Monitoring data identification method and device, medium and equipment

By building a monitoring data identification model for dense connection blocks and multi-scale feature extraction networks, the problems of low efficiency and insufficient accuracy in monitoring data identification are solved, efficient and accurate monitoring data identification is achieved, and the efficiency and accuracy of medical monitoring are improved. sex.

CN118887705AInactive Publication Date: 2024-11-01XIAN KEWEI RAILWAY INTELLIGENT TECH CO LTD
View PDF 16 Cites 0 Cited by

Patent Information

Application Number
CN202411280930.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-11-01
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art has problems of low recognition efficiency and insufficient accuracy in automatic identification of monitoring data, especially when processing image data with diversity and noise, it is difficult to meet the real-time and accuracy requirements of medical monitoring.

Method used

The monitoring data identification model with dense connection blocks is constructed, and feature multiplexing is achieved by superimposing convolution layers and jump connections of different convolution kernels, and combining multi-scale feature extraction networks, multi-head attention mechanisms, feature enhancement and denoising modules, The capsule network layer is used to perform image preprocessing and model training to improve recognition effect.

Benefits of technology

It realizes efficient and accurate identification of key data from the monitoring input image, reduces manual workload, improves the accuracy and efficiency of monitoring data processing, and avoids errors caused by manual transcription.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118887705A_ABST
    Figure CN118887705A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of image processing, and particularly discloses a monitoring data identification method and device, a medium and equipment, and the method comprises the steps: collecting a monitoring input image; the monitoring input image is preprocessed; constructing a monitoring data identification model and performing training; and inputting the preprocessed monitoring input image into a trained monitoring data identification model to identify monitoring data in the monitoring input image. According to the invention, the monitoring data can be accurately and efficiently identified from the monitoring image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of image processing technology, and specifically relates to a monitoring data recognition method, device, medium and equipment. Background Art

[0002] In the existing technology, automatic recognition of monitoring data faces many challenges, especially when extracting and identifying these key data from images efficiently and accurately. Due to the diversity of monitoring data (such as heart rate, blood pressure, blood oxygen saturation, etc.), the differences in image sources (the format, resolution and clarity of the output images of monitors of different brands and models are different), and the possible noise and interference factors in the image (such as light changes, occlusion, stains, etc.), traditional methods often cannot achieve ideal recognition results.

[0003] Specifically, traditional image recognition technology often relies on manually designed feature extractors, which have relatively weak generalization and robustness for complex and changeable monitoring input images. In addition, due to the extremely high real-time and accuracy requirements of monitoring data, traditional methods are also difficult to meet the needs of clinical applications in terms of processing speed and accuracy. Summary of the invention

[0004] In view of the deficiencies in the prior art, the purpose of this application is to provide a monitoring data recognition method, device, medium and equipment, which can automatically and accurately identify key monitoring data from monitoring images, thereby improving the efficiency and accuracy of medical monitoring.

[0005] To achieve the above objectives, this application provides the following technical solutions:

[0006] A monitoring data recognition method, the method comprising: collecting a monitoring input image; preprocessing the monitoring input image; constructing a monitoring data recognition model and training it; wherein the monitoring data recognition model introduces a dense connection block, the dense connection block comprises superimposed convolutional layers with different convolutional kernels, and the superimposed convolutional layers with different convolutional kernels use jump connections to achieve feature reuse; the preprocessed monitoring input image is input into the trained monitoring data recognition model to recognize the monitoring data in the monitoring input image.

[0007] Optionally, the input image is preprocessed, including: adjusting the monitoring input image to a uniform size; performing image enhancement on the monitoring input image after the uniform size; and normalizing the monitoring input image after the image enhancement.

[0008] Optionally, the monitoring data recognition model is trained through the following steps: collecting a large number of monitoring input images to form a monitoring input image set; using different boxes to mark each data item in each monitoring input image; dividing the marked input images into a training set and a verification set in proportion; setting training parameters, training the model through the training set, calculating the cross entropy loss function of the model during the training process, and using the Adam optimization algorithm to optimize the loss function until the loss function converges; verifying the model with the verification set, and during the verification process, using accuracy, recall rate and F1 score as evaluation indicators to evaluate the performance of the model. When all indicators reach the preset values, the model verification is passed; otherwise, adjusting the training parameters or expanding the training samples to re-train the model until the model verification is passed.

