Method and device for locating abnormalities in sequenced data based on reagent reactions

Abstract

The present invention discloses a method and device for locating anomalies in serialized data based on reagent reactions. The method comprises: obtaining an original data sequence and converting the original data sequence into a two-dimensional image; inputting the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model; if the prediction result indicates the presence of an anomaly, post-processing the prediction result to obtain the sample type and anomaly location corresponding to the anomaly result; wherein the anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels. This method solves the problem of low anomaly location accuracy in the prior art due to the poor robustness of statistical models, thereby improving the robustness of the location model and further improving the accuracy of anomaly location in serialized data based on reagent reactions.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and device for locating anomalies in serialized data based on reagent reactions. Background Art

[0002] When locating the abnormal position of the serialized data, it can be done with the help of specific reagents. Add multiple reagents to the sample, obtain the sequence after the reagent reaction, and compare the sequence before the reagent reaction with the sequence after the reagent reaction. Figure 1 , curve L1 represents the sample before the reagent reaction, and curve L2 represents the sample after the reagent reaction. If the sample has an abnormality at a certain position, and the sample becomes normal at that position after reacting with the reagent, then this sample is called positive for the reagent, for example Figure 1 The sample shown in the figure is positive for reagents B and D, and the shaded area is marked as the abnormal location.

[0003] Existing techniques typically use statistical methods to detect anomalies based on reagent reactions. During implementation, these methods calculate sample features, such as peak prominence, peak width, and peak area. Anomalies are then detected by setting corresponding thresholds. If the sample features meet the threshold specifications, no anomaly exists; otherwise, an anomaly exists. However, due to the diverse morphologies of samples, this statistically based approach lacks robustness. The same sample feature calculation method and threshold setting may not be effective for a different set of samples.

[0004] In view of this, the present invention provides a method and device for locating anomalies in serialized data based on reagent reactions, so as to solve the problem of low anomaly locating accuracy due to poor robustness of statistical models in the prior art. Summary of the Invention

[0005] To this end, an embodiment of the present invention provides a method and device for locating anomalies in serialized data based on reagent reactions to solve the problem of low anomaly positioning accuracy in the prior art due to poor robustness of statistical models, thereby improving the robustness of the positioning model and further improving the accuracy of anomaly positioning in serialized data based on reagent reactions.

[0006] In order to achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0007] The present invention provides a method for locating anomalies in sequenced data based on reagent reactions, the method comprising:

[0008] Acquire an original data sequence, and convert the original data sequence into a two-dimensional image;

[0009] Inputting the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model;

[0010] If the prediction result indicates that an abnormality exists, post-processing the prediction result to obtain a sample type and an abnormal location corresponding to the abnormal result;

[0011] The anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels.

[0012] In some embodiments, the network architecture of the deep learning network includes:

[0013] The encoder is used to extract features from the image of the input data sample. It can be a convolution or a combination of convolution and transformer. Convolution can fuse local features, and transformer can fuse global features.

[0014] The decoder is used to decode the features encoded by the encoder. Decoding generally uses upsampling methods such as interpolation and deconvolution;

[0015] The feature fusion layer is used to enrich feature information. In the decoding process, not only decoding features but also encoding features are used, which is conducive to improving the effect of semantic segmentation.

[0016] The semantic segmentation head is used to perform semantic segmentation using the fused features. It generally outputs an array of size H×W×C, where H, W, and C represent the height, width, and channel dimensions (number of categories) of the input image.

[0017] In some embodiments, a pre-built deep learning network is trained using images of data samples and corresponding result labels to obtain the anomaly prediction model, specifically including:

[0018] Acquire one-dimensional data of a data sample, and convert the one-dimensional data into a two-dimensional image;

[0019] Construct a dataset using two-dimensional images of data samples and corresponding result labels;

[0020] Dividing the dataset into a training set, a validation set, and a test set;

[0021] Using the training set to train a pre-built deep learning network to obtain an initial model;

[0022] The initial model is evaluated using a validation set to obtain the abnormality prediction model.

[0023] In some embodiments, the initial model is optimized using a binary cross entropy loss function.

