Construction method of water seepage detection model, water seepage detection method, device and electronic equipment

Abstract

The invention relates to the technical field of environment monitoring, and discloses a construction method of a water seepage detection model, a water seepage detection method, a device and electronic equipment. The construction method of the water seepage detection model comprises the steps of obtaining a training image set; the training image set is a temperature image set comprising a water seepage temperature image of the to-be-detected area in a water seepage state and a normal temperature image of the to-be-detected area in a normal state; classifying the temperature images in the training image set according to the time stamp information of the temperature images to obtain a plurality of sub-image training sets; training a plurality of pre-constructed sub-detection models based on the plurality of sub-image training sets to obtain a plurality of trained first detection models; and fusing the plurality of trained first detection models to obtain a water seepage detection model. According to the invention, the capturing capability of the water seepage detection model on the water seepage characteristics in different time periods is improved, so that the water seepage detection precision is improved.

Description

Technical Field

[0001] This application relates to the technical field of environmental monitoring, for example, to a method for constructing a water seepage detection model, a water seepage detection method, a device, and an electronic device. Background Art

[0002] For some places, such as factories or power distribution rooms equipped with a large number of electrical equipment, warehouses for storing and keeping materials, etc., it is crucial to conduct building water seepage detection to ensure electrical safety and property safety.

[0003] The related art discloses a method for predicting anti-seepage and water leakage based on a distributed temperature measurement optical fiber system, including: determining the start time and end time corresponding to a leakage event according to historical monitoring data, and determining the leakage time interval and the leakage state corresponding to the leakage time interval according to the start time and the end time, determining first temperature data according to the leakage time interval and historical temperature data, where the historical monitoring data includes monitoring data on the leakage situation of the target object before the current moment, the historical temperature data is monitoring data on the temperature change of the target object, and the first temperature data is the historical temperature data with the same acquisition period as the leakage time interval; constructing an alternative LSTM model, and constructing a training data set according to each of the first temperature data and the corresponding leakage state, training the alternative LSTM model based on the training data set to obtain a target LSTM model, where the alternative LSTM model is used to predict the corresponding leakage state according to the temperature change of the target object; obtaining second temperature data, and inputting the second temperature data into the target LSTM model to obtain a predicted leakage state, where the second temperature data is real-time collected temperature data.

[0004] Although the related art realizes the pre-judgment of anti-seepage and water leakage by considering the continuity of time and space, the accuracy of this prediction method depends on the accuracy of the temperature parameters measured by the distributed temperature measurement optical fiber system. And the temperature measurement accuracy of the distributed temperature measurement optical fiber system is easily affected by various factors, such as the position of the optical fiber, system noise, spatial resolution, etc. The insufficient temperature measurement accuracy is likely to lead to misjudgment or missed judgment of leakage events. Therefore, the accuracy rate of the prediction method in the related art is relatively low. Summary of the Invention

[0005] To have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important constituent elements or delineate the protection scope of these embodiments, but rather serves as a preface to the subsequent detailed description.

[0006] Embodiments of the present disclosure provide a method for constructing a water seepage detection model, a water seepage detection method, an apparatus, and an electronic device, which improve the ability of the water seepage detection model to capture water seepage characteristics at different time periods, thereby improving the water seepage detection accuracy.

[0007] In some embodiments, a method for constructing a water seepage detection model is provided, including: obtaining a training image set; the training image set is a temperature image set including water seepage temperature images and normal temperature images in a normal state of the area to be detected; classifying the temperature images in the training image set according to the time marking information of the temperature images to obtain a plurality of sub-image training sets; training a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets to obtain a plurality of trained first detection models; and fusing the plurality of trained first detection models to obtain a water seepage detection model.

[0008] Optionally, the pre-constructed sub-detection model includes a convolutional neural network; training a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets includes: preprocessing the temperature images in each sub-image training set; the preprocessing includes illumination correction and / or contrast enhancement; and inputting the preprocessed temperature images into the convolutional neural network for training.

[0009] Optionally, the pre-constructed sub-detection model includes a long short-term memory network; training a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets includes: extracting temperature features of the temperature images in each sub-image training set; sorting the extracted temperature features according to the time marking information of the temperature images to obtain a target temperature feature sequence; and inputting the target temperature feature sequence into the long short-term memory network for training.