[0009] The present application also provides a monitoring data recognition device, which includes: an acquisition module for acquiring monitoring input images; a preprocessing module for preprocessing the monitoring input images; a model construction and training module for constructing a monitoring data recognition model and training it; wherein the monitoring data recognition model introduces a densely connected block, and the densely connected block includes superimposed convolutional layers with different convolutional kernels, and the superimposed convolutional layers with different convolutional kernels use jump connections to achieve feature reuse; a recognition module is used to input the preprocessed monitoring input image into the trained monitoring data recognition model to identify the monitoring data in the input image.

[0010] Optionally, the preprocessing module includes: a size adjustment submodule, used to adjust the monitoring input image to a uniform size; an enhancement submodule, used to perform image enhancement on the monitoring input image after the uniform size; and a normalization submodule, used to normalize the monitoring input image after image enhancement.

[0011] The present application also provides a monitoring data identification system, which includes the device as described above.

[0012] The present application also provides a storage medium, which includes instructions, and when the instructions are executed on a computer, the computer executes the method as described in any of the preceding items.

[0013] The present application also provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above methods when executing the program.

[0014] Compared with the prior art, the beneficial effects brought by the present application are as follows: by constructing a monitoring data recognition model, the present application can automatically and accurately and efficiently identify and output various monitoring data from the monitoring input image, which greatly reduces the manual workload, and can avoid the errors caused by manual transcription of data, thereby improving work efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a monitoring data identification method provided by one embodiment of the present application;

[0016] Figure 2 A schematic diagram of the structure of a monitoring data recognition model provided by another embodiment of the present application;

[0017] Figure 3 A schematic diagram of the structure of a multi-scale feature extraction network provided in another embodiment of the present application;

[0018] Figure 4 A schematic diagram of the structure of a densely connected block provided in another embodiment of the present application;

[0019] Figure 5 A schematic diagram of the structure of a conversion layer provided in another embodiment of the present application;

[0020] Figure 6 A schematic diagram of the structure of a feature enhancement and denoising module provided in another embodiment of the present application;

[0021] Figure 7 A schematic diagram of the structure of a capsule network provided in another embodiment of the present application;

[0022] Figure 8 An original monitoring input image provided by another embodiment of the present application;

[0023] Fig. 9 A labeled monitoring input image is provided in another embodiment of the present application.

[0024] Fig.10 A schematic diagram of the structure of a storage medium provided in another embodiment of the present application;

[0025] Fig.11 A schematic diagram of the structure of an electronic device provided in another embodiment of the present application; DETAILED DESCRIPTION

[0026] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although the specific embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0027] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the implementation of the present application, but the description is based on the general principles of the specification and is not used to limit the scope of the present application. The scope of protection of the present application shall be determined by the attached claims.

[0028] To facilitate understanding of the embodiments of the present application, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and the various drawings do not constitute a limitation on the embodiments of the present application.

[0029] Figure 1 FIG. 1 is a flow chart of a monitoring data identification method provided by an exemplary embodiment of the present application. Figure 1 As shown, the method includes:

[0030] S100: collecting monitoring input images;

[0031] S200: Preprocessing the monitoring input image;

[0032] S300: constructing a monitoring data recognition model and performing training;

[0033] S400: Inputting the preprocessed monitoring input image into a trained monitoring data recognition model to recognize monitoring data in the monitoring input image.

[0034] This application can automatically, accurately and efficiently extract and identify key monitoring data from monitoring input images, which can not only greatly reduce the workload of medical staff and related personnel, but also significantly improve the accuracy and efficiency of monitoring data processing.

[0035] In another exemplary embodiment, in step S200, preprocessing the input image includes:

[0036] S201: adjusting the monitoring input image to a uniform size;

[0037] In this step, since the monitoring input images may come from monitors with different resolutions, these images need to be adjusted to a uniform size to ensure the consistency of the model input. Specifically, all images can be adjusted to the same width and height through image resampling techniques (such as bilinear interpolation, nearest neighbor interpolation, etc.). This step helps to eliminate the impact of different image sizes, allowing the model to focus on learning features in the image rather than the size of the image.

[0038] S202: performing image enhancement on the monitoring input image after the unified size;

[0039] In this step, the unified-size monitoring input image can be enhanced by methods such as brightness adjustment, contrast enhancement, color conversion, rotation, translation, scaling, shearing, etc. For example, the brightness and contrast of the image can be randomly changed, or the image can be rotated and translated to simulate shooting conditions in different scenes. By enhancing the input image, the robustness of the model can be increased, so that the model can still maintain a high recognition accuracy under different lighting conditions, shooting angles, etc.

[0040] S203: Normalizing the monitoring input image after image enhancement.

