Method and device for automatically establishing time sequence of ultraviolet radiation photon number

Abstract

The invention provides a method and equipment for automatically establishing an ultraviolet light radiation photon number time sequence, and relates to the field of photon number identification, and the method comprises the steps: carrying out the preprocessing of each sample ultraviolet video, and obtaining a single sample photon number image set; constructing a training image set through the single sample photon number image set; training the discharge photon number identification network through the training image set to obtain a trained discharge photon number identification network; and preprocessing the ultraviolet video to be measured to obtain a single photon number image set to be measured, and inputting the single photon number image set to be measured into the trained discharge photon number recognition network to obtain a photon number time sequence. The trained discharge photon number identification network can realize the automatic identification of the photon number, and forms a photon number time sequence, thereby greatly improving the generation speed and efficiency of the photon number sequence, facilitating the extraction of subsequent characteristic parameters, and improving the accuracy of pollution discharge evaluation.

Description

Technical Field

[0001] The invention relates to the field of photon number recognition, and in particular to a method and device for automatically establishing a time series of ultraviolet radiation photon numbers. Background Art

[0002] Insulators are one of the most widely used high-voltage insulating components in power grids. They are used to fix conductors with potential differences to form electrical isolation. Insulators are exposed to the natural environment all year round, and dirt will inevitably accumulate on their surfaces. In weather conditions such as fog, drizzle, light rain, snowmelt, and condensation, the salt substances in the dirt on the surface will gradually dissolve in the water, causing discharge on the surface. In severe cases, it will cause pollution flashover. In recent years, pollution accidents have occurred on many lines in my country, seriously threatening the safe and stable operation of the power grid.

[0003] In order to prevent insulator flashover from causing grid failures or even large-scale power outages, it is necessary to conduct regular inspections of insulators running on the line. The discharge process is accompanied by discharge phenomena. Discharge detection is an important means to evaluate the operating status of contaminated insulators. Conventional discharge detection methods, such as ultrasonic method and ultra-high frequency method, are difficult to locate the specific discharge position and are easily interfered by the outside world. The discharge process is accompanied by the radiation of ultraviolet light signals, some of which are located in the day-blind ultraviolet band of 240-280nm. In recent years, ultraviolet imaging methods originally used in the military field have gradually been promoted and applied in power grid discharge detection. This method is accurate, intuitive, safe, and less affected by the environment. It has excellent discharge positioning capabilities and detection sensitivity far exceeding conventional methods, and has been widely used in engineering.

[0004] Ultraviolet imagers can record discharge videos. In engineering, the number of photons is often used as a quantitative parameter to characterize its discharge characteristics. The number of photons is displayed in digital form on the instrument screen, but this parameter needs to be read manually. Since the intensity of dirty discharge has a large randomness, the number of photons also changes dynamically over time. It is difficult to fully reflect the dynamic process of its discharge using only a single photon number parameter. It is necessary to read the number of photons over a period of time to form a photon number sequence to better characterize its discharge characteristics. It is necessary to read the number of photons in hundreds or thousands of frames of images in the video stream. At this time, if manual methods are still used to read the number, the efficiency will be very low, the workload will be very large, and it will take a long time to meet the actual needs of the project. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method and device for automatically establishing a time series of ultraviolet radiation photon numbers, which is used to solve the problem that the photon number parameters on instruments in engineering currently need to be read manually, which is very inefficient, the workload is very large, and it takes a long time to meet the actual needs of the project.

[0006] The present invention provides a method for automatically establishing a time series of ultraviolet radiation photon numbers, comprising the steps of:

[0007] S1: Obtain a set of sample ultraviolet videos, preprocess each sample ultraviolet video, and obtain a single sample photon number image set;

[0008] S2: construct a training image set through a single sample photon number image set;

[0009] S3: constructing a discharge photon number recognition network, training the discharge photon number recognition network through a training image set, and obtaining a trained discharge photon number recognition network;

[0010] S4: Obtain the ultraviolet video to be tested, pre-process the ultraviolet video to be tested, obtain a single set of images of photon numbers to be tested, input the single set of images of photon numbers to be tested into a trained discharge photon number recognition network, and obtain a photon number time series.

