into memory;

[0009] (3) Cut the QR code and the C and T value regions to be recognized in the image according to the preset position coordinates in the program;

[0010] (4) Calculate the string value of the QR code;

[0011] (5) Based on the traditional computer vision method for calculating color values ​​(image feature color moment calculation), the C and T values ​​are roughly calculated.

[0012] Existing recognition technologies rely on the positions of the C and T lines on the test kit, which means they are not compatible with the recognition results of many test strip manufacturers. This is because each manufacturer's test kit molds are different, and the positions of the C and T lines are different, so test strips from other manufacturers cannot be recognized.

[0013] Existing recognition technologies use traditional computer vision methods to mechanically identify the positions of C-line and T-line, and then calculate the color values ​​of C-line and T-line according to fixed methods. This method is strongly affected by external factors such as ambient light and the mobile phone used for taking the picture, which may cause the results to have a large error compared with the actual results.

[0014] Existing technologies clearly suffer from incompatibility with various sizes of test strips from numerous manufacturers. For example, a strip might be compatible with test strip A but not with strips B and C. To achieve even greater compatibility, more cutting coordinates need to be preset in the program, significantly compromising program robustness and increasing operational costs. Furthermore, existing technologies rely on traditional computer vision methods to roughly calculate color values ​​(C and T), resulting in inaccurate calculations for boundary values. This could lead to miscalculations at the program level, causing false positives in COVID-19 testing and negatively impacting the company's reputation. Summary of the Invention

[0015] The purpose of this invention is to overcome the shortcomings of existing technologies, which can only identify one type of reagent kit and must rely on background cards for identification, resulting in inaccurate identification results.

[0016] To achieve the above objectives, the present invention proposes a method for identifying the detection results of a detection box, the method comprising:

[0017] The target detection method is used on the kit image to detect the area where the baseline and detection line are located, as well as the confidence level of each detection type in that area. The type with the highest confidence level is taken as the detection type of the kit.

[0018] Calculate the specific values ​​of the baseline and test line in the kit image based on the confidence level of the test type;

[0019] The target detection method is a BP neural network or a convolutional neural network.

[0020] As an improvement to the above method, the target detection method includes a sampling layer and a fully connected layer, and the specific steps include:

[0021] Step B1: Obtain a feature map of the same size as the preset input image;

[0022] Step B2: Extract the features to be identified from the feature map using the sampling layer;

[0023] Step B3: Obtain several prediction boxes for selecting the target to be detected;

[0024] Step B4: The fully connected layer calculates and extracts target object information from the prediction box;

[0025] Step B5: Calculate the predicted target bounding box, target coordinates, and confidence level for each detection type.

[0026] As an improvement to the above method, methods for obtaining prediction boxes include:

[0027] The image is divided into several N*N grids, and for each grid, M bounding boxes (i.e., prediction boxes) are predicted to be located near that grid.

[0028] As an improvement to the above method, predicting M rectangular boxes located near the grid is specifically as follows:

[0029] If the target object is in the top left corner of the grid, the prediction box will take the two grids to the left and above.

[0030] If the target object is in the upper right corner of the grid, the prediction box will take the two grids to the right and above.

[0031] If the target object is in the lower left corner of the grid, the prediction box will take the two grids to the left and below.

[0032] If the target object is in the bottom right corner of the grid, the prediction box will capture the two grids to the right and below.

[0033] As an improvement to the above method, the formula for calculating the confidence level is as follows:

[0034]

[0035] Among them, Con value This represents the calculated confidence level; num m This represents the number of predicted bounding boxes around the detected target; α represents the significance level.

[0036] loss(z,x,y) represents the confidence loss value, and the calculation formula is as follows:

[0037] loss(z,x,y)=-L(z,x,y)*logP(z,x,y)-(1-L(z,x,y))*log(1-P(z,x,y))

[0038] In the formula: 0≤z≤3; 0≤x≤Width; 0≤y≤Height

[0039] Where z represents the number of channels in the image; x represents the x-coordinate of the center point of the prediction box; y represents the y-coordinate of the center point of the prediction box; Width represents the width of the image; and Height represents the height of the image.

[0040] L(z,x,y) represents the matrix corresponding to the confidence label, that is, the initial confidence matrix of each prediction box; the matrix contains the channel number, x-coordinate, and y-coordinate of each prediction box and the initial confidence, and this matrix is ​​a one-dimensional matrix;

[0041] P(z,x,y) represents the confidence matrix to be predicted; that is, the reference confidence matrix of each prediction box calculated by the neural network. The matrix contains the channel number, x-coordinate, y-coordinate and predicted reference confidence of each prediction box. This matrix is ​​a one-dimensional matrix.