[0041] In this step, the enhanced monitoring input image can be normalized by subtracting the mean value from each pixel value in the monitoring input image and dividing it by the standard deviation. For example, for an RGB image, the mean and standard deviation of the three channels R, G, and B can be calculated respectively, and then the formula (pixel represents the original value of a pixel in the image; mean represents the average value of all pixel values ​​in the image; std represents the standard deviation of all pixel values ​​in the image) is normalized to make the pixel values ​​of different images comparable, and to avoid the problem of gradient disappearance or explosion caused by too large or too small values. It should be noted that based on the above normalization formula, the pixel values ​​of the image can be adjusted from the original distribution to a normal distribution with a mean of 0 and a standard deviation of 1. The benefits of this treatment are: first, it can eliminate the influence of the numerical range: different images may have different pixel value ranges, and normalization can unify these values ​​into a standard range, so that the model will not be affected by different numerical ranges when processing different images; second, it can speed up the training of the model: the normalized data distribution is closer to the normal distribution, which helps the gradient descent algorithm converge faster, thereby speeding up the training of the model; third, it can improve the performance of the model: normalization can reduce the influence of the numerical range on the update of the model weights, making the model more stable during the training process, thereby improving the performance of the model.

[0042] In summary, through the above three steps, it can be ensured that the monitoring input images input into the model are both consistent and diverse, thereby improving the accuracy and stability of the monitoring data recognition model.

[0043] In another exemplary embodiment, in step S300, Figure 2 As shown, the monitoring data recognition model includes:

[0044] Input layer, multi-scale feature extraction network, multi-head attention mechanism, feature enhancement and denoising module, capsule network layer and output layer.

[0045] Specifically, Figure 3 As shown, the multi-scale feature extraction network includes a dense connection block, a conversion layer, a dense connection block, a conversion layer and a dense connection block connected in sequence, wherein, Figure 4 As shown, the densely connected block includes an input layer, a batch normalization layer, a ReLU activation function, a 3×3 convolution layer, a batch normalization layer, a 5×5 convolution layer, a batch normalization layer, a 7×7 convolution layer, a batch normalization layer and an output layer, which are connected in sequence. In addition, the densely connected block is internally connected via a skip connection (SkipConnection) to form a pyramid structure, that is, the 5×5 convolution layer receives input data and the first feature map generated by the 3×3 convolution layer, and generates a second feature map through a convolution operation; the 7×7 convolution layer receives input data, the first feature map generated by the 3×3 convolution layer, and the second map generated by the 5×5 convolution layer, and generates a third map through a convolution operation, and the output layer splices the maps generated by all the convolution layers and outputs them. In the example Figure 4In the densely connected block shown in FIG. 1 , the input layer is used to receive the original input data; the batch normalization layer is used to normalize the original input data to reduce the internal covariate shift. Shift), accelerating the model training process; the ReLU activation function enables the model to learn more complex patterns by introducing nonlinearity; the 3×3 convolution layer is used to extract local features in the input data and keep the spatial dimension unchanged; the 5×5 convolution layer is used to further extract more complex features in the input data and has a larger receptive field than the 3×3 convolution layer. Therefore, the 5×5 convolution layer can cover a larger area in the input data, thereby being able to capture a wider range of contextual information and more complex features, thereby enabling the model to better understand the patterns and structures in the image, especially when dealing with tasks such as image recognition that require comprehensive analysis of a larger area, which helps to improve the recognition accuracy and robustness of the model; the 7×7 convolution layer has a larger receptive field than the 5×5 convolution layer, and can further extract features in a larger range in the input data. In addition, due to the jump connection of each convolution layer, the 7×7 convolution layer can synthesize information from the input data, the feature map generated by the 3×3 convolution layer, and the feature map generated by the 5×5 convolution layer, further enhancing the diversity and richness of the features. Therefore, the 7×7 convolution layer can capture a wider range of contextual information and is suitable for situations where it is necessary to understand image features in a larger range, such as identifying large-sized objects in an image or understanding the macroscopic structure of an image. The above convolution layers are connected to each other through skip connections to form a pyramid structure. Specifically, the 5×5 convolution layer not only extracts features directly from the input data, but also extracts feature maps from the 3×3 convolution layer and combines them with its own feature maps, so that information can flow between different scales. Finally, the feature maps generated by all convolution layers are spliced ​​together at the output layer to form a multi-channel feature map. Through skip connections, each convolution layer can directly access the feature maps of all previous layers, which means that each layer in the network can learn from the entire historical layer, not just from the previous layer. This direct connection method promotes feature reuse and allows feature maps to be directly transferred between different convolution layers instead of having to be transferred layer by layer. In this way, the features of the shallower layers can be directly transferred to the deeper layers, so that the deep network can use these features without re-learning. This direct access to features allows each layer to benefit from global information, not just local information. In addition, skip connections also help gradients flow through the entire network, alleviating the vanishing gradient problem common in deep networks, allowing effective training even when the network becomes very deep.