[0011] Preferably, step S1 specifically comprises:

[0012] S11: extracting frames from the sample ultraviolet video to obtain sample ultraviolet images of each frame;

[0013] S12: cropping the sample ultraviolet image of each frame to obtain a combined sample photon number image of each frame;

[0014] S13: grayscale transformation, binarization and single image digital extraction are performed on the combined sample photon number images of each frame to obtain all single sample photon number images of each frame;

[0015] S14: Repeat steps S11-S13 until all sample ultraviolet videos are traversed to obtain a single sample photon number image set.

[0016] Preferably, step S2 specifically comprises:

[0017] S21: construct a training image subset of digits "0" to "9";

[0018] S22: extracting a preset number of single sample photon number images from the single sample photon number image set and putting them into a training image subset of corresponding numbers;

[0019] S33: compose a training image set from the training image subsets.

[0020] Preferred:

[0021] The discharge photon number recognition network includes an image input layer, a first convolutional network, a second convolutional network, a fully connected layer, a Softmax layer, and a classification layer connected in sequence;

[0022] The first convolutional network and the second convolutional network have the same structure;

[0023] The first convolutional network includes a convolutional layer, a batch normalization layer, an activation function layer, and a pooling layer connected in sequence.

[0024] Preferably, step S3 specifically comprises:

[0025] S31: input the training image into the discharge photon number recognition network, output a single predicted photon number and calculate the loss value;

[0026] S32: Obtain a single real photon number, compare the single predicted photon number with the single real photon number, and calculate the accuracy;

[0027] S33: adjusting parameters of the discharge photon number recognition network through an optimizer;

[0028] S34: Repeat steps S31-S33 until the loss value is less than the first preset value and the accuracy is greater than the second preset value, thereby obtaining a trained discharge photon number recognition network.

[0029] Preferably, step S4 is specifically:

[0030] S41: extracting frames of the ultraviolet video to be tested to obtain ultraviolet images of each frame to be tested;

[0031] S42: performing cropping, grayscale conversion, binarization, and single image number extraction on the ultraviolet image to be measured in the i-th frame in sequence to obtain all single photon number images to be measured in the i-th frame;

[0032] S43: inputting each single photon number image to be measured into a trained discharge photon number recognition network to obtain all single photon numbers of the i-th frame, and sequentially combining each single photon number to obtain a combined photon number of the i-th frame;

[0033] S44: Repeat steps S42-S43 until the combined photon number of each frame is obtained, and sort the combined photon numbers according to the time series to obtain the photon number time series.

[0034] A storage medium stores instructions and data for realizing the method for automatically establishing a time series of ultraviolet radiation photon numbers.

[0035] A device for automatically establishing a time series of ultraviolet radiation photon numbers comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the method for automatically establishing a time series of ultraviolet radiation photon numbers.

[0036] The present invention has the following beneficial effects:

[0037] A discharge photon number recognition network was constructed based on a deep convolutional neural network. The discharge photon number recognition network was trained by using a set of extracted single sample photon number images as a training image set. The trained discharge photon number recognition network can realize automatic recognition of photon numbers and form a photon number time series, which greatly improves the generation speed and efficiency of the photon number sequence, facilitates the subsequent extraction of feature parameters, and improves the accuracy of contaminated discharge assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0039] Figure 2 is the binary image of the photon number;

[0040] Figure 3 is the sub-region of the binary image;

[0041] Figure 4 Structural diagram of the network for identifying the number of discharge photons;

[0042] Figure 5 It is a curve chart of the change of accuracy and loss value;

[0043] Figure 6 The result diagram of the identification confusion matrix for the training set and the validation set;

[0044] Figure 7 It is the photon count value change diagram of the photon number time series;

[0045] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0046] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0047] Reference Figure 1 The present invention provides a method for automatically establishing a time series of ultraviolet radiation photon numbers, comprising the steps of:

[0048] S1: Obtain a set of sample ultraviolet videos, preprocess each sample ultraviolet video, and obtain a single sample photon number image set;

[0049] As an embodiment, the photon count of the UV imager is directly displayed on its image. Taking the CoroCAM504 UV imager commonly used in engineering as an example, its photon count is generally displayed in the lower right corner of the counting box. For example, the photon count in the image is 316.

[0050] The number of photons is actually part of the ultraviolet image and is stored in the calculation in the format of an image. To use a deep learning network for automatic recognition, it is necessary to first establish a digital sample database of the number of photons in the ultraviolet image and input it into the deep network system for training. The number of photons is often a multi-digit number, so before establishing the sample database, it is necessary to segment its individual data, and the segmentation process is shown in step S1.