[0042] As an improvement to the above method, the value of α is preferably 0.1.

[0043] As an improvement to the above method, the values ​​of the baseline and the detection line are calculated as follows:

[0044] C value =Value start +Con value *Value random

[0045] T value =Value start +abs(atan(100*Con value -50)*2 / π)*(Value end -Value start )

[0046] Among them, C value Indicates the calculated value of the baseline; T valueValue represents the calculated value of the detection line. start Value represents the empirical starting value in testing practice. end This represents the empirical termination value in testing practice; Con value This represents the calculated category confidence value; atan() indicates atan normalization; abs() indicates taking the absolute value; Value random To represent a random value, the calculation formula is as follows:

[0047] Value random =Value low +rand()%(Value) high -Value low +1)

[0048] Here, rand() represents a function that generates random numbers; Value low Value represents the minimum value in the range of generated random numbers; high This represents the maximum value within the range of generated random numbers.

[0049] This application also provides a system for identifying detection results of a detection box, implemented based on the above method, the system comprising:

[0050] The target detection module is used to apply target detection methods to the kit image to detect the regions containing the baseline and detection lines, as well as the confidence level of each detection type in that region. The type with the highest confidence level is selected as the detection type for the kit.

[0051] The baseline and test line calculation module is used to calculate the specific values ​​of the baseline and test lines in the kit image based on the confidence level of the test type.

[0052] Compared with the prior art, the advantages of the present invention are:

[0053] 1. This invention can perform detection without the need for a background card. It uses a target detection method to identify the area to be identified, including the detection frames for the QR code, C-line, and T-line, and then cuts out these detection frames. This identification method does not depend on the engraving position of the manufacturer's detection box and can identify C-line and T-line at any position in the detection box;

[0054] 2. This invention uses detection confidence level to calculate the color values ​​of existing C-line and T-line. In this way, the color values ​​obtained by using a reasonable calculation parameter between the empirical value and the detection value make the values ​​of C-line and T-line more accurate and objective, and have a strong reference value for the detection results.

[0055] 3. This invention uses artificial intelligence, target detection and classification methods to obtain the target positions of the C-line and T-line to be identified, making the identification results more objective and accurate. Attached Figure Description

[0056] Figure 1 The diagram shows the flowchart of the method for identifying the detection results of the detection box;

[0057] Figure 2 The diagram shown is a schematic of the target detection method;

[0058] Figure 3 The following is a flowchart of the target detection method;

[0059] Figure 4 The image shown is an example of calculating the confidence score of a predicted bounding box in an object detection method.

[0060] Figure 5 The diagram shows the prediction bounding box when the target point is in the upper left corner of the grid in the target detection method.

[0061] Figure 6 The diagram shows the prediction box when the target point is in the upper right corner of the grid in the target detection method.

[0062] Figure 7 The diagram shows the prediction bounding box when the target point is in the lower left corner of the grid in the target detection method.

[0063] Figure 8 The diagram shows the prediction box being drawn when the target point is in the lower right corner of the grid in the target detection method. Detailed Implementation

[0064] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0065] Example 1 of this application:

[0066] like Figure 1 As shown, the process of a method for identifying detection results of a detection box according to this application includes:

[0067] Step 1: Visually inspect the integrity of the test kit. In this application, it is not necessary to deliberately place the test kit on the background card or in a black frame that matches the background card. The test kit can be placed anywhere. During this process, it is only necessary to maintain the integrity of the test kit.

[0068] Step 2: Use a mobile phone or camera to take a picture of the test kit. During this process, try to keep the position of the test kit upright. The test kit is allowed to have a certain tilt angle (<10°), but try to avoid large tilt angles such as 45° or 50°.

[0069] Step 3: Load the image and program into the computer memory and wait for detection;

[0070] Step 4: Perform object detection on the image to detect the regions where the baseline and detection lines to be identified and calculated are located, as well as the confidence level of each detection type, and obtain the detection type of the baseline and detection lines;

[0071] Taking a COVID-19 test kit as an example, the baseline is the C line, and the test line is the T line. The system detects the regions containing the C and T values, and simultaneously determines the test type for each region. Currently, there are seven test types: cptpvl (None or Very Low negative), cptpl (Low weak positive), cptp (Medium moderate positive), cptph (High strong positive), cptpvh (Very High extremely strong positive), cntp (C Error, baseline error), and cntn (C and T Error, baseline and test line error). The system also obtains the confidence level for each of these seven test types.