[0046] In summary, since monitoring data often contains a variety of different information and patterns such as heart rate, blood pressure, and blood oxygen saturation, these information are usually distributed at different locations and scales of the monitoring input image. The densely connected block can well integrate feature information of different scales through multi-level feature reuse and effective gradient propagation mechanism, so that the model can better understand the patterns and structures in the input image, thereby improving the learning ability and generalization ability of the model, making the model more efficient and accurate in processing monitoring input image recognition tasks.

[0047] like Figure 5 As shown, the conversion layer includes a 1×1 convolution layer, a 3×3 convolution layer, a batch normalization layer, a ReLU activation function, a maximum average pooling layer and an output layer connected in sequence. The 1×1 convolution layer is used to reduce the number of channels of the input feature map. It generates a new feature map by linearly combining the channels of the input feature map, thereby reducing the complexity of subsequent calculations. The 3×3 convolution layer is responsible for extracting local spatial features based on the 1×1 convolution layer. It performs convolution operations on the local area of ​​the input feature map in a sliding window manner to extract features with spatial correlation, such as edges, textures and other local structures. In addition, compared with the 1×1 convolution layer, the 3×3 convolution layer has a larger receptive field and can capture the spatial relationship between adjacent pixels in the input feature map. The combined use of the 1×1 convolution layer and the 3×3 convolution layer enables the conversion layer to effectively perform feature dimensionality reduction and spatial feature extraction, which not only reduces the computational burden, but also retains important local feature information, which helps to improve the computational efficiency and feature expression ability of the network. The maximum average pooling layer (Max AveragePooling) is used to reduce the spatial size of the feature map. This spatial downsampling operation helps to extract more abstract features, while also reducing the amount of calculation, allowing the model to process deeper information. Furthermore, through batch normalization (BatchNormalization) and ReLU activation function, the conversion layer can maintain the validity of features while reducing dimensions, ensuring that the feature map after dimensionality reduction is still rich in information. Batch normalization helps to speed up the training process and improve the generalization ability of the model, and the ReLU activation function introduces nonlinearity to the model, allowing the network to learn more complex patterns.

[0048] In summary, the conversion layer acts as a bridge between densely connected blocks, which not only reduces the dimension of the feature map, but also plays a role in feature selection, helping to pass important features to the next densely connected block. In this way, the model can focus more on important features while reducing unnecessary redundant information.

[0049] The multi-head attention mechanism first divides the features of the input image (usually the output feature map of the previous layer of the network) into multiple smaller "heads", each of which is responsible for processing part of the features. This division enables the model to focus on different aspects of the input features in parallel, thereby improving the model's ability to handle complex features. For each head, the multi-head attention mechanism calculates an attention score, which represents the relevance between different positions. Specifically, it calculates these scores based on the concepts of "query", "key" and "value", where query, key and value all come from different linear transformations of the input features. By calculating the dot product of query and all keys and then applying the softmax function for normalization, the attention weight of each position to all other positions can be obtained. These weights are then used to weighted sum the values ​​to generate the output of each head. The outputs of all heads are concatenated and then restored to the original feature dimension through a linear transformation (usually a fully connected layer). This process ensures that the multi-head attention mechanism can output feature maps of the same dimension as the input features, which is convenient for processing in subsequent layers. The multi-head attention mechanism can capture more feature information by focusing on different aspects of the input features in parallel, thereby enhancing the model's feature expression ability. This is particularly important for processing complex monitoring input images, because monitoring data often contains a variety of different information and patterns. In summary, the multi-head attention mechanism processes the input image efficiently and accurately through steps such as segmenting features, calculating attention scores, and splicing results, thereby improving the performance and effect of the monitoring data recognition model.