[0051] Step S1 is specifically as follows:

[0052] S11: extracting frames from the sample ultraviolet video to obtain sample ultraviolet images of each frame;

[0053] Specifically, frame extraction: in order to obtain multiple image sample data, the video file collected by the ultraviolet imager is imported into the computer. The present invention uses video analysis software to continuously extract multiple frames of images from the video. The time interval between adjacent image frames is 40ms, and the size of each original image frame is 784*576 pixels.

[0054] S12: cropping the sample ultraviolet image of each frame to obtain a combined sample photon number image of each frame;

[0055] Specifically, image cropping: the background image outside the discharge area in the ultraviolet image is often complex. In the ultraviolet image, the position of the photon number in the image area is fixed. In order to improve the speed of extracting the image number, this invention crops and extracts the photon number area.

[0056] The analysis of the photon count of a large number of dirty discharge ultraviolet images shows that the photon count varies with the discharge intensity, so the number of photon digits will also change, and the maximum number of digits can reach 4. Therefore, the size of the rectangular area of ​​the image intercepted in the present invention is 64*25 pixels, that is, the width is 64 pixels and the height is 25 pixels.

[0057] S13: grayscale transformation, binarization and single image digital extraction are performed on the combined sample photon number images of each frame to obtain all single sample photon number images of each frame;

[0058] Specifically, grayscale transformation and binarization:

[0059] The cropped image is an RGB color image with three color components. Its color information is not helpful for the training and recognition of the subsequent model, and also increases the amount of calculation. Based on this, the present invention performs a grayscale transformation on the cropped image, and the transformation formula is as follows:

[0060] Y=0.299R+0.587G+0.114B(1)

[0061] In formula (1), Y is the gray value, which ranges from 0 to 255; R, G, and B are the three color component values ​​of red, green, and blue of the original color image. The above algorithm is implemented using the rgb2gray function.

[0062] The grayscale value of each pixel in the grayscale image ranges from (0 to 255). The present invention performs binarization on the grayscale image, converting each pixel into white and black. The algorithm is as follows:

[0063]

[0064] x, y represent the horizontal and vertical coordinates of the pixel in the image respectively; fx, y represent the values ​​of the pixel at the corresponding position. The relevant algorithm is implemented using the imbinarize function in digital image processing, and its threshold is set to 0.55.

[0065] Extraction of single image digits:

[0066] In order to facilitate network training, it is necessary to extract single digits. Since the number of photons is not fixed, such as the photon number 316 is a 3-digit number, and the image is intercepted according to the image width of 4 digits, the area before the number "3" is an area without digits and an invalid area. Therefore, before extracting a single digit, the present invention first identifies the valid digit area.

[0067] Binary images are represented by numbers 0 and 1 in a computer matrix, with black areas displayed as 0 and white areas displayed as 1. Since the areas without numbers are displayed in black except for a few noise points, the resulting binary image is as follows: Figure 2 shown.

[0068] Based on this, the present invention proposes the following effective digital region extraction method.

[0069] Image sub-region division:

[0070] The above 64*25 rectangular area is divided into four image sub-areas with the same width and height. The size of each area is 16*25 pixels. The image sub-areas extracted from the above four areas are as follows: Figure 3 shown.

[0071] Extraction of effective area:

[0072] The number of pixels containing "1" in each area after the above segmentation is counted separately. The number of pixels in the four sub-areas (a)-(d) are 7, 72, 49 and 54 respectively. It is obvious that area (a) has only 7 pixels because there are no numbers but some scattered noise points. The other areas have valid numbers and therefore have a larger number of pixels. The present invention analyzes a large number of 0-9 images and finds that the image of "1" is relatively simple and has the least pixel value, which is about 50. Therefore, the threshold value is taken as 25 in the present invention. It is believed that when the number of pixels in a sub-area is less than 25, the area is an invalid digital area. Figure 3 The sub-region (a) is determined to be an invalid region, and thus the number of valid digits is 3.

[0073] S14: Repeat steps S11-S13 until all sample ultraviolet videos are traversed to obtain a single sample photon number image set.

[0074] S2: construct a training image set through a single sample photon number image set;

[0075] As an embodiment, step S2 is specifically as follows:

[0076] S21: construct a training image subset of digits "0" to "9";

[0077] S22: extracting a preset number of single sample photon number images from the single sample photon number image set and putting them into a training image subset of corresponding numbers;

[0078] S33: compose a training image set from the training image subsets.