[0010] Optionally, the pre-constructed sub-detection model includes a convolutional neural network and a long short-term memory network; training a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets includes: obtaining a prediction result of the area to be detected at the prediction moment output by the long short-term memory network; obtaining an actual measurement result of the area to be detected output by the convolutional neural network corresponding to the temperature image at the prediction moment; when the correlation coefficient between the prediction result and the actual measurement result is less than the coefficient threshold, determining a target sub-detection model from the plurality of pre-constructed sub-detection models and optimizing the target sub-detection model.

[0011] Optionally, when the target sub-detection model includes a long short-term memory network, optimizing the target sub-detection model includes: calculating the mean square error value of the long short-term memory network; updating the model parameters of the long short-term memory network according to the mean square error value, and recalculating the mean square error value.

[0012] Optionally, the training image set further includes the patched temperature image after the water seepage repair of the area to be detected; fusing a plurality of trained first detection models to obtain a water seepage detection model, including: training a pre-constructed sub-detection model based on the patched temperature image to obtain a trained second detection model; fusing a plurality of trained first detection models and the second detection model to obtain a water seepage detection model.

[0013] In some embodiments, a water seepage detection method is provided, including: obtaining a real-time temperature image of the area to be detected; determining a target first detection model in the water seepage detection model according to the time information of the real-time temperature image; wherein, the water seepage detection model is constructed by using the construction method of the water seepage detection model described in the above embodiments; adjusting the target first detection model to an active state and adjusting the non-target first detection models to a deactivated state; inputting the real-time temperature image into the adjusted water seepage detection model to obtain a detection result.

[0014] Optionally, when the detection result is a water seepage state, the water seepage detection method further includes: extracting feature points and brightness information from the real-time temperature image; transforming the real-time temperature image according to the feature points and brightness information to obtain a transformed image; comparing the transformed image with the real-time temperature image to obtain and output an image of the area to be patched.

[0015] In some embodiments, a device for constructing a water seepage detection model is provided, including: an acquisition module configured to acquire a training image set; the training image set is a temperature image set including water seepage temperature images and normal temperature images of the area to be detected in a water seepage state and a normal state respectively; a classification module configured to classify the temperature images in the training image set according to the time marking information of the temperature images to obtain a plurality of sub-image training sets; a training module configured to train a plurality of pre-constructed sub-detection models based on the plurality of sub-image training sets respectively to obtain a plurality of trained first detection models; a fusion module configured to fuse the plurality of trained first detection models to obtain a water seepage detection model.

[0016] In some embodiments, a water seepage detection device is provided, including: a collection module configured to obtain a real-time temperature image of the area to be detected; an analysis module configured to determine a target first detection model in the water seepage detection model according to the time information of the real-time temperature image; wherein, the water seepage detection model is constructed by using the construction method of the water seepage detection model described in the above embodiments; an adjustment module configured to adjust the target first detection model to an active state and adjust the non-target first detection models to a deactivated state; a detection module configured to input the real-time temperature image into the adjusted water seepage detection model to obtain a detection result.

[0017] In some embodiments, an electronic device is provided, including: a memory storing program instructions; and a processor configured to execute the method for constructing a water seepage detection model as described in the above embodiments and / or the water seepage detection method as described in the above embodiments when running the program instructions.

[0018] The method for constructing a water seepage detection model, the water seepage detection method, the device and the electronic device provided by the embodiments of the present disclosure can achieve the following technical effects:

[0019] The method for constructing a water seepage detection model provided by the embodiments of the present disclosure can obtain a temperature image of a region to be detected as a training image set, and classify the temperature image according to time marking information to obtain a sub-image training set, so that each sub-image training set corresponds to a specific time period. Using the classified multiple sub-image training sets, multiple pre-constructed sub-detection models are respectively trained, so that the corresponding sub-detection models can learn the water seepage characteristics of a specific time period. By training and fusing multiple sub-detection models, the feature information of the sub-image training sets in different time periods can be fully utilized to improve the accuracy of water seepage detection of the water seepage detection model. Each sub-model can learn specific water seepage characteristics, and the fused water seepage detection model can more comprehensively capture the water seepage characteristics in different time periods, thereby improving the water seepage detection accuracy.