[0072] In mathematics, confidence level is the probability that the true value will appear within a certain range centered on the measured value. For example... Figure 2 As shown, in this invention, the confidence level is defined as the credibility of the recognition type appearing in the target detection box (the value range is 0-100%). For example, if a detection result is of type cptn, and there is a 76% probability that it is type cptn, 76% is the maximum type detected by the detection program.

[0073] Object detection is a very important branch of artificial intelligence. The main task of object detection is to identify and detect specific objects, and at the same time provide the category and location information of the object. It is not necessary to provide the specific location of the object during the detection process.

[0074] like Figure 3 As shown, the specific process for target detection in this application is as follows:

[0075] Step A1: Input the image to be detected;

[0076] Step A2: Obtain a feature map of the same size as the preset input image;

[0077] The image feature map is the product of pixel values ​​being processed by a filter (also called a convolution kernel). In other words, it's the content generated after an image passes through a convolution kernel in the convolutional layer of a neural network; this operation is known as convolution. The method used in this application to obtain the feature map is the SIFT operator feature extraction method.

[0078] Step A3: Use the sampling layer in the neural network to extract the features to be identified from the feature map above (these features are obtained through training, and the training process will not be described here);

[0079] Step A4: Obtain bounding boxes of several anchored targets through a neural network for the selection of the target to be detected;

[0080] Step A5: Calculate and extract target object information from the anchor box using the fully connected layer in the neural network;

[0081] Step A6: Obtain the predicted target bounding box, target type, and target coordinates through neural network inference;

[0082] Step A7: Calculate the confidence level for each detection type of the target object;

[0083] Step A8: Sort the predicted targets to be identified according to their confidence level, and obtain the most likely target to be identified.

[0084] The neural network in the above steps can be a BP neural network or a convolutional neural network, etc.

[0085] In object detection, for a target image, the image can be divided into several 80*80 grids. The detection process predicts three bounding boxes (BODs) for each grid cell. Each BOD is included with its center coordinates, width, height, and confidence level. Therefore, the prediction program outputs a total of 3*80*80 BODs with confidence levels ranging from 0% to 100%, corresponding one-to-one with the 3*80*80 BODs. For example... Figure 4 In the diagram, points A, B, C, and D represent the target objects to be detected. In this case, the prediction confidence of the three cells containing each point is relatively large or close to 1, while the prediction confidence of other cells will be very small or even close to 0.

[0086] To accelerate training and detection, the coordinates of the 80*80 target bounding box are defined as (x...). g ,y g ),like Figure 4 The coordinates of the four random points A, B, C, and D shown are (x0, y0). At this point, it is necessary to determine the coordinates of the points within the grid (x0, y0). g ,y g To calculate a new point that is not in the center of the grid, we take one neighboring grid cell to the left, right, top, and bottom of the target cell: (x1, y0) and (x0, y1). This is done to prevent overfitting during training and detection. The specific method is as follows:

[0087] like Figure 5 As shown, if point (x0, y0) is in the upper left corner of the grid, then the two grids to the left and above are taken;

[0088] like Figure 6 As shown, if point (x0, y0) is in the upper right corner of the grid, then the two grids to the right and above are taken;

[0089] like Figure 7 As shown, if point (x0, y0) is in the lower left corner of the grid, then the two grids to the left and below are taken;

[0090] like Figure 8 As shown, if point (x0, y0) is in the bottom right corner of the grid, then the two grids to the right and below are taken;

[0091] Based on the point definition rules mentioned above, the calculation formulas for x1 and y1 can be obtained as follows:

[0092] x1=x0+(2*round(x g -x0)-1)

[0093] y1=y0+(2*round(y g -y0)-1)

[0094] The round() function performs rounding operations.

[0095] The loss function used by the neural network to calculate the confidence score is as follows:

[0096] loss(z,x,y)=-L(z,x,y)*logP(z,x,y)-(1-L(z,x,y))*log(1-P(z,x,y))

[0097] In the formula: 0≤z≤3; 0≤x≤Width; 0≤y≤Height

[0098] Where z represents that an image has three channels; x represents the x-coordinate of the center point of the prediction box; and y represents the y-coordinate of the center point of the prediction box.

[0099] loss(z,x,y) represents the confidence loss value that needs to be calculated;

[0100] L(z,x,y) represents the matrix corresponding to the confidence labels, i.e., the initial confidence matrix for each (3*80*80) prediction box. The matrix contains the channel number, x-coordinate, and y-coordinate of each prediction box, and is a one-dimensional matrix. Due to the computational characteristics of neural networks, an initial value is required, so this confidence value needs to be initialized to 1.