[0079] Specifically, the convolutional deep learning network needs to establish the above-mentioned sample database of photon numbers to train the model. In the present invention, ultraviolet video of dirty discharge is collected through experiments, 100 frames of ultraviolet images are captured using video analysis software, and then each digital image is captured separately using the method shown in the above step S1 and stored as an image file, and a total of about 300 independent pictures of photon numbers are obtained.

[0080] To facilitate model training, the present invention creates 10 folders under the UV Phtonumber Samples folder. Each folder is named with the corresponding training number 0-9, which is also the label value of the sample.

[0081] Sample data of 0-9 were randomly selected from the above 300 independent digital images, and sample images of the same number were placed in the same folder. They were then encoded sequentially from the numbers “0” to “9” starting from 1. The number of different digital samples and the encoding range are shown in Table 1.

[0082] Table 1 Sample size distribution and coding of different numbers

[0083]

[0084]

[0085] Based on the above method, a total of 153 groups of sample data were obtained, and each folder stored relevant image data to be trained.

[0086] Taking the image of the number "0" stored in the folder "0" as an example, the 14 sample images are all binary images with a size of 16*25 pixels. The methods for establishing samples of other numbers are similar and will not be repeated here.

[0087] S3: constructing a discharge photon number recognition network, training the discharge photon number recognition network through a training image set, and obtaining a trained discharge photon number recognition network;

[0088] As an example:

[0089] The discharge photon number recognition network includes an image input layer, a first convolutional network, a second convolutional network, a fully connected layer, a Softmax layer, and a classification layer connected in sequence;

[0090] The first convolutional network and the second convolutional network have the same structure;

[0091] The first convolutional network includes a convolutional layer, a batch normalization layer, an activation function layer, and a pooling layer connected in sequence.

[0092] Specifically, the present invention constructs a discharge photon number recognition network based on a deep convolutional neural network. The structure of the discharge photon number recognition network is as follows: Figure 4 shown.

[0093] (1) Image input layer: There are 153 sample image samples in the present invention, 90% of which are randomly selected as training samples and 10% as verification samples. All images are binary images with a size of 25*16 pixels. In the image input layer, the size of the input image is set to 16*25 and the number of channels is set to 1.

[0094] (2) Convolutional layer: The function of the convolutional layer is to extract image features by sliding the convolution kernel in the above image samples. The sizes of convolutional layer 1 and convolutional layer 2 are both set to 3*3 square matrices. Convolutional layer 1 has 8 convolutional kernels and convolutional layer 2 has 16 convolutional kernels.

[0095] (3) Batch Normalization Layer: In order to keep the data in the same distribution during the training process, batch normalization is performed on one or more layers of the middle layer (hidden layer) of the neural network. For each small batch of data (mini-batch), the mean and variance of the batch are calculated. Before the linear calculation result is sent to the activation function, the calculation result is first batch normalized, that is, the mean is subtracted and the standard deviation is divided to ensure that the calculation result conforms to the standard normal distribution with a mean of 0 and a variance of 1. The calculation result is then used as the input value of the activation function for calculation. This can keep the input of the middle layer of the network relatively stable, which helps to solve the problem of gradient disappearance or gradient explosion during the training process.

[0096] (4) Pooling layer: In image processing, since there is a lot of redundant information in the image, the pooling layer 1 and the pooling layer 2 of the present invention both use maximum pooling. The method is to use the maximum value of the statistical information of a certain area sub-block to characterize the spatial distribution pattern of all pixels in the area, so as to replace the values ​​of all pixels in the area sub-block. The extracted image feature information can be reduced in dimension to achieve downsampling, while retaining the main information in the feature map. On the one hand, the feature map is made smaller and the complexity of network calculation is simplified; on the other hand, feature compression is performed to extract the main features, increase translation invariance, and reduce the risk of overfitting.

[0097] (5) Activation function: The activation function layer is also called the non-linearity mapping layer. Its function is to increase the expression and abstraction capabilities of the entire network. The present invention uses the ReLU activation function.

[0098] (6) Fully connected layer: The fully connected layer integrates these local features by connecting to all nodes of the previous neural network to form a global feature that can classify the input data. Due to its fully connected characteristics, the fully connected layer generally has the most parameters. In the present invention, since the output of its model is 10 numbers from 0 to 9, its output is set to 10.