[0050] like Figure 6As shown, the feature enhancement and denoising module includes a feature enhancement part and a denoising part, wherein the feature enhancement part includes an input layer, a convolution layer, a batch normalization layer (BN), a ReLU activation function, a convolution layer, a batch normalization layer (BN), a ReLU activation function and an output layer connected in sequence, wherein the feature enhancement part adopts a residual structure, that is, the input feature map skips the first convolution layer and the second convolution layer and is added to the output of the second convolution; the denoising part includes an encoder and a decoder, wherein the encoder includes a convolution layer, a batch normalization layer (BN) and a ReLU activation function, and the decoder includes a deconvolution layer, a batch normalization layer (BN), a ReLU activation function and an output layer, and the denoising part also adopts a residual structure, that is, the output of the feature enhancement part is added to the output of the denoising part. In the feature enhancement part, through two convolution operations, each accompanied by batch normalization and activation functions, the useful features in the input image (such as specific values ​​including heart rate, blood pressure, blood oxygen saturation, etc., and another example including waveforms or bar graphs reflecting physiological signals displayed on monitoring equipment, etc.) can be highlighted to make them more obvious, so that the model can more easily detect and identify these key data from the image. The purpose of the residual connection is to retain the original input information, allowing the input information to be transmitted directly from the input to the output, avoiding the gradient vanishing or gradient exploding problems in the deep network, so as to help the model better learn and maintain features. The encoder reduces the dimension of the input feature map to remove unnecessary noise, and the decoder attempts to reconstruct the image from the reduced-dimensional features, further clean up the noise and restore useful details, thereby helping to improve the purity of the features, so that the model can more clearly and accurately identify the target information. The outputs of the feature enhancement part and the denoising part are superimposed through residual connections, which can achieve simultaneous feature enhancement and noise reduction, so that the denoised features and the enhanced features can be better integrated, so that the final output is an enhanced and clean feature map, thereby enhancing the model's ability to capture key features in the image. In summary, through the above mechanism, the model can focus more on important features when processing images and reduce unnecessary redundant information, so that it can more accurately identify and extract key monitoring data from the input image, and improve the accuracy and efficiency of image recognition.

[0051] like Figure 7As shown, the capsule network layer includes an input layer, a primary capsule layer, a dynamic routing layer, a high-level capsule layer and a reconstruction layer connected in sequence, wherein the input layer is used to receive the output of the feature enhancement and denoising module; the primary capsule layer uses a convolution operation to extract basic features from the input feature map, and the primary capsule layer contains multiple neurons (8 convolution kernels are used, each convolution kernel generates a primary capsule, and each capsule contains 8 neurons) for encoding the existence and properties of basic features. Multiple convolution kernels can be used, each convolution kernel generates a primary capsule, and each primary capsule corresponds to a specific type of simple feature, such as edge, texture, etc., and is passed to the dynamic routing layer; the dynamic routing layer is used to determine the primary capsule and the high-level The connection weights between capsules, the dynamic routing layer iteratively updates the connection weights to ensure that the high-level capsules can obtain the most valuable information from the most relevant primary capsules; the high-level capsule layer is composed of more complex capsules, which are used to detect higher-level objects or parts of objects in the image. The high-level capsules obtain information from the primary capsule layer through the dynamic routing layer, and integrate this information to detect more complex features or objects. The high-level capsules also contain multiple neurons for encoding the attributes of the object, such as position, direction, size, etc.; the reconstruction layer decodes the output of the high-level capsule to generate a reconstructed image similar to the input image. The reconstruction layer is used to assist in training the capsule network and verify its ability to understand the input image. The reconstruction layer can evaluate the learning effect of the capsule network by comparing the similarity between the reconstructed image and the original input image.

[0052] In summary, the capsule network layer can effectively capture high-level features in the image, such as the shape and posture of the object, and can handle objects with different postures. Even if the target object is rotated or deformed, the capsule network can still accurately identify them, thereby improving the robustness of image recognition.

[0053] In another exemplary embodiment, in step S300, the monitoring data recognition model is trained by the following steps:

[0054] S301: Collect a large number of monitoring input images to form a monitoring input image set. For example, one monitoring input image is Figure 8 As shown;

[0055] S302: Use different boxes to mark each data item in each monitoring input image. For example, Figure 8 The pulse rate-blood oxygen data in the input image shown is annotated with a solid line box, the blood pressure data is annotated with a dashed line box, the patient ID is annotated with a dotted line box, and so on. The annotated input image is as follows Fig. 9 As shown;

[0056] In this step, by using different boxes to label each data item, the model can learn the characteristics of different data items more quickly, so that the model can more easily find and identify different types of data; in addition, different boxes can enhance the contrast between different data items in the input image, so that the model can more easily distinguish the position and content of each data item, which helps the model learn clearer and more specific feature expressions during the training process; further, during the model training process, by viewing the labeling results, it is possible to intuitively find out where the model has problems, so that the model can be adjusted and optimized in a targeted manner, thereby improving the model's recognition efficiency and accuracy for monitoring data.