[0024] In some embodiments, the initial model is optimized using a Focal loss function.

[0025] In some embodiments, the initial model is optimized using a weighted value of Focal loss and Dice loss as a loss function.

[0026] In some embodiments, the post-processing specifically includes:

[0027] The prediction result is subjected to a sigmoid function operation to obtain a probability value of an abnormality at a certain position;

[0028] If the probability value reaches a preset threshold, the location is an abnormal location.

[0029] The present invention also provides a device for locating anomalies in serialized data based on reagent reactions, the device comprising:

[0030] A data processing unit, configured to obtain an original data sequence and convert the original data sequence into a two-dimensional image;

[0031] an anomaly prediction unit, configured to input the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model;

[0032] an anomaly locating unit, configured to, when the prediction result indicates that an anomaly exists, perform post-processing on the prediction result to obtain a sample type and an anomaly location corresponding to the anomaly result;

[0033] The anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels.

[0034] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0036] The reagent reaction-based anomaly localization method and device for serialized data provided by the present invention obtains a raw data sequence, converts the raw data sequence into a two-dimensional image, and inputs the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model. If the prediction result indicates an anomaly, the prediction result is post-processed to obtain the sample type and anomaly location corresponding to the anomaly result. This solves the problem of low anomaly localization accuracy caused by the poor robustness of statistical models in the prior art, thereby improving the robustness of the localization model and, in turn, the accuracy of reagent reaction-based anomaly localization for serialized data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0038] The structures, proportions, sizes, etc. illustrated in this specification are intended only to complement the contents disclosed herein and to facilitate understanding and reading by persons familiar with the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall still fall within the scope of the technical contents disclosed herein.

[0039] Figure 1 It is a comparison diagram of the sequence before the reagent reaction and the sequence after the reagent reaction;

[0040] Figure 2 This is one of the flow charts of the method for locating anomalies in serialized data based on reagent reactions provided by the present invention;

[0041] Figure 3 This is the second flow chart of the method for locating anomalies in serialized data based on reagent reactions provided by the present invention;

[0042] Figure 4 This is a structural block diagram of the device for locating abnormalities in serialized data based on reagent reactions provided by the present invention;

[0043] Figure 5 This is a structural block diagram of a computer device provided by the present invention. DETAILED DESCRIPTION

[0044] The following describes the implementation of the present invention using specific embodiments. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. Obviously, the embodiments described are only a portion of the present invention, not all of it. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0045] When locating abnormalities in serialized data, there are two situations: 1) The original sample does not have abnormal performance; 2) After adding a certain reagent to the original sample, the sequence before the reagent reaction and the sequence after the reagent reaction are compared, and it is found that the sample has abnormal performance. For situation 1), the method provided by the present invention can identify that the sample does not have abnormal performance on these reagents; for situation 2), the method provided by the present invention determines its positive type by comparing the differences between the sample before the reagent reaction and the sample after the reagent reaction, such as reagent A positive, reagent B positive, etc., and if it is positive on a certain reagent, it can locate the positive position in the sample. In all embodiments provided by the present invention, the sample before the reagent reaction represents the original sample, which will not be repeated.

[0046] In order to solve the problems existing in the prior art, the present invention converts 1D sequence data into 2D images through data processing, and the method provided by the present invention can not only identify the type of positive sample with the help of reagent reaction, but also locate the abnormal position. In the embodiment provided by the present invention, it is set that there are N kinds of reagents that react with the sample, then it can be identified whether the sample is positive on these N reagents, specifically, it can be positive on one or several reagents, and the specific positive position can be identified.

[0047] The following combination Figure 2-Figure 4 The present invention introduces a method and device for locating anomalies in serialized data based on reagent reactions.

[0048] Please refer to Figure 2 , Figure 2 This is one of the flow charts of the method for locating anomalies in serialized data based on reagent reactions provided by the present invention.