[0099] (7) Softmax layer:

[0100] The Softmax layer is mainly used in deep learning to convert the K-dimensional vector output by the previous layer (usually the output of the fully connected layer) into a probability distribution, that is, to output a K-dimensional probability vector, in which each element represents the probability that the sample belongs to the corresponding category.

[0101] (8) Classification layer: Based on the probability output of softmax, the result with the highest probability is taken as the output of the class, that is, the final output is a specific number between 0 and 9.

[0102] As an embodiment, step S3 is specifically as follows:

[0103] S31: input the training image into the discharge photon number recognition network, output a single predicted photon number and calculate the loss value;

[0104] S32: Obtain a single real photon number, compare the single predicted photon number with the single real photon number, and calculate the accuracy;

[0105] S33: adjusting parameters of the discharge photon number recognition network through an optimizer;

[0106] S34: Repeat steps S31-S33 until the loss value is less than the first preset value and the accuracy is greater than the second preset value, thereby obtaining a trained discharge photon number recognition network.

[0107] Specifically, before training the deep network, it is necessary to set the relevant hyperparameters. The optimizer of the model in the present invention adopts sgdm, and the stochastic gradient descent method with momentum is used to train the network; the initial learning rate 'InitialLearnRate' is set to 0.01; the maximum number of training rounds 'MaxEpochs' is set to 4; 'Shuffle' is set to 'every-epoch', which means that the data is shuffled once in each round of training; and the validation frequency 'ValidationFrequency' is set to 30.

[0108] After completing the above parameter settings, the accuracy and loss value during the training process change with the iteration rounds as shown in the following curves: Figure 5 As shown in the figure, the accuracy increases with the increase of iterative calculation steps. When the iterative calculation reaches 15 steps, the accuracy reaches 100%.

[0109] refer to Figure 6 , the result of the recognition confusion matrix of the training set is Figure 6 (a), the recognition confusion matrix of the validation set is Figure 6 (b). The analysis shows that the model accurately achieves classification and recognition of all samples in the training set and the validation set, with good results.

[0110] S4: Obtain the ultraviolet video to be tested, pre-process the ultraviolet video to be tested, obtain a single set of images of photon numbers to be tested, input the single set of images of photon numbers to be tested into a trained discharge photon number recognition network, and obtain a photon number time series.

[0111] As an embodiment, step S4 is specifically as follows:

[0112] S41: extracting frames of the ultraviolet video to be tested to obtain ultraviolet images of each frame to be tested;

[0113] Specifically, the number of image frames to be identified is preferably set to 100-1000 frames, that is, the length of the discharge video is 4s-40s, which can better reflect the time characteristics of the insulator surface discharge.

[0114] S42: performing cropping, grayscale conversion, binarization, and single image number extraction on the ultraviolet image to be measured in the i-th frame in sequence to obtain all single photon number images to be measured in the i-th frame;

[0115] Specifically, the image frames to be analyzed captured from the ultraviolet video are read sequentially and continuously using the imread function, and then digital image segmentation and effective digital bit extraction are performed according to the method shown in step S1 of the present invention.

[0116] S43: inputting each single photon number image to be measured into a trained discharge photon number recognition network to obtain all single photon numbers of the i-th frame, and sequentially combining each single photon number to obtain a combined photon number of the i-th frame;

[0117] Specifically, the extracted digital image is input into the trained discharge photon number recognition network, and the network model can automatically recognize the numbers on the image as Arabic numerals. Since the recognized numbers are independent 1-digit numbers, it is necessary to combine the different digits and store the results in the array for storing the photon value.

[0118] S44: Repeat steps S42-S43 until the combined photon number of each frame is obtained, and sort the combined photon numbers according to the time series to obtain the photon number time series.

[0119] Specifically, after all the image frames are processed in sequence according to the above steps, the photon numbers on all the images are stored in an array in sequence, and the array is the photon number time series.

[0120] Taking the discharge of a contaminated insulator in the small arc stage as an example, the present invention is used to automatically extract the number of photons in 500 frames of images, and the obtained photon count value change diagram of the photon number time series is as follows: Figure 7 As shown, the entire calculation process takes 7.6675 seconds, which is much shorter than the time required for reading the number of photons by manual methods, greatly reducing labor intensity and improving efficiency.

[0121] The accuracy of comparing the number of photons extracted by the computer with the manual reading results reached 100%. The photon number sequence can effectively reflect the dynamic change characteristics of the discharge and provide relevant parameters for the subsequent operating status evaluation of contaminated insulators based on ultraviolet imaging.