[0020] In the embodiments of the present disclosure, the temperature image is used as training data. The temperature image shows the temperature distribution on the surface of an object through different colors or gray levels, and can intuitively show the temperature difference. Compared with the temperature parameters measured by the distributed temperature measurement optical fiber system in the related art, it is not affected by factors such as spatial resolution and measurement position. When the water seepage detection model in the embodiments of the present disclosure is applied to the process of water seepage detection, the stability and reliability are higher. In addition, since the water seepage detection model in the embodiments of the present disclosure includes multiple trained first detection models, and each first detection model learns the water seepage characteristics of a specific time period, the characteristic changes of the temperature images of the region to be detected in different time periods are considered, the influence of the time period on the water seepage event is considered, and the ability of the water seepage detection model to capture the water seepage characteristics in different time periods is improved, thereby improving the water seepage detection accuracy.

[0021] The above general description and the following description are only exemplary and explanatory, and are not used to limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] One or more embodiments are exemplarily illustrated by corresponding drawings. These exemplary illustrations and the drawings do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements. The drawings do not constitute a scale limitation, and wherein:

[0023] Figure 1Schematic diagram of an electronic device provided by an embodiment of the present disclosure;

[0024] Figure 2 Schematic diagram of a controller provided by an embodiment of the present disclosure;

[0025] Figure 3 Schematic diagram of a method for constructing a water seepage detection model provided by an embodiment of the present disclosure;

[0026] Figure 4 Schematic diagram of a method for constructing a water seepage detection model provided by another embodiment of the present disclosure;

[0027] Figure 5 Schematic diagram of a water seepage detection method provided by an embodiment of the present disclosure;

[0028] Figure 6 Schematic diagram of a device for constructing a water seepage detection model provided by an embodiment of the present disclosure;

[0029] Figure 7 Schematic diagram of a water seepage detection device provided by an embodiment of the present disclosure. Detailed implementation manners

[0030] In order to be able to understand the features and technical content of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure will be described in detail below in conjunction with the accompanying drawings. The attached drawings are only for reference and explanation purposes, and are not used to limit the embodiments of the present disclosure. In the following technical description, for the convenience of explanation, multiple details are provided to provide a full understanding of the disclosed embodiments. However, one or more embodiments can still be implemented without these details. In other cases, well-known structures and devices can be shown in a simplified manner to simplify the drawings.

[0031] In the embodiments of the present disclosure, terms such as "first" and "second" in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to implement the embodiments of the present disclosure described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion.

[0032] Unless otherwise specified, the term "plurality" means two or more.

[0033] In the embodiments of the present disclosure, the character " / " indicates that the objects before and after are in an "or" relationship. For example, A / B means: A or B.

[0034] The term "and / or" is an associative relationship describing objects, indicating that three relationships can exist. For example, A and / or B means: A, B, and A and B these three relationships.

[0035] The term "corresponding" may refer to an association relationship or a binding relationship. That A corresponds to B means there is an association relationship or a binding relationship between A and B.

[0036] It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments may be combined with each other.

[0037] Combined with Figure 1 As shown, the embodiments of the present disclosure provide an electronic device 1, including: a device body 10 and a controller 20. The controller 20 is installed on the device body 10.

[0038] In the embodiments of the present disclosure, the controller 20 is installed on the device body 10. The installation relationship described here is not limited to being placed inside the device body 10, but also includes the installation connection with other components of the electronic device 1, including but not limited to physical connection, electrical connection, or signal transmission connection, etc. Those skilled in the art can understand that the controller 20 can be adapted to a feasible electronic device 1, and thus other feasible embodiments can be realized. In some embodiments, the electronic device 1 includes a robot or a computer, etc. The specific types of the electronic device 1 are not limited in this application.

[0039] Optionally, combined with Figure 2 As shown, the controller 20 includes a processor 200. The processor 200 can obtain a training image set; the training image set is a temperature image set including a seepage temperature image and a normal temperature image of the area to be detected in a seepage state and a normal state respectively; it can classify the temperature images in the training image set according to the time marking information of the temperature images to obtain a plurality of sub-image training sets; it can train a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets to obtain a plurality of trained first detection models; it can fuse the plurality of trained first detection models to obtain a seepage detection model.

[0040] Combined with Figure 1 and Figure 2 As shown, the embodiments of the present disclosure provide a method for constructing a seepage detection model