Disease detection method, device and computer equipment

Abstract

The present application relates to a disease detection method, device, computer equipment, computer-readable storage medium and computer program product. The method includes: obtaining response vibration waves of multiple target positions in a target object; the response vibration waves are generated by the collision of a percussion device of different materials with the target object; extracting and screening the features of the response vibration waves according to the timing analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration waves of each target position; performing correlation analysis on the first waveform features corresponding to each response vibration wave of the target position based on the spatial analysis structure of the disease detection model to determine the disease detection result of the target object. The use of this method can improve the accuracy of the disease detection results.

Description

Technical Field

[0001] The present application relates to the technical field of structural detection, and in particular to a disease detection method, device and computer equipment. Background Art

[0002] With the development of bridge engineering, most of the axial force at the steel-concrete joint is transmitted by the pressure plate. Whether the steel cell pressure plate and concrete are dense plays an important role in the bearing capacity of the joint. However, internal defects such as concrete voids are prone to occur during the pouring construction of steel cell concrete.

[0003] In traditional technology, for the diagnosis of internal defects of steel plate concrete, the inspector knocks on the surface of the steel plate concrete and evaluates the vibration feedback of the steel plate concrete after being knocked based on the inspector's experience to determine whether there are defects inside the steel plate concrete and obtain the test results.

[0004] However, in traditional technologies, due to the subjectivity of inspectors, the accuracy of concrete void detection is poor. Summary of the invention

[0005] Based on this, it is necessary to provide a disease detection method, device, computer equipment, computer readable storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a disease detection method, comprising:

[0007] Acquiring response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object;

[0008] Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position;

[0009] Based on the spatial analysis structure of the disease detection model, a correlation analysis is performed on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object.

[0010] In one embodiment, before obtaining the response vibration waves of multiple target positions in the target object, the method includes:

[0011] Obtaining multiple training sample sets and sample labels corresponding to each of the training sample sets;

[0012] Performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each of the training sample sets to obtain a loss value of the disease detection model to be trained;

[0013] Based on the loss value and the gradient descent algorithm, the disease detection model to be trained is back-propagated, and the model parameters of the disease detection model to be trained are adjusted until the loss value meets the preset loss condition, thereby obtaining a trained disease detection model.

[0014] In one embodiment, the training sample set includes vibration wave samples obtained at target positions by percussion devices made of multiple different materials and position information of the vibration wave samples, and the sample labels include detection labels indicating whether a target disease exists at each of the target positions.

[0015] In one embodiment, the loss value includes a first loss corresponding to a time series analysis structure to be trained and a second loss corresponding to a space analysis structure to be trained in the disease detection model to be trained; the feature extraction and prediction processing of the disease detection model to be trained according to the training sample set and the sample labels corresponding to each of the training sample sets to obtain the loss value of the disease detection model to be trained includes:

[0016] For each of the training sample sets, extracting features from each of the vibration wave samples according to the time series analysis structure to be trained of the disease detection model to be trained, to obtain a second waveform feature;

[0017] Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each of the target positions;

[0018] A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

[0019] In one embodiment, the back propagation of the disease detection model to be trained is performed based on the loss value and the gradient descent algorithm, and the model parameters of the disease detection model to be trained are adjusted until the loss value meets the preset loss condition to obtain the trained disease detection model, including:

[0020] Back-propagating the spatial analysis structure to be trained according to the second loss, and updating the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss satisfies a preset loss condition, and outputting a target feature extraction result that satisfies the preset loss condition;

[0021] The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure, until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

[0022] In one embodiment, the feature extraction and feature screening of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform feature corresponding to the response vibration wave at each target position includes:

[0023] Extracting features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features;

[0024] The initial waveform feature is subjected to feature screening based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure to obtain a first waveform feature.

[0025] In one embodiment, the spatial analysis structure based on the disease detection model performs correlation analysis on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object, including:

[0026] Extracting features of the first waveform features and the target positions corresponding to each of the response vibration waves based on the spatial analysis structure of the disease detection model, and combining the extracted features to obtain spatial features between the response vibration waves at each of the target positions;

[0027] Determine the detection result of each target location according to the association relationship between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease;

[0028] The disease detection result of the target object is determined based on the detection results of each of the target positions.

[0029] In one embodiment, determining the disease detection result of the target object based on the detection results of each of the target positions includes:

[0030] If the target disease does not exist in the detection results of each of the target positions, determine the disease detection result of the target object as not having the target disease;

[0031] If the target disease exists in any of the target positions, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

[0032] In a second aspect, the present application also provides a disease detection device, comprising:

[0033] A first acquisition module is used to acquire response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object;

[0034] A feature extraction module, used for extracting and screening features of the response vibration wave according to the time series analysis structure of the disease detection model, to obtain a first waveform feature corresponding to the response vibration wave at each target position;

[0035] The correlation analysis module is used to perform correlation analysis on the first waveform features corresponding to each of the response vibration waves at the target position based on the spatial analysis structure of the disease detection model to determine the disease detection result of the target object.