[0049] In a specific embodiment, the method for locating anomalies in sequenced data based on reagent reaction provided by the present invention comprises the following steps:

[0050] S210: Obtain an original data sequence, and convert the original data sequence into a two-dimensional image; that is, the obtained original data is one-dimensional sequence data. In order to facilitate subsequent model processing, the one-dimensional sequence data is converted into a two-dimensional image, and the two-dimensional image is used as the input of the model.

[0051] S220: Input the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model; wherein the anomaly prediction model is obtained by training a pre-constructed deep learning network using images of data samples and corresponding result labels; it should be understood that the so-called samples in model training are data samples used as training.

[0052] S230: When the prediction result indicates that an abnormality exists, the prediction result is post-processed to obtain the sample type and abnormal location corresponding to the abnormal result; the prediction result can be of two types, one is that there is no abnormality, in which case there is no need to analyze and locate the abnormal result, and the other is that there is an abnormality, in which case post-processing is required to determine the abnormal sample type and the location of the abnormality.

[0053] In step S230, the post-processing specifically includes:

[0054] The prediction result is subjected to a sigmoid function operation to obtain a probability value of an abnormality at a certain position;

[0055] If the probability value reaches a preset threshold, the location is an abnormal location.

[0056] In a specific usage scenario, first, according to the algorithm input adjustment method, the 1D sequence is converted into an image, and then the trained algorithm is used for calculation, and finally an output of 224×224×N is obtained. The output at this time is a floating-point number, which needs to be post-processed to obtain the positive sample type and the abnormal position. The post-processing method is as follows: 1) The obtained output is subjected to a sigmoid function operation to limit the output range between 0 and 1, indicating the probability of an abnormality at a certain position; 2) A threshold is set, such as 0.5. The position greater than the threshold indicates the abnormal position, and the position less than the threshold indicates the normal position; 3) The result is analyzed. If all output values are less than 0.5, it means that there is no abnormality in the sample. If there is a position with an output greater than 0.5 on some channels, the sample has a positive performance on the reagent represented by this channel, and the positive position is the position where the output is greater than 0.5.

[0057] The artificial intelligence algorithm used in the present invention can be a semantic segmentation algorithm Unet, TransUnet, etc. In some embodiments, the network architecture of the deep learning network includes:

[0058] The encoder is used to extract features from the input data sample image. The encoder is a key component in deep learning models, and its primary function is to extract useful features from the input data, which can be used for subsequent tasks such as segmentation. The encoder extracts features through convolution or a combination of convolution and transformers. Convolution is used to fuse local image features, while transformers are used to fuse global image features. In other words, the encoder can be a convolutional or a combination of convolution and transformers, with convolution fusing local features and transformers fusing global features.

[0059] The decoder is used to decode the features encoded by the encoder. The decoder is another key component in deep learning models and typically works alongside the encoder. Its primary function is to use the features extracted by the encoder to generate predictions through multiple calculations. Decoding typically uses upsampling methods such as interpolation and deconvolution.

[0060] The feature fusion layer is used to fuse encoded and decoded features. It combines features from different levels to create a richer, more comprehensive feature representation. For example, in image processing, fusing low-level texture features with high-level semantic features can help the model better understand image content. This feature fusion layer enriches feature information. During the decoding process, it uses both decoded and encoded features, which helps improve semantic segmentation.

[0061] The semantic segmentation head is used to perform semantic segmentation using the fused features. It generally outputs an array of size H×W×C, where H, W, and C represent the height, width, and channel dimensions (number of categories) of the input image. The semantic segmentation head is the part of the deep learning model specifically used for semantic segmentation tasks. In semantic segmentation, the purpose of the model is to assign a category label to each pixel in the image, thereby achieving semantic segmentation of the image. The semantic segmentation head usually receives the output from the feature fusion layer and generates a segmentation map with pixel-level classification.