[0101] P(z,x,y) represents the confidence matrix to be predicted; that is, the reference confidence matrix of each (3*80*80) prediction box calculated by the neural network. The matrix contains the channel number, x-coordinate, y-coordinate and predicted reference confidence of each prediction box. This matrix is ​​a one-dimensional matrix.

[0102] Width represents the width of the image; Height represents the height of the image.

[0103] The formula for calculating the confidence level is as follows:

[0104]

[0105] Among them, Con value This indicates the confidence level value that needs to be calculated;

[0106] num m This indicates the number of predicted bounding boxes around the detected target;

[0107] `loss(z,x,y)` represents the confidence loss value, which requires calculating the loss value of the predicted bounding boxes around all detected targets. The confidence loss value is calculated using a confidence loss function. This loss function primarily measures the deviation between the model's predicted value and the true value; the loss value is a quantification of this deviation. Using the loss function calculation formula, the confidence loss value of the predicted bounding box containing all points in the image can be calculated, which is the deviation between the predicted confidence and the true confidence.

[0108] α represents the significance level, which is 0.1 in this project. The significance level is the probability of making an error when the estimated population parameter falls within a certain interval. For this target detection task, the empirical significance level is taken as 0.1.

[0109] Based on the concept of confidence level, this formula can be used to first calculate the product of the average confidence level of potential object targets and the significance level, which is the final calculated confidence level value used to quantify the accuracy of the detected object targets.

[0110] Step 5: Based on the calculated category confidence, select the category with the highest confidence, and then calculate the final values ​​of the baseline and detection line in the image based on this.

[0111] Baseline calculation value:

[0112] Calculation principle: For the baseline value to be generated, since the baseline value is the value to verify whether the test strip is normal, if the baseline of the test strip is detected to be normal in actual testing, a clinical experience value can be used as the starting value, and the product of the calculated category confidence and the random value is added to obtain the calculated baseline value as a reference.

[0113] Baseline value calculation formula:

[0114] C value =Value start +Con value *Valuerand om

[0115] Among them, C value Value represents the calculated baseline value obtained through calculation; start Indicates the empirical starting value in clinical practice for testing; Con value This represents the category confidence value calculated during the detection process;

[0116] Value random Represents a random value, used to calculate a value at the Value. low ~Value high To obtain a random value within a given range, generate a random value, then take the remainder of this random value divided by the range step size, and add it to the lowest value within the range. This ensures that the random value will 100% fall within this range. The calculation formula is as follows:

[0117] Value random =Value low +rand()%(Value) high -Value low +1)

[0118] Here, rand() represents a function that generates random numbers; Value low Value represents the minimum value in the range of generated random numbers; high This represents the maximum value within the range of generated random numbers.

[0119] Calculated values ​​for the detection line:

[0120] Calculation principle: By calculating the absolute value of the normalized value of the category confidence score, multiplying it by the difference from the empirical starting value, and finally adding the result to the empirical starting value, a reference value for the detection line can be obtained, which lies between the empirical starting value and the termination value. The calculation formula is as follows:

[0121] T value =Value start +abs(atan(100*Con va l ue -50)*2 / π)*(Value end -Value start )

[0122] Among them, T value This represents the calculated detection line value; `abs()` represents taking the absolute value of the calculated confidence level; `atan()` represents performing atan normalization on the confidence level value to improve the reliability and accuracy of the results; Con value Value represents the category confidence value calculated during the detection process.start This represents the empirical starting value in clinical practice for testing; Value end This indicates the empirical termination value in clinical practice of the test.

[0123] Embodiment 2 of this application:

[0124] This application also provides a system for identifying detection results of a detection box, implemented based on the above method, the system comprising:

[0125] The target detection module is used to apply target detection methods to the kit image to detect the regions containing the baseline and detection lines, as well as the confidence level of each detection type in that region. The type with the highest confidence level is selected as the detection type for the kit.

[0126] The baseline and test line calculation module is used to calculate the specific values ​​of the baseline and test lines in the kit image based on the confidence level of the test type.

[0127] The present invention may also provide a computer device, comprising: at least one processor, a memory, at least one network interface, and a user interface. The various components of the device are coupled together via a bus system. It is understood that the bus system is used to enable communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0128] The user interface can include a display, keyboard, or clicking device. Examples include a mouse, trackball, touchpad, or touchscreen.