[0036] In one embodiment, the device further comprises:

[0037] A second acquisition module is used to acquire multiple training sample sets and sample labels corresponding to each of the training sample sets;

[0038] A prediction module, used for performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample set and the sample labels corresponding to each of the training sample sets, to obtain the loss value of the disease detection model to be trained;

[0039] The training module is used to perform back propagation on the disease detection model to be trained based on the loss value and the gradient descent algorithm, adjust the model parameters of the disease detection model to be trained, until the loss value meets the preset loss condition, and obtain the trained disease detection model.

[0040] In one embodiment, the training sample set includes vibration wave samples obtained at target positions by percussion devices made of multiple different materials and position information of the vibration wave samples, and the sample labels include detection labels indicating whether a target disease exists at each of the target positions.

[0041] In one embodiment, the loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; the prediction module is specifically used to extract features of each vibration wave sample according to the time series analysis structure to be trained of the disease detection model to be trained for each training sample set, so as to obtain a second waveform feature;

[0042] Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each of the target positions;

[0043] A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

[0044] In one of the embodiments, the training module is specifically used to perform back propagation on the spatial analysis structure to be trained according to the second loss, and update the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss satisfies a preset loss condition, and output a target feature extraction result that satisfies the preset loss condition;

[0045] The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure, until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

[0046] In one embodiment, the feature extraction module is specifically used to extract features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features;

[0047] The initial waveform feature is subjected to feature screening based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure to obtain a first waveform feature.

[0048] In one embodiment, the association analysis module is specifically used to extract the first waveform features corresponding to each of the response vibration waves and the target position based on the spatial analysis structure of the disease detection model, and combine the extracted features to obtain the spatial features between the response vibration waves of each of the target positions;

[0049] Determine the detection result of each target location according to the association relationship between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease;

[0050] The disease detection result of the target object is determined based on the detection results of each of the target positions.

[0051] In one embodiment, the association analysis module is specifically used to determine the disease detection result of the target object as the absence of the target disease if the detection results of each of the target positions do not contain the target disease;

[0052] If the target disease exists in any of the target positions, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

[0053] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0054] Acquiring response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object;

[0055] Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position;

[0056] Based on the spatial analysis structure of the disease detection model, a correlation analysis is performed on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object.

[0057] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0058] Acquiring response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object;

[0059] Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position;

[0060] Based on the spatial analysis structure of the disease detection model, a correlation analysis is performed on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object.

[0061] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0062] Acquiring response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object;

[0063] Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position;

[0064] Based on the spatial analysis structure of the disease detection model, a correlation analysis is performed on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object.

[0065] The above-mentioned disease detection method, device and computer equipment can improve the diversity of waveform features and the reliability of disease detection in complex environments and for complex target objects by analyzing the response vibration waves generated by the collision of knocking devices of different materials with the target object; according to the timing analysis structure of the disease detection model, feature extraction and feature screening of the response vibration waves are performed to obtain the first waveform features with stronger reflection ability for the target disease, and then the first waveform features of the response vibration waves generated by each target position are correlated and analyzed using the spatial analysis structure to determine the disease detection results of the target object, thereby improving the accuracy of the disease detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0067] Figure 1 Schematic diagram of a disease detection method in one embodiment;

[0068] Figure 2 A schematic diagram of a process for training a disease detection model to be trained in one embodiment;

[0069] Figure 3 A schematic diagram of a process of calculating the loss of a time series analysis structure to be trained and a spatial analysis structure to be trained in one embodiment;

[0070] Figure 4 A schematic diagram of a process of adjusting model parameters during training of a disease detection model to be trained in one embodiment;

[0071] Figure 5 A schematic diagram of a process of extracting features from a timing analysis structure in one embodiment;

[0072] Figure 6 A schematic diagram of a process of predicting a spatial analysis structure in one embodiment;

[0073] Figure 7 A schematic diagram of a convolutional layer of a spatial analysis structure in one embodiment;

[0074] Figure 8 A schematic diagram of a process for determining a disease detection result in another embodiment;

[0075] Fig. 9 is a structural block diagram of a disease detection device in one embodiment;

[0076] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0078] In one embodiment, Figure 1 As shown, a disease detection method is provided. This embodiment uses the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0079] Step 102: Acquire response vibration waves of multiple target positions in the target object.

[0080] Among them, the response vibration wave is generated by the collision of the striking device made of different materials with the target object.

[0081] In the embodiment of the present application, the target object may be the steel plate concrete part of the steel-concrete joint in the bridge structure. It is understandable that the target object may also be a concrete structure or a steel plate concrete structure in other building structures. First, the terminal obtains the response vibration waves of multiple target positions in the target object through an acquisition sensor, and the acquisition sensor may be one or more. The response vibration wave may be generated by tapping the surface of the target object by manual tapping or mechanical tapping. The tapping by mechanical equipment is used as an example for explanation. The mechanical equipment includes tapping equipment of different materials, such as a wooden hammer head, a rubber hammer head, a nylon hammer head, a steel hammer head, etc. The target position may be a tapping point pre-constructed by the mechanical equipment, and the tapping sequence for each target position is pre-set in the mechanical equipment. By using tapping equipment of different materials to tap the surface of the target object, a response vibration wave is generated on the structure of the target object. Furthermore, the acquisition sensor can collect the response vibration wave generated by the target object, and convert the response vibration wave into a digital signal and transmit it to the terminal. Optionally, after receiving the digital signal of the response vibration wave, the terminal normalizes the response vibration waves generated by the striking devices at different positions and with different materials to obtain the response vibration waves at each target position.