[0057] S303: Divide the labeled input image into a training set and a validation set in proportion, and the division ratio may be 7:3;

[0058] S304: Setting training parameters, for example, setting the training batch size to 32, the number of training rounds to 100, and the learning rate to 0.001, training the model using the training set, calculating the cross entropy loss function of the model during the training process, and using the Adam optimization algorithm to optimize the loss function until the loss function converges;

[0059] S305: Use the validation set to validate the model. During the validation process, accuracy, recall and F1 score are used as evaluation indicators to evaluate the performance of the model. When all indicators reach the preset values ​​(for example, the accuracy must meet 95%, the recall must meet 90%, and the F1 score must meet 92%), the model validation is passed; otherwise, adjust the training parameters (for example, adjust the number of training rounds to 150, adjust the learning rate to 0.05) or expand the training samples (for example, adjust the division ratio of the training set and the validation set to 8:2) to re-train the model until the model validation is passed.

[0060] In another exemplary embodiment, the cross entropy loss function is expressed as:

[0061]

[0062] Among them, L confidence-CE represents the loss function, λ represents the penalty weight, δ represents the confidence threshold, y i The i-th element in the one-hot encoding vector representing the true label, p i represents the predicted probability of the model for the i-th category, C represents the total number of categories, max(0,δ-p i ) represents the specific implementation of the confidence penalty term, which calculates the predicted probability p i The difference between the confidence threshold δ and the maximum value between the difference and 0 is taken. iWhen it is less than δ, the difference is positive, thus producing a penalty effect. i If it is greater than or equal to δ, the penalty term is 0 and has no effect on the total loss.

[0063] In the training process of the monitoring data recognition model described in this application, by adopting the cross entropy loss function with confidence penalty, the following special effects can be brought about:

[0064] 1. Improve the model's penalty for uncertain predictions: The traditional cross entropy loss function focuses on the difference between the model prediction and the true label, while the cross entropy loss function with confidence penalty adds consideration of the model's prediction confidence on this basis. When the model's prediction probability for a certain category is lower than the set confidence threshold, an additional penalty term will be generated, which prompts the model to not only pursue prediction accuracy during training, but also strive to improve prediction confidence and reduce uncertain predictions.

[0065] 2. Enhance the robustness of the model: By introducing confidence penalties, the model will be more cautious about predictions that it is not confident of during training, thereby avoiding errors caused by overconfidence in testing or actual applications. This mechanism helps improve the robustness of the model in complex or noisy environments.

[0066] 3. Optimize model performance: During the training process, the model will constantly try to adjust its parameters to minimize the loss function. Since the cross-entropy loss function with confidence penalty will generate additional penalties when the predicted probability is lower than the confidence threshold, this prompts the model to pay more attention to those difficult-to-distinguish samples during training, which helps to improve the overall performance of the model, including evaluation indicators such as accuracy, recall, and F1 score.

[0067] 4. Promote model convergence: In the early stages of training, the model may have vague predictions for all categories, that is, the prediction probability distribution is relatively uniform. By introducing confidence penalties, the model will gradually tend to produce clearer and more confident predictions during training, which helps accelerate the model's convergence process and improve training efficiency.

[0068] 5. Adapt to different data distributions: In practical applications such as medical monitoring data recognition, data distribution may be complex and unbalanced. The cross entropy loss function with confidence penalty can alleviate the problem of data imbalance to a certain extent. By penalizing low-confidence predictions, the model is prompted to pay more attention to samples that are difficult to identify, thereby improving the model's recognition ability under complex data distribution.

[0069] In summary, the use of the cross entropy loss function with confidence penalty has a significant improvement effect on the training of the monitoring data recognition model proposed in this application, which can enhance the robustness of the model, optimize the model performance, promote model convergence and adapt to different data distributions.

[0070] Below, still with Fig.10 Taking as an example, the recognition process of the monitoring data recognition model described in this application is described:

[0071] First, the densely connected blocks in the multi-scale feature extraction network extract multi-scale features in the image through convolutional layers of different scales such as 3×3, 5×5, and 7×7, and reuse features through skip connections. The conversion layer further reduces the dimension and extracts features from the feature map, reducing the computational burden and retaining important features.

[0072] Secondly, the feature enhancement and denoising module highlights useful features in the image (such as heart rate, blood pressure, blood oxygen saturation, etc.) through convolution and residual connection, and uses the encoder and decoder structure to remove noise in the image to improve the purity of the features.