[0062] The method provided by the present invention converts the problem of locating anomalies in serialized data into a semantic segmentation problem. The algorithm used is not limited to a specific semantic segmentation method, and can be any semantic segmentation algorithm, such as the pure convolution algorithm Unet, the algorithm combining convolution and transformer TransUnet, etc. The network architecture mainly includes an encoder, a decoder, a feature fusion layer, and a semantic segmentation head. The encoder is mainly used for feature extraction and can be a convolution or a combination of convolution and transformer. Convolution can fuse local features, and transformer can fuse global features. The decoder mainly decodes the encoded features. Decoding generally uses upsampling methods such as interpolation and deconvolution. The feature fusion layer is used to enrich feature information. During the decoding process, not only decoded features but also encoded features are used, which is conducive to improving the effect of semantic segmentation. The semantic segmentation head generally outputs an array of size H×W×C, where H, W, and C represent the height, width, and channel dimensions (number of categories) of the input.

[0063] In some embodiments, the pre-built deep learning network is trained using the images of the data samples and the corresponding result labels to obtain the abnormality prediction model, such as Figure 3 As shown, the specific steps include:

[0064] S310: Obtain one-dimensional data of the data sample and convert the one-dimensional data into a two-dimensional image; in a specific usage scenario, for each reagent, convert the reagent pre-reaction sequence ( Figure 1 The middle curve L1) and the sequence after the reagent reaction ( Figure 1 The middle curve L2 is plotted on the same graph. The area between the two sequences is filled with white, and the rest of the image is filled with black. The image size is set to 224×224. Since there are N reagents that react with the sample, the above image can be obtained after reacting with each reagent. All images are spliced together on the channel to obtain a sample input of size 224×224×N.

[0065] S320: Construct a data set using the two-dimensional image of the data sample and the corresponding result label; when labeling the sample, there are two labeling methods. One method is to label which reagent the sample is positive for, for example Figure 1 Reagents B and D are marked as positive in the sample. Another mark is to locate the position where a certain reagent is positive, which is represented by an interval. For example, reagent B is added at this time and reagent B is positive. The marked interval is [a0, a1], which means that in the sample, reagent B is positive in the interval from a0 to a1.

[0066] Furthermore, the above annotations need to be converted into a format that can adapt to the algorithm. The dimension of the annotation is the same as the input, which is also 224×224×N. First, the annotation is initialized to 0, and then the positive position is updated to 1. For example, if the annotated sample is positive under reagent D, and the positive position is [b0, b1], since the annotated channel dimension represents N reagents, first locate the channel where reagent D is located, between b0 and b1, and update the area between the sample before the reagent reaction and the sample after the reagent reaction to 1. It should be noted that if the sample is negative under N reagents, then the annotation does not need to be updated.

[0067] S330: Divide the dataset into a training set, a validation set, and a test set in a ratio of 6:3:1 or 8:1:1.

[0068] S340: Using the training set to train a pre-built deep learning network to obtain an initial model;

[0069] S350: Evaluate the initial model using a validation set to obtain the abnormality prediction model.

[0070] Since a sample can be positive on one or more reagents, this method is a multi-classification task. In the process of optimizing the initial model, a binary cross entropy loss function can be used, as shown in formula (1):

[0071]

[0072] Among them, p t Represents the predicted probability, p represents the probability of the algorithm predicting the positive class, y represents whether it is marked as foreground, 1 represents the foreground, CE(p t ) represents the binary cross entropy loss function, log(p t ) represents taking the logarithm of the predicted probability.

[0073] Considering that the area occupied by the abnormal region is relatively small and some abnormal locations are difficult to detect, the formula (1) is expanded and two parameters α are introduced. t ,γ,α t is the positive and negative sample imbalance parameter, γ is the sample difficulty parameter, and the improved Focal loss function is shown in the following formula (2):

[0074] FL(p t )= -α t (1-p t ) γ log (p t ) (2)

[0075] Among them, α tIt can suppress the impact caused by the excessive proportion of normal areas, and γ can increase the optimization of samples that are difficult to detect in abnormal locations.

[0076] The present invention can also use the optimization target Dice loss function, and its calculation formula is shown in the following formula (3):

[0077]

[0078] Among them, P represents the mask predicted by the algorithm, and Y represents the real labeled mask.

[0079] During the training process, Focal loss and Dice loss are combined in different proportions. The proportion can be adjusted according to the actual situation to obtain the final optimization target. Then, the optimizer is used for backpropagation to continuously update the algorithm parameters until convergence.