[0082] Step 104 , extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model, and obtaining the first waveform features corresponding to the response vibration wave at each target position.

[0083] In an embodiment of the present application, the disease detection model may be a disease detection model for concrete voiding disease, and the disease detection model includes a time series analysis structure and a space analysis structure, wherein the time series analysis structure may be an LSTM (Long Short-Term Memory) structure.

[0084] First, after the terminal obtains the response vibration waves generated at different target positions, the terminal inputs the response vibration waves into the defect detection model, and performs a preliminary analysis on the waveform characteristics of each response vibration through the time series analysis structure. The influence of the presence of concrete voids on the response vibration waves includes dimensions such as spectrum characteristics, amplitude energy, and waveform shape. For target positions with defects, there may be an increase or decrease in specific frequencies in the spectrum, reflecting the change in the structural natural frequency of the target object or the change in the vibration propagation path caused by the defect; the amplitude of the response vibration wave may be higher, and the energy distribution may be more uneven, indicating a loss of local structural rigidity or durability; the waveform of the response vibration wave may show more harmonics or nonlinear characteristics, reflecting the internal structural unevenness or degree of damage of the target object.

[0085] The timing analysis structure is based on the correlation between the waveform features learned during the training process and the concrete void disease. The timing analysis structure pays more attention to the features that can reflect the more prominent features of the concrete void disease, and extracts the features of the response vibration wave of each target position. According to the correlation between each characteristic index in the waveform features learned by the timing analysis structure and the concrete void disease, the weight of each characteristic index of the response vibration wave is determined to achieve feature screening for the response vibration wave, and then the first waveform feature corresponding to the response vibration wave of each target position is obtained. The first waveform feature is a one-dimensional feature vector, which represents the signal feature of the response vibration wave extracted by the LSTM model. The signal feature is related to the spectrum, amplitude and other characteristics of the response wave, and can more prominently reflect whether there is concrete void disease in the structure of the target object.

[0086] Step 106 , based on the spatial analysis structure of the disease detection model, correlation analysis is performed on the first waveform features corresponding to the response vibration waves at the target position to determine the disease detection result of the target object.

[0087] In an embodiment of the present application, after the terminal obtains the first waveform feature corresponding to each response vibration wave, the waveform feature is input into the spatial analysis structure of the disease analysis model, wherein the spatial analysis structure of the disease detection model can be a CNN (Convolutional Neural Network) structure.

[0088] The response vibration waves at different positions reflect the physical characteristics and structural health status of the target concrete structure. The spatial correlation analysis of the spatial analysis structure can analyze the similarities and differences between the target positions by comparing the vibration wave characteristics at different positions. Based on the similarity analysis, it can be inferred whether the abnormal vibration wave in a certain area of ​​the structure is caused by a disease, as well as the distribution range and impact of the disease. The response vibration waves at different target positions are correlated and analyzed, and the vibration waves at different positions are comprehensively analyzed to identify the abnormalities in the structure. First, under normal conditions, the response vibration waves at different positions usually show a certain degree of consistency or predictable change pattern. When there is a disease (concrete degassing) in the target structure, not only will the response vibration waves at the diseased position be abnormal, but the response vibration waves at some other target positions may also be damaged or distorted, which is manifested as different degrees of changes in the characteristics of the vibration waves at different positions, such as spectrum changes, amplitude abnormalities, or waveform distortion.

[0089] The terminal conducts a comprehensive analysis of the response vibration waves of each target position through the spatial analysis structure, which can provide a more comprehensive and reliable structural health status assessment, obtain the confidence level of whether there is concrete voids at each target position, and then obtain the disease detection results of the target object. Through data fusion, it can reduce the misjudgment that may be caused by point data at a single target position, and at the same time increase the understanding and grasp of the overall structural status of the target object by the disease detection model.

[0090] In an optional embodiment, the terminal can also determine the concrete hollowing disease outside the target position in the target object by correlating and analyzing the first waveform features of the response vibration waves of all different target positions. Specifically, the terminal captures the timing features of the response vibration waves through the timing analysis structure in the disease detection model, and can infer the direction information of the concrete hollowing disease relative to the target position of the knocking by analyzing the phase difference between the vibration signals at different positions, and uses the spatial analysis structure to capture the local mode and spatial distribution information of each response vibration wave, thereby determining the target vibration source corresponding to the concrete hollowing disease, and then uses the beamforming technology to analyze the local mode and spatial distribution information of each response vibration wave, determine the direction and position of the target vibration source relative to the receiving position of the response vibration wave of different acquisition devices, and finally determine the specific position of the concrete hollowing disease outside the target position based on the direction information of the concrete hollowing disease relative to the knocking target position and the direction and position of the target vibration source relative to the receiving position of the response vibration wave of different acquisition devices.

[0091] In the above-mentioned disease detection method, by analyzing the response vibration waves generated by the collision of knocking devices of different materials with the target object, the diversity of waveform characteristics can be improved, and the reliability of disease detection in complex environments and for complex target objects can be improved; according to the timing analysis structure of the disease detection model, the characteristics of the response vibration waves are extracted and screened, and the first waveform characteristics with stronger reflection ability for the target disease are screened, and then the first waveform characteristics of the response vibration waves generated at each target position are correlated and analyzed using the spatial analysis structure to determine the disease detection results of the target object, thereby improving the accuracy of the disease detection results.