[0073] Secondly, in the capsule network layer, the primary capsule layer uses convolution operations to extract basic features from the feature map and generate primary capsules. The dynamic routing layer iteratively updates the connection weights to ensure that high-level capsules obtain valuable information from the most relevant primary capsules. The high-level capsule layer is responsible for detecting higher-level objects or parts of objects in the image and generating high-level features by integrating information. The reconstruction layer generates a reconstructed image by decoding the high-level capsule output to verify the model's ability to understand the input image.

[0074] Finally, the output layer is responsible for outputting Fig.10 The monitoring data in the

[0075] Blood oxygen: 97%

[0076] Blood pressure: 100 / 62 mmHg

[0077] Pulse rate-blood oxygen: 95bpm

[0078] Patient ID: 3000858263

[0079] It can be seen from the output results that the monitoring data recognition model proposed in this application can clearly identify Fig.10 All monitoring data are accurate and there is no error.

[0080] Furthermore, the present application also compares and analyzes the monitoring data recognition model with the convolutional neural network CNN, the recurrent neural network RNN ​​and the traditional image processing and machine learning combined method. Specifically, taking a monitoring data set as an example, the data set contains images of different formats, resolutions and clarity obtained from monitors of various brands and models. These images contain key data such as heart rate, blood pressure, and blood oxygen saturation, and the data set contains a large number of labeled images, and the labels are specific values ​​contained in the images. The comparison results are shown in Table 1:

[0081] Table 1

[0082]

[0083] It can be seen from Table 1 that the monitoring data recognition model proposed in this application benefits from the application of technologies such as multi-scale feature extraction network and multi-head attention mechanism, which makes its various indicators far higher than traditional methods. Therefore, it can be seen that the monitoring data recognition model proposed in this application is superior in processing monitoring input images.

[0084] In another exemplary embodiment, the present application also provides a monitoring data recognition device, which includes: an acquisition module for acquiring monitoring input images; a preprocessing module for preprocessing the monitoring input images; a model building and training module for building a monitoring data recognition model and training it; and a recognition module for inputting the preprocessed monitoring input image into a trained monitoring data recognition model to identify the monitoring data in the monitoring input image.

[0085] In another exemplary embodiment, the preprocessing module includes: a size adjustment submodule, used to adjust the monitoring input image to a uniform size; an enhancement submodule, used to perform image enhancement on the monitoring input image after the uniform size; and a normalization submodule, used to normalize the monitoring input image after image enhancement.

[0086] In another exemplary embodiment, the present application also provides a monitoring data identification system, which includes the device as described above.

[0087] Based on the above embodiments, Fig.10 , for an explanation of the computer-readable storage medium of the exemplary embodiment of the present application, please refer to Fig.10, the computer-readable storage medium shown is a CD 40, on which a computer program (i.e., a program product) is stored. When the computer program is executed by the processor, it will implement the steps described in the above method implementation, for example, collecting monitoring input images; preprocessing the monitoring input images; building a monitoring data recognition model and training it; inputting the preprocessed monitoring input images into the trained monitoring data recognition model to recognize the monitoring data in the monitoring input images. It should be noted that the computer-readable storage medium includes, but is not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which will not be described one by one here.

[0088] Based on the above embodiments, the present application also provides an electronic device, as shown below: Fig.11 An electronic device for downloading a file according to an exemplary embodiment of the present application is described.

[0089] Fig.11 A block diagram of an exemplary electronic device 50 suitable for implementing the embodiments of the present application is shown, and the electronic device 50 may be a computer system or a cloud server. Fig.11 The electronic device 50 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0090] like Fig.11 As shown, the electronic device 50 includes but is not limited to: one or more processors or processing units 501, a system memory 502, and a bus 503 connecting different system components (including the system memory 502 and the processing unit 501).

[0091] The electronic device 50 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 50, including volatile and non-volatile media, removable and non-removable media.

[0092] The system memory 502 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 5021 and / or cache memory 5022. The electronic device 50 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the ROM 5023 may be used to read and write non-removable, non-volatile magnetic media ( Fig.11 is not shown in the Fig.11As shown in FIG. 5 , a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) can be provided. In these cases, each drive can be connected to bus 503 via one or more data medium interfaces. System memory 502 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of each embodiment of the present application.

[0093] A program / utility 5025 having a set (at least one) of program modules 5024 may be stored, for example, in system memory 502, and such program modules 5024 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 5024 generally perform the functions and / or methods of the embodiments described herein.