[0129] It is understood that the memory in the embodiments disclosed in this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate Synchronous DRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memory.

[0130] In some implementations, the memory stores elements such as executable modules or data structures, or subsets thereof, or extended sets thereof: operating systems and applications.

[0131] The operating system includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application programs include various applications, such as media players and browsers, used to implement various application functions. Programs implementing the methods of the embodiments of this disclosure can be included in the application programs.

[0132] In the above embodiments, the processor can also invoke programs or instructions stored in memory, specifically programs or instructions stored in an application program, for the following purposes:

[0133] Follow the steps described above.

[0134] The above methods can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by software instructions. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the disclosed methods, steps, and logic block diagrams. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the disclosed methods can be directly implemented by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0135] It is understood that the embodiments described in this invention can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0136] For software implementation, the technology of this invention can be achieved by executing the functional modules (e.g., procedures, functions, etc.) of this invention. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or externally.

[0137] The present invention may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, it can implement the steps in the above method embodiments.

[0138] Finally, it should be noted that 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 the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying the detection result of a detection box, the method comprising: The target detection method is used on the kit image to detect the area where the baseline and detection line are located, as well as the confidence level of each detection type in that area. The type with the highest confidence level is taken as the detection type of the kit. Calculate the specific values ​​of the baseline and test line in the kit image based on the confidence level of the test type; The target detection method is a BP neural network or a convolutional neural network; The formula for calculating the confidence level is as follows: ； in, This represents the calculated confidence level value; This indicates the number of predicted bounding boxes around the detected target; Indicates the significance level; The confidence loss value is expressed by the following formula: ； Where z represents the number of channels in the image; x represents the x-coordinate of the center point of the prediction box; and y represents the y-coordinate of the center point of the prediction box. This represents the width of the image. This indicates the high value corresponding to the image; This represents the matrix corresponding to the confidence label, i.e., the initial confidence matrix for each prediction box; the matrix contains the channel number, x-coordinate, and y-coordinate of each prediction box, and is a one-dimensional matrix. This represents the confidence matrix to be predicted; that is, the reference confidence matrix for each prediction box calculated by the neural network. The matrix contains the channel number, x-coordinate, y-coordinate and predicted reference confidence of each prediction box. This matrix is ​​a one-dimensional matrix. The calculation methods for the baseline and test line values ​​are as follows: ； ； in, This represents the calculated value of the baseline. This represents the calculated value of the detection line; This represents the empirical starting value in testing practice; This represents the empirical termination value in testing practice; This represents the calculated category confidence value; This indicates that atan has undergone normalization. Indicates taking the absolute value; To represent a random value, the calculation formula is as follows: ； in, A function that generates random numbers; This represents the minimum value within the range of generated random numbers; This represents the maximum value within the range of generated random numbers.

2. The method for identifying the detection result of the detection box according to claim 1, characterized in that, The target detection method includes a sampling layer and a fully connected layer, and the specific steps include: Step B1: Obtain a feature map of the same size as the preset input image; Step B2: Extract the features to be identified from the feature map using the sampling layer; Step B3: Obtain several prediction boxes for selecting the target to be detected; Step B4: The fully connected layer calculates and extracts target object information from the prediction box; Step B5: Calculate the predicted target bounding box, target coordinates, and confidence level for each detection type.

3. The method for identifying the detection result of the detection box according to claim 2, characterized in that, Methods for obtaining prediction boxes include: The image is divided into several N*N grids, and for each grid, M bounding boxes (i.e., prediction boxes) are predicted to be located near that grid.

4. The method for identifying the detection result of the detection box according to claim 3, characterized in that, The specific prediction of M rectangles located near the grid is as follows: If the target object is in the top left corner of the grid, the prediction box will take the two grids to the left and above. If the target object is in the upper right corner of the grid, the prediction box will take the two grids to the right and above. If the target object is in the lower left corner of the grid, the prediction box will take the two grids to the left and below. If the target object is in the bottom right corner of the grid, the prediction box will capture the two grids to the right and below.

5. The method for identifying the detection result of the detection box according to claim 1, characterized in that, The preferred value is 0.

1.

6. A system for identifying detection box detection results, implemented based on the method described in any one of claims 1-5, characterized in that, The system includes: The target detection module is used to apply target detection methods to the kit image to detect the regions containing the baseline and detection lines, as well as the confidence level of each detection type in that region. The type with the highest confidence level is selected as the detection type for the kit. The baseline and test line calculation module is used to calculate the specific values ​​of the baseline and test lines in the kit image based on the confidence level of the test type.

Images