[0092] In an exemplary embodiment, Figure 2 As shown, before step 102, the disease detection model needs to be trained, and the method also includes steps 202 to 206. Among them:

[0093] Step 202: Obtain multiple training sample sets and sample labels corresponding to each training sample set.

[0094] In the embodiment of the present application, the terminal uses a steel plate concrete non-voided specimen and a series of steel plate concrete with different void depths at different target positions, and uses a tapping device of different materials to tap according to a preset target position and a preset tapping order, and obtains multiple sets of training sample sets reflecting whether there is concrete void disease and multiple target positions with different void depths according to the acquisition sensor. Optionally, after the terminal receives the sample data collected by the acquisition sensor, each response vibration wave is normalized to obtain a training sample set for training the disease detection model to be trained.

[0095] The terminal obtains multiple training sample sets, each of which corresponds to a material of a striking device, whether there is a concrete void disease, and if there is a concrete void disease, a corresponding void depth. Each training sample set contains response vibration waves in multiple target positions, and the sample label of the training sample set is whether there is a concrete void disease at each target position.

[0096] Step 204 , performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each training sample set, to obtain the loss value of the disease detection model to be trained.

[0097] In an embodiment of the present application, the terminal inputs each training sample set into the disease detection model to be trained respectively. For each training sample set, the terminal performs feature extraction on each vibration wave sample and the position information corresponding to the vibration wave sample in the training sample set through the disease detection model to be trained, and performs prediction processing based on the feature extraction results of the vibration wave and position information of each sample, determines the confidence level of the existence of concrete void disease at each target position, and determines the loss value of the disease detection model to be trained based on the confidence level and the sample label. The loss value is used to reflect the data processing accuracy of the current disease detection model to be trained, so as to train the disease detection model to be trained.

[0098] Step 206, back-propagation is performed on the disease detection model to be trained based on the loss value and the gradient descent algorithm, and the model parameters of the disease detection model to be trained are adjusted until the loss value meets the preset loss condition, thereby obtaining a trained disease detection model.

[0099] In an embodiment of the present application, when the loss value of the disease detection model to be trained does not meet the preset threshold, the terminal back-propagates the disease detection model to be trained through the gradient descent algorithm, iteratively adjusts its internal weights, and inputs the next training sample set into the disease detection model to be trained, executes the above step 204, and obtains a new loss value until the loss value meets the preset loss condition, so that the prediction result of the disease detection model to be trained is fitted to the real result consistent with the sample label of the training sample set, and a trained disease detection model is obtained.

[0100] In this embodiment, the disease detection model to be trained is trained by a gradient descent algorithm to obtain a loss value that meets a preset loss condition, and then a trained disease detection model is obtained, which can improve the accuracy of disease detection by the trained detection model in applications.

[0101] In an exemplary embodiment, the training sample set includes vibration wave samples and position information of the vibration wave samples obtained at target positions by using multiple striking devices of different materials, and the sample labels include detection labels indicating whether a target disease exists at each target position.

[0102] Among them, striking devices made of different materials will result in vibration waveforms within different spectral ranges. For example, a hard hammer head may produce higher-frequency vibration waves, while a soft hammer head may produce lower-frequency vibration waves. By using hammer heads made of different materials, a wider range of spectral characteristics can be captured, thereby providing more comprehensive vibration signal information.

[0103] In an exemplary embodiment, the loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; Figure 3 As shown, step 204 includes steps 302 to 306. Among them:

[0104] Step 302: for each training sample set, extract features from each vibration wave sample according to the time series analysis structure to be trained of the defect detection model to be trained, and obtain a second waveform feature.

[0105] In the embodiment of the present application, for each training sample set, the terminal extracts features of each vibration wave sample according to the same principle as step 104 and the to-be-trained timing analysis structure of the to-be-trained disease detection model to obtain the second waveform feature. In the present embodiment, the data processing process of the to-be-trained timing analysis structure is not further described.

[0106] Step 304 , based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each target position.

[0107] In the embodiment of the present application, after the spatial analysis structure to be trained obtains the second waveform feature, the terminal further extracts the second waveform feature through the convolution layer to obtain a feature map of each second waveform feature as the initial feature extraction result. Then, the terminal extracts and predicts the second waveform feature and the target position corresponding to the second waveform feature based on the spatial analysis structure to be trained of the disease detection model to be trained according to the same principle as step 106, and obtains the initial feature extraction result and the prediction result corresponding to each target position. In this embodiment, the data processing process of the spatial analysis structure to be trained will not be described in detail.

[0108] Step 306, determining a first loss of the time series analysis structure to be trained according to the second waveform feature and the initial feature extraction result, and determining a second loss of the space analysis structure to be trained according to the prediction result and the detection label.

[0109] In the embodiment of the present application, the initial feature extraction result is the result of the spatial analysis structure extracting the second waveform feature. Because the spatial analysis structure can capture higher-level abstract features and combine the characteristics of multiple sample vibration waves to determine the relationship between more significant sample vibration waves and sample labels, the difference between the second waveform feature and the initial feature extraction result can guide the training of the timing analysis model. By comparing the difference between the second waveform feature and the initial feature extraction result, the terminal uses the binary cross entropy function (Cross Entropy Loss, CE Loss) as the loss function. The calculation formula of the loss function L is as follows:

[0110]

[0111] Where T is the length of the one-dimensional vector output by the timing analysis structure, The confidence level for the correct result category when making the i-th prediction for the disease detection model. The goal of this loss function is to minimize the gap between the second waveform features and the expected features to ensure that the model can accurately extract the vibration wave features in the time series analysis structure.