[0094] The electronic device 50 may also communicate with one or more external devices 504 (e.g., a keyboard, a pointing device, a display, etc.). Such communication may be performed via an input / output (I / O) interface 505. Furthermore, the electronic device 50 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 506. Fig.11 As shown, the network adapter 506 communicates with other modules (such as the processing unit 501, etc.) of the electronic device 50 via the bus 503. It should be understood that although Fig.11 Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 50 .

[0095] The processing unit 501 executes various functional applications and data processing by running the program stored in the system memory 502, for example, collecting monitoring input images; preprocessing the monitoring input images; constructing and training monitoring data recognition models; inputting the preprocessed monitoring input images into the trained monitoring data recognition model to identify the monitoring data in the monitoring input images. The specific implementation of each step is not repeated here. It should be noted that although several units / modules or sub-units / sub-modules of the file concurrent downloading device are mentioned in the above detailed description, this division is only exemplary and not mandatory. In fact, according to the implementation of the present application, the features and functions of two or more units / modules described above can be concretized in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be concretized.

[0096] In the description of the present application, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0097] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0098] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0099] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0101] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that can be executed by a processor. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a cloud server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0102] The above embodiments are only for illustrating the technical concept and features of the present application, and their purpose is to enable people familiar with the technology to understand the content of the present application and implement it accordingly, and they cannot be used to limit the protection scope of the present application. Any equivalent changes or modifications made according to the spirit of the present application should be included in the protection scope of the present application.

Claims

1. A monitoring data identification method, characterized in that: The method comprises: Collect monitoring input images; Preprocessing the monitoring input image; Constructing a monitoring data recognition model and training it; wherein the monitoring data recognition model introduces a dense connection block, the dense connection block includes superimposed convolutional layers with different convolutional kernels, and the superimposed convolutional layers with different convolutional kernels use jump connections to achieve feature reuse; The preprocessed monitoring input image is input into a trained monitoring data recognition model to recognize the monitoring data in the monitoring input image.

2. The method according to claim 1, characterized in that Preprocessing the monitoring input image includes: Adjusting the monitoring input image to a uniform size; Performing image enhancement on the monitoring input image after the unified size; The monitoring input image after image enhancement is normalized.

3. The method according to claim 1, characterized in that The monitoring data recognition model is trained by the following steps: Collect a large number of monitoring input images to form a monitoring input image set; Each data item in each monitoring input image is marked with a different box; Divide the labeled input images into training and validation sets in proportion; Set training parameters, train the model using the training set, calculate the cross entropy loss function of the model during the training process, and use the Adam optimization algorithm to optimize the loss function until the loss function converges; The model is verified using the validation set. During the verification process, accuracy, recall and F1 score are used as evaluation indicators to evaluate the model performance. When all indicators reach the preset values, the model is verified. Otherwise, the training parameters are adjusted or the training samples are expanded to retrain the model until the model is verified.

4. A monitoring data identification device, characterized in that: The device comprises: An acquisition module, used for acquiring monitoring input images; A preprocessing module, used for preprocessing the monitoring input image; A model building and training module, used to build and train a monitoring data recognition model; wherein the monitoring data recognition model introduces a dense connection block, the dense connection block includes superimposed convolutional layers with different convolutional kernels, and the superimposed convolutional layers with different convolutional kernels use jump connections to achieve feature reuse; The recognition module is used to input the preprocessed monitoring input image into the trained monitoring data recognition model to recognize the monitoring data in the monitoring input image.

5. The device according to claim 4, characterized in that The preprocessing module comprises: A size adjustment submodule, used for adjusting the monitoring input image to a uniform size; An enhancement submodule, used for performing image enhancement on the monitoring input image after the image is unified in size; The normalization submodule is used to normalize the monitoring input image after image enhancement.

6. A monitoring data identification system, characterized in that: The system comprises the apparatus of claim 4.

7. A storage medium, characterized in that: The method comprises instructions, which, when executed on a computer, enable the computer to execute the method according to any one of claims 1 to 3.

8. An electronic device, characterized in that: The electronic device comprises: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 3 when executing the program.

Citation Information

Patent Citations

  • Image segmentation method, system and electronic device based on depth learning

    CN109118491A

  • Image processing method and device, medium and computing equipment

    CN110163205A

  • Lunar meteorite crater identification method and device based on depth residual U-Net and storage medium

    CN113139592A

  • Two-stage Brillouin frequency shift extraction method

    CN113465676A

  • Method and device for reading vital sign data in monitoring equipment

    CN113647920A