[0122] The invention provides an ultraviolet radiation photon number time series automatic establishment device, a processor and a storage medium.

[0123] A device for automatically establishing a time series of ultraviolet radiation photon numbers: the device for automatically establishing a time series of ultraviolet radiation photon numbers implements the method for automatically establishing a time series of ultraviolet radiation photon numbers.

[0124] Processor: The processor loads and executes the instructions and data in the storage medium to implement the method for automatically establishing the ultraviolet radiation photon number time series.

[0125] Storage medium: The storage medium stores instructions and data; the storage medium is used to implement the method for automatically establishing the ultraviolet radiation photon number time series.

[0126] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0127] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0128] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for automatically establishing a time series of ultraviolet radiation photon numbers, characterized in that: Includes steps: S1: Obtain a set of sample ultraviolet videos, preprocess each sample ultraviolet video, and obtain a single sample photon number image set; S2: construct a training image set through a single sample photon number image set; S3: constructing a discharge photon number recognition network, training the discharge photon number recognition network through a training image set, and obtaining a trained discharge photon number recognition network; S4: Obtain the ultraviolet video to be tested, pre-process the ultraviolet video to be tested, obtain a single set of images of photon numbers to be tested, input the single set of images of photon numbers to be tested into a trained discharge photon number recognition network, and obtain a photon number time series.

2. The method for automatically establishing a time series of ultraviolet radiation photon numbers according to claim 1, characterized in that: Step S1 is specifically as follows: S11: extracting frames from the sample ultraviolet video to obtain sample ultraviolet images of each frame; S12: cropping the sample ultraviolet image of each frame to obtain a combined sample photon number image of each frame; S13: grayscale transformation, binarization and single image digital extraction are performed on the combined sample photon number images of each frame to obtain all single sample photon number images of each frame; S14: Repeat steps S11-S13 until all sample ultraviolet videos are traversed to obtain a single sample photon number image set.

3. The method for automatically establishing a time series of ultraviolet radiation photon numbers according to claim 1, characterized in that: Step S2 is specifically as follows: S21: construct a training image subset of digits "0" to "9"; S22: extracting a preset number of single sample photon number images from the single sample photon number image set and putting them into a training image subset of corresponding numbers; S33: compose a training image set from the training image subsets.

4. The method for automatically establishing a time series of ultraviolet radiation photon numbers according to claim 1, characterized in that: The discharge photon number recognition network includes an image input layer, a first convolutional network, a second convolutional network, a fully connected layer, a Softmax layer, and a classification layer connected in sequence; The first convolutional network and the second convolutional network have the same structure; The first convolutional network includes a convolutional layer, a batch normalization layer, an activation function layer, and a pooling layer connected in sequence.

5. The method for automatically establishing a time series of ultraviolet radiation photon numbers according to claim 1, characterized in that: Step S3 is specifically as follows: S31: input the training image into the discharge photon number recognition network, output a single predicted photon number and calculate the loss value; S32: Obtain a single real photon number, compare the single predicted photon number with the single real photon number, and calculate the accuracy; S33: adjusting parameters of the discharge photon number recognition network through an optimizer; S34: Repeat steps S31-S33 until the loss value is less than the first preset value and the accuracy is greater than the second preset value, thereby obtaining a trained discharge photon number recognition network.

6. The method for automatically establishing a time series of ultraviolet radiation photon numbers according to claim 1, characterized in that: Step S4 is specifically as follows: S41: extracting frames of the ultraviolet video to be tested to obtain ultraviolet images of each frame to be tested; S42: performing cropping, grayscale conversion, binarization, and single image number extraction on the ultraviolet image to be measured in the i-th frame in sequence to obtain all single photon number images to be measured in the i-th frame; S43: inputting each single photon number image to be measured into a trained discharge photon number recognition network to obtain all single photon numbers of the i-th frame, and sequentially combining each single photon number to obtain a combined photon number of the i-th frame; S44: Repeat steps S42-S43 until the combined photon number of each frame is obtained, and sort the combined photon numbers according to the time series to obtain the photon number time series.

7. A storage medium, characterized in that: The storage medium stores instructions and data for implementing the method for automatically establishing a time series of ultraviolet radiation photon numbers as described in any one of claims 1 to 6.

8. An automatic device for establishing a time series of ultraviolet radiation photon numbers, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the method for automatically establishing a time series of ultraviolet radiation photon numbers as described in any one of claims 1 to 6.

Images