[0112] For the prediction results and detection labels of the training space analysis structure, the terminal calculates the second loss through the cross entropy loss function. The second loss calculation formula is:

[0113]

[0114] Where N is the number of samples at the target location, is the actual detection label of the jth sample (usually 0 or 1), is the predicted probability of the jth vibration wave sample (the output of the spatial analysis structure).

[0115] In this embodiment, the second waveform feature of the sample vibration wave is subjected to feature extraction through the spatial analysis structure to obtain an initial feature extraction result, which causes the feature representation of the second waveform feature to change. Then, the first loss of the timing analysis structure is constructed according to the initial feature extraction result and the second waveform feature, which accurately reflects the accuracy of feature extraction by the timing analysis structure. In combination with the loss value of the spatial analysis structure obtained by the prediction result and the detection label, the disease detection model to be trained is trained in reverse order to ensure that the timing analysis structure and the spatial analysis structure of the disease detection model to be trained can meet the preset loss conditions, avoid local overfitting, and thereby improve the accuracy and reliability of the disease detection model.

[0116] In an exemplary embodiment, Figure 4 As shown, step 206 includes steps 402 to 404. Among them:

[0117] Step 402, back-propagating the spatial analysis structure to be trained according to the second loss, and updating the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss meets the preset loss condition, and outputting the target feature extraction result that meets the preset loss condition.

[0118] In the embodiment of the present application, first, the terminal calculates the gradient of the second loss to the parameters in the spatial analysis structure through the back propagation algorithm, including the weights and biases of each neural network layer. Then, after obtaining the gradient, the terminal uses the gradient descent algorithm to update the first model parameters of the spatial analysis structure, and adjusts the parameters in the opposite direction of the gradient to reduce the second loss. The gradient descent update is repeated until the second loss meets the preset loss condition, indicating that the spatial analysis structure training is completed, and the currently trained spatial analysis model can extract the most influential factors for concrete voiding. Among them, the preset loss condition can be that during the iterative training process, the change rate of the second loss is less than the preset threshold, indicating that the current spatial analysis structure has fully converged, or a loss threshold is set in advance. When the second loss is less than the loss threshold, it means that the loss has been reduced to an acceptable level, or it stops after a certain number of iterations. Finally, the terminal outputs the target feature extraction result that meets the preset loss condition. The target feature extraction result reflects the optimal feature representation learned by the spatial analysis structure during the training process, and the result obtained by extracting the second waveform feature is used as a guiding basis for adjusting the second model parameters of the timing analysis structure.

[0119] Step 404, back-propagation is performed on the time series analysis structure to be trained based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure, until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

[0120] In an embodiment of the present application, the terminal adjusts the second model parameters of the timing analysis structure based on the first loss and gradient descent algorithm, combined with the result obtained by feature extraction of the second waveform features by the spatial analysis structure, so that when the timing analysis structure extracts features from the sample vibration wave, it pays more attention to the optimal feature representation that is concerned about in the feature extraction by the spatial analysis structure, until the first loss of the timing analysis structure meets the preset loss condition, and a trained disease detection model is obtained.

[0121] In this embodiment, by performing back propagation and gradient descent optimization on the spatial analysis structure and the temporal analysis structure, and guiding the adjustment of the second model parameters based on the target feature extraction result obtained from the trained spatial analysis structure, the data processing of the spatial analysis structure and the temporal analysis structure can be jointly trained, and the feature extractions of different structures can be correlated, thereby improving the overall training effect of the disease detection model to be trained, thereby improving the accuracy of disease detection by the disease detection model.

[0122] In an exemplary embodiment, Figure 5 As shown, step 104 includes steps 502 to 504. Among them:

[0123] Step 502: extract features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features.

[0124] In the embodiment of the present application, taking the timing analysis structure as an LSTM structure as an example for explanation, the terminal learns the response vibration wave feature extraction logic during training through the gating mechanism of the LSTM result, extracts features of the response vibration wave, and obtains the timing features of the response vibration wave as the initial waveform features, including changes in peaks, troughs, and amplitude.

[0125] Step 504 , based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure, feature screening is performed on the initial waveform feature to obtain the first waveform feature.

[0126] In the embodiment of the present application, in the process of extracting features from sample vibration waves using the LSTM structure, first, the terminal further processes the initial waveform features according to the LSTM structure. Based on the correlation between the response vibration wave and the target disease learned by the timing analysis structure, the terminal evaluates the importance of the initial waveform features of the response vibration wave at each time step through the forget gate and sigmoid (an activation function for output) layer in the LSTM structure, and determines which factors with less influence need to be forgotten. The input gate then decides to add new timing feature information of the response vibration wave to the cell state, the sigmoid layer guides the information update, and the tanh layer generates a candidate value vector. After the information is updated by the cell state, the output gate uses the sigmoid layer to determine which parts of the cell state should be output, and then processes it through the tanh function to finally obtain the output of the current time step, and obtains the first waveform feature in the form of a one-dimensional vector, which can effectively capture the temporal changes of the timing features of the response vibration wave and the influence relationship on the concrete degassing disease.