[0080] In the above-mentioned specific embodiment, the method for locating anomalies in serialized data based on reagent reactions provided by the present invention obtains the original data sequence, converts the original data sequence into a two-dimensional image, and inputs the two-dimensional image into a pre-trained anomaly prediction model to obtain the prediction result output by the anomaly prediction model; if the prediction result indicates an anomaly, the prediction result is post-processed to obtain the sample type and anomaly location corresponding to the anomaly result. This solves the problem of low anomaly location accuracy in the prior art due to the poor robustness of the statistical model, thereby improving the robustness of the location model and further improving the accuracy of anomaly location in serialized data of reagent reactions.

[0081] In addition to the above method, the present invention also provides a device for locating abnormalities in serialized data based on reagent reaction, such as Figure 4 As shown, the device includes:

[0082] The data processing unit 410 is configured to obtain an original data sequence and convert the original data sequence into a two-dimensional image;

[0083] An anomaly prediction unit 420 is used to input the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model;

[0084] Anomaly locating unit 430, configured to, when the prediction result indicates that an anomaly exists, perform post-processing on the prediction result to obtain a sample type and an anomaly location corresponding to the anomaly result;

[0085] The anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels.

[0086] In some embodiments, the network architecture of the deep learning network includes:

[0087] The encoder is used to extract features from the image of the input data sample. It can be a convolution or a combination of convolution and transformer. Convolution can fuse local features, and transformer can fuse global features.

[0088] The decoder is used to decode the features encoded by the encoder. Decoding generally uses upsampling methods such as interpolation and deconvolution;

[0089] The feature fusion layer is used to enrich feature information. In the decoding process, not only decoding features but also encoding features are used, which is conducive to improving the effect of semantic segmentation.

[0090] The semantic segmentation head is used to perform semantic segmentation using the fused features. It generally outputs an array of size H×W×C, where H, W, and C represent the height, width, and channel dimensions (number of categories) of the input image.

[0091] In some embodiments, a pre-built deep learning network is trained using images of data samples and corresponding result labels to obtain the anomaly prediction model, specifically including:

[0092] Acquire one-dimensional data of a data sample, and convert the one-dimensional data into a two-dimensional image;

[0093] Construct a dataset using two-dimensional images of data samples and corresponding result labels;

[0094] Dividing the dataset into a training set, a validation set, and a test set;

[0095] Using the training set to train a pre-built deep learning network to obtain an initial model;

[0096] The initial model is evaluated using a validation set to obtain the abnormality prediction model.

[0097] In some embodiments, the initial model is optimized using a binary cross entropy loss function.

[0098] In some embodiments, the initial model is optimized using a Focal loss function.

[0099] In some embodiments, the initial model is optimized using a weighted value of Focal loss and Dice loss as a loss function.

[0100] In some embodiments, the post-processing specifically includes:

[0101] The prediction result is subjected to a sigmoid function operation to obtain a probability value of an abnormality at a certain position;

[0102] If the probability value reaches a preset threshold, the location is an abnormal location.

[0103] In the above-mentioned specific embodiment, the reagent reaction-based serialized data anomaly locating device provided by the present invention obtains the original data sequence, converts the original data sequence into a two-dimensional image, and inputs the two-dimensional image into a pre-trained anomaly prediction model to obtain the prediction result output by the anomaly prediction model; if the prediction result indicates the presence of an anomaly, the prediction result is post-processed to obtain the sample type and anomaly location corresponding to the anomaly result. This solves the problem of low anomaly locating accuracy due to the poor robustness of the statistical model in the prior art, thereby improving the robustness of the locating model and further improving the accuracy of the reagent reaction-based serialized data anomaly locating device.

[0104] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment are implemented.

[0105] Those skilled in the art will understand that Figure 5 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0106] Corresponding to the above embodiment, an embodiment of the present invention further provides a computer storage medium, which contains one or more program instructions, wherein the one or more program instructions are used to execute the above method.

[0107] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the above method.