[0127] In this embodiment, the response vibration wave data is subjected to time series feature extraction and screening through a time series analysis structure to obtain a first waveform feature that can effectively distinguish the target disease. In the waveform analysis stage, the feature extraction accuracy of the first waveform feature input to the spatial analysis structure is improved, making the first waveform feature more influential relative to the concrete void disease.

[0128] In an exemplary embodiment, Figure 6 As shown, step 106 includes steps 602 to 606. Among them:

[0129] Step 602 , based on the spatial analysis structure of the disease detection model, feature extraction is performed on the first waveform features and target positions corresponding to each response vibration wave, and the extracted features are combined to obtain the spatial features between the response vibration waves of each target position.

[0130] In the embodiment of the present application, the spatial analysis structure is a CNN structure as an example for explanation. Figure 7 As shown, the terminal performs convolution processing on the first waveform feature and the target position of each target position through the convolution layer of the CNN structure, combines the first waveform features of different target positions with the target positions where the response vibration waves corresponding to each first waveform feature are located, and constructs a spatial feature of the correlation between the first waveform features of different target positions. Specifically, CNN extracts spatial features from the first waveform features through operations such as convolution layers, pooling layers, and fully connected layers. The convolution layer can capture local spatial relationships, the pooling layer is used to reduce the dimension of the features and extract the main features, and the fully connected layer is used to integrate the position information of different target positions and combine the position information with the first waveform features.

[0131] Step 604, determining the detection result of each target location according to the correlation between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease.

[0132] In the embodiment of the present application, when a disease exists at a certain target position of the target object, it not only affects the response vibration wave of the target position, but also has an impact on the waveform characteristics of the response vibration waves of other target positions. On this basis, the terminal predicts whether there is a target disease at each target position through a classifier or a regressor based on the mapping relationship between the spatial characteristics of the response vibration waves of multiple target positions and the target disease learned by the spatial analysis structure, and obtains the detection results of each target position.

[0133] Step 606: Determine the disease detection result of the target object based on the detection results of each target position.

[0134] In the embodiment of the present application, the terminal can determine the disease detection result of the target object based on the detection results of each target position, and comprehensively consider the disease detection conditions of all target positions, so as to obtain the disease detection result of the entire target object (for example, a specific part of a concrete structure). Optionally, after obtaining the disease detection result of the target object, the terminal can determine the disease assessment of the target object based on the disease detection result, including overall assessment of the target object based on the number of target diseases and outputting corresponding treatment measures.

[0135] In this embodiment, the first waveform feature of the response vibration wave is extracted through a spatial analysis structure, and the spatial feature is used to achieve accurate detection and positioning of the target disease in the target object.

[0136] In an exemplary embodiment, Figure 8 As shown, step 606 includes steps 802 to 804. Among them:

[0137] Step 802: If the detection results of each target position do not contain the target disease, the disease detection result of the target object is determined as not containing the target disease.

[0138] In an embodiment of the present application, the terminal determines the final detection result of the target object based on the collected disease detection results of all target positions. That is, when the detection results of each target position do not contain the target disease, it means that the overall structure of the target object is intact and there is no concrete void disease. The terminal then determines the final disease detection result of the target object as the absence of the target disease.

[0139] Step 804: If a target disease exists in any target position, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

[0140] In an embodiment of the present application, when the detection result of each target position is that a target disease exists at a certain target position, it means that the overall structure of the target object has defects and there is a concrete void disease. The terminal determines the final disease detection result of the target object as the existence of the target disease, and outputs the target position where the target disease exists for further processing.

[0141] In this embodiment, the final disease detection result of the target object is determined based on the detection results of each target position, which can quickly and effectively evaluate the overall disease condition of the target object and provide a basis and guidance for subsequent maintenance and management decisions.

[0142] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0143] Based on the same inventive concept, the embodiment of the present application also provides a disease detection device for implementing the disease detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more disease detection device embodiments provided below can refer to the limitations of the disease detection method above, and will not be repeated here.

[0144] In an exemplary embodiment, Fig. 9 As shown, a disease detection device 900 is provided, comprising: a first acquisition module 901, a feature extraction module 902 and a correlation analysis module 903, wherein:

[0145] The first acquisition module 901 is used to acquire response vibration waves of multiple target positions in the target object; the response vibration waves are generated by the collision between the striking devices of different materials and the target object;

[0146] The feature extraction module 902 is used to extract and screen the response vibration wave according to the time series analysis structure of the disease detection model, and obtain the first waveform feature corresponding to the response vibration wave at each target position;

[0147] The correlation analysis module 903 is used to perform correlation analysis on the first waveform features corresponding to each response vibration wave at the target position based on the spatial analysis structure of the disease detection model to determine the disease detection result of the target object.

[0148] In one embodiment, the apparatus 900 further includes:

[0149] A second acquisition module is used to acquire multiple training sample sets and sample labels corresponding to each training sample set;

[0150] A prediction module is used to extract features and perform prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each training sample set, so as to obtain the loss value of the disease detection model to be trained;

[0151] The training module is used to perform back propagation on the disease detection model to be trained based on the loss value and the gradient descent algorithm, adjust the model parameters of the disease detection model to be trained, until the loss value meets the preset loss condition, and obtain the trained disease detection model.

[0152] In one embodiment, the training sample set includes vibration wave samples and position information of the vibration wave samples obtained at the target position by using multiple striking devices of different materials, and the sample labels include detection labels indicating whether a target disease exists at each target position.

[0153] In one embodiment, the loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; the prediction module is specifically used to extract features of each vibration wave sample according to the time series analysis structure to be trained of the disease detection model to be trained for each training sample set, so as to obtain a second waveform feature;

[0154] Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each target position;

[0155] A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

[0156] In one embodiment, the training module is specifically used to perform back propagation on the spatial analysis structure to be trained according to the second loss, and update the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss satisfies a preset loss condition, and output a target feature extraction result that satisfies the preset loss condition;

[0157] The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

[0158] In one embodiment, the feature extraction module 902 is specifically used to extract features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features;

[0159] Based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure, the initial waveform features are screened to obtain the first waveform features.

[0160] In one embodiment, the correlation analysis module 903 is specifically used to extract the first waveform features and target positions corresponding to each response vibration wave based on the spatial analysis structure of the disease detection model, and combine the extracted features to obtain the spatial features between the response vibration waves of each target position;

[0161] Determine the detection results of each target location based on the correlation between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease;

[0162] The disease detection result of the target object is determined based on the detection results of each target position.

[0163] In one embodiment, the association analysis module 903 is specifically used to determine the disease detection result of the target object as the absence of the target disease if the detection results of each target position do not have the target disease;

[0164] If a target disease exists in any target position, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

[0165] Each module in the above disease detection device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0166] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.10As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. 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 the computer program in the non-volatile storage medium. The database of the computer device is used to store response vibration waves and target position data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a disease detection method is implemented.

[0167] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0168] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0169] Acquire response vibration waves of multiple target positions in the target object; the response vibration waves are generated by the collision of the striking devices of different materials with the target object;

[0170] Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position;

[0171] Based on the spatial analysis structure of the disease detection model, a correlation analysis is performed on the first waveform features corresponding to each response vibration wave at the target position to determine the disease detection result of the target object.

[0172] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0173] Obtain multiple training sample sets and sample labels corresponding to each training sample set;

[0174] Perform feature extraction and prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each training sample set, and obtain the loss value of the disease detection model to be trained;

[0175] Based on the loss value and the gradient descent algorithm, the disease detection model to be trained is back-propagated, and the model parameters of the disease detection model to be trained are adjusted until the loss value meets the preset loss condition, thereby obtaining a trained disease detection model.

[0176] In one embodiment, the training sample set includes vibration wave samples and position information of the vibration wave samples obtained at the target position by using multiple striking devices of different materials, and the sample labels include detection labels indicating whether a target disease exists at each target position.

[0177] In one embodiment, the loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; when the processor executes the computer program, the following steps are also implemented:

[0178] For each training sample set, extract features of each vibration wave sample according to the time series analysis structure to be trained of the disease detection model to be trained, and obtain a second waveform feature;

[0179] Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each target position;

[0180] A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

[0181] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0182] Back-propagating the spatial analysis structure to be trained according to the second loss, and updating the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss satisfies a preset loss condition, and outputting a target feature extraction result that satisfies the preset loss condition;

[0183] The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

[0184] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0185] The response vibration wave is feature extracted according to the time series analysis structure of the disease detection model to obtain the initial waveform feature;

[0186] Based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure, the initial waveform features are screened to obtain the first waveform features.

[0187] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0188] Based on the spatial analysis structure of the disease detection model, the first waveform features and target positions corresponding to each response vibration wave are extracted, and the extracted features are combined to obtain the spatial features between the response vibration waves of each target position;

[0189] Determine the detection results of each target location based on the correlation between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease;

[0190] The disease detection result of the target object is determined based on the detection results of each target position.

[0191] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0192] If the detection results of each target position do not contain the target disease, the disease detection result of the target object is determined as the target disease not existing;

[0193] If a target disease exists in any target position, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

[0194] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0195] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0196] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0197] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0198] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0199] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A disease detection method, characterized in that: The method comprises: Acquiring response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object; Extracting and screening the features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform features corresponding to the response vibration wave at each target position; Performing correlation analysis on the first waveform features corresponding to the response vibration waves at the target position based on the spatial analysis structure of the disease detection model to determine the disease detection result of the target object; Before obtaining the response vibration waves of multiple target positions in the target object, the method includes: Obtaining multiple training sample sets and sample labels corresponding to each of the training sample sets; Performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each of the training sample sets to obtain a loss value of the disease detection model to be trained; Based on the loss value and the gradient descent algorithm, back-propagation is performed on the disease detection model to be trained, and model parameters of the disease detection model to be trained are adjusted until the loss value meets a preset loss condition, thereby obtaining a trained disease detection model; The training sample set includes vibration wave samples obtained at target positions by a plurality of striking devices of different materials and position information of the vibration wave samples, and the sample labels include detection labels of whether target diseases exist at each of the target positions; The loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; The step of performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample sets and the sample labels corresponding to each of the training sample sets to obtain the loss value of the disease detection model to be trained includes: For each of the training sample sets, extracting features from each of the vibration wave samples according to the time series analysis structure to be trained of the disease detection model to be trained, to obtain a second waveform feature; Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each of the target positions; A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