[0108] In the embodiments of the present invention, the processor may be an integrated circuit chip having signal processing capabilities. The processor may 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.

[0109] The methods, steps, and logic diagrams disclosed in the embodiments of the present invention can be implemented or executed. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules within the decoding processor. The software modules can be located in a storage medium well-established in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The processor reads the information from the storage medium and, in conjunction with its hardware, completes the steps of the aforementioned methods.

[0110] The storage medium may be a memory and may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0111] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0112] The volatile memory may be random access memory (RAM), which is used as an external cache memory. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DRRAM).

[0113] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0114] Those skilled in the art will appreciate that in one or more of the above examples, the functions described herein can be implemented using a combination of hardware and software. When software is used, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media includes any medium that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0115] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for locating anomalies in serialized data based on reagent reaction, characterized in that: The method comprises: Acquire an original data sequence, and convert the original data sequence into a two-dimensional image; Inputting the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model; If the prediction result indicates that an abnormality exists, post-processing the prediction result to obtain a sample type and an abnormal location corresponding to the abnormal result; The anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels; The post-processing specifically includes: The prediction result is subjected to a sigmoid function operation to obtain a probability value of an abnormality at a certain position; If the probability value reaches a preset threshold, the location is an abnormal location; In the post-processing process, the obtained output is subjected to a sigmoid function operation, and the output range is limited to between 0 and 1, indicating the probability of an abnormality at a certain position; a threshold is set, and positions greater than the threshold represent abnormal positions, and positions less than the threshold represent normal positions; the result is analyzed, if all output values are less than the threshold, it means that there is no abnormality in the sample, if there is a position where the output is greater than the threshold on some channels, then the sample has a positive performance on the reagent represented by this channel, and the positive position is the position where the output is greater than the threshold.

2. The method for locating anomalies in serialized data based on reagent reaction according to claim 1, characterized in that: The network architecture of a deep learning network includes: An encoder is used to extract features from the image of the input data sample. The encoder performs feature extraction through convolution or a combination of convolution and transformer, wherein convolution is used to fuse local features of the image and transformer is used to fuse global features of the image; The decoder is used to decode the features encoded by the encoder through an upsampling method; Feature fusion layer; The semantic segmentation head is used to perform semantic segmentation using the fused features. The semantic segmentation head outputs an array of size H×W×C, where H, W, and C represent the height, width, and channel dimensions of the input image, respectively.

3. The method for locating anomalies in serialized data based on reagent reaction according to claim 2, characterized in that: The pre-built deep learning network is trained using the images of the data samples and the corresponding result labels to obtain the anomaly prediction model, which specifically includes: Acquire one-dimensional data of a data sample, and convert the one-dimensional data into a two-dimensional image; Construct a dataset using two-dimensional images of data samples and corresponding result labels; Dividing the dataset into a training set, a validation set, and a test set; Using the training set to train a pre-built deep learning network to obtain an initial model; The initial model is evaluated using a validation set to obtain the abnormality prediction model.

4. The method for locating anomalies in serialized data based on reagent reaction according to claim 3, characterized in that: The initial model is optimized using a binary cross entropy loss function.

5. The method for locating anomalies in serialized data based on reagent reaction according to claim 3, characterized in that: The initial model is optimized using the Focalloss loss function.

6. The method for locating anomalies in serialized data based on reagent reaction according to claim 3, characterized in that: The initial model is optimized using a weighted value of Focalloss loss and Dice loss as a loss function.

7. A device for locating abnormalities in serialized data based on reagent reaction, characterized in that: The device comprises: A data processing unit, configured to obtain an original data sequence and convert the original data sequence into a two-dimensional image; an anomaly prediction unit, configured to input the two-dimensional image into a pre-trained anomaly prediction model to obtain a prediction result output by the anomaly prediction model; an anomaly locating unit, configured to, when the prediction result indicates that an anomaly exists, perform post-processing on the prediction result to obtain a sample type and an anomaly location corresponding to the anomaly result; The anomaly prediction model is obtained by training a pre-built deep learning network using images of data samples and corresponding result labels.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images