2. The method according to claim 1, characterized in that The back propagation of the disease detection model to be trained is performed based on the loss value and the gradient descent algorithm, and the model parameters of the disease detection model to be trained are adjusted until the loss value meets the preset loss condition, thereby obtaining a trained disease detection model, including: Back-propagating the spatial analysis structure to be trained according to the second loss, and updating the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss satisfies a preset loss condition, and outputting a target feature extraction result that satisfies the preset loss condition; The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure, until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

3. The method according to claim 1, characterized in that The feature extraction and feature screening of the response vibration wave according to the time series analysis structure of the disease detection model to obtain the first waveform feature corresponding to the response vibration wave at each target position includes: Extracting features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features; The initial waveform feature is subjected to feature screening based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure to obtain a first waveform feature.

4. The method according to claim 1, characterized in that: The spatial analysis structure based on the disease detection model performs correlation analysis on the first waveform features corresponding to each of the response vibration waves at the target position to determine the disease detection result of the target object, including: Extracting features of the first waveform features and the target positions corresponding to each of the response vibration waves based on the spatial analysis structure of the disease detection model, and combining the extracted features to obtain spatial features between the response vibration waves at each of the target positions; Determine the detection result of each target location according to the association relationship between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease; The disease detection result of the target object is determined based on the detection results of each of the target positions.

5. The method according to claim 4, characterized in that The determining the disease detection result of the target object based on the detection results of each of the target positions includes: If the target disease does not exist in the detection results of each of the target positions, determine the disease detection result of the target object as not having the target disease; If the target disease exists in any of the target positions, the disease detection result of the target object is determined as the presence of the target disease, and the target position where the target disease exists is displayed.

6. A disease detection device, characterized in that: The device comprises: A first acquisition module is used to acquire response vibration waves of multiple target positions in a target object; the response vibration waves are generated by collision between a striking device made of different materials and the target object; A feature extraction module, used for extracting and screening features of the response vibration wave according to the time series analysis structure of the disease detection model, to obtain a first waveform feature corresponding to the response vibration wave at each target position; A correlation analysis module, configured to perform correlation analysis on the first waveform features corresponding to the response vibration waves at the target position based on the spatial analysis structure of the disease detection model, so as to determine the disease detection result of the target object; The device also includes: A second acquisition module is used to acquire multiple training sample sets and sample labels corresponding to each of the training sample sets; A prediction module, used for performing feature extraction and prediction processing on the disease detection model to be trained according to the training sample set and the sample labels corresponding to each of the training sample sets, to obtain the loss value of the disease detection model to be trained; A training module, used for performing back propagation on the disease detection model to be trained based on the loss value and the gradient descent algorithm, adjusting the model parameters of the disease detection model to be trained, until the loss value meets the preset loss condition, thereby obtaining a trained disease detection model; The training sample set includes vibration wave samples obtained at target positions by a plurality of striking devices of different materials and position information of the vibration wave samples, and the sample labels include detection labels of whether target diseases exist at each of the target positions; The loss value includes a first loss corresponding to the time series analysis structure to be trained and a second loss corresponding to the space analysis structure to be trained in the disease detection model to be trained; The prediction module is specifically used to extract features of each of the vibration wave samples according to the time series analysis structure to be trained of the disease detection model to be trained for each of the training sample sets, so as to obtain a second waveform feature; Based on the to-be-trained spatial analysis structure of the to-be-trained disease detection model, feature extraction and prediction processing are performed on the second waveform feature and the target position corresponding to the second waveform feature to obtain an initial feature extraction result and a prediction result corresponding to each of the target positions; A first loss of the time series analysis structure to be trained is determined according to the second waveform feature and the initial feature extraction result, and a second loss of the space analysis structure to be trained is determined according to the prediction result and the detection label.

7. The device according to claim 6, characterized in that The training module is specifically used to perform back propagation on the spatial analysis structure to be trained according to the second loss, and update the first model parameters of the spatial analysis structure to be trained according to the gradient descent algorithm until the second loss meets the preset loss condition, and output the target feature extraction result that meets the preset loss condition; The time series analysis structure to be trained is back-propagated based on the first loss, and the second model parameters of the time series analysis structure to be trained are updated according to the target feature extraction results of the gradient descent algorithm and the spatial analysis structure, until the first loss meets the preset loss condition, thereby obtaining a trained disease detection model.

8. The device according to claim 6, characterized in that The feature extraction module is specifically used to extract features of the response vibration wave according to the time series analysis structure of the disease detection model to obtain initial waveform features; The initial waveform feature is subjected to feature screening based on the correlation between the response vibration wave and the target disease learned by the time series analysis structure to obtain a first waveform feature.

9. The device according to claim 6, characterized in that The association analysis module is specifically used to extract the first waveform features corresponding to each of the response vibration waves and the target position based on the spatial analysis structure of the disease detection model, and combine the extracted features to obtain the spatial features between the response vibration waves of each of the target positions; Determine the detection result of each target location according to the association relationship between the spatial features learned by the spatial analysis structure and the target disease and the location of the target disease; The disease detection result of the target object is determined based on the detection results of each of the target positions.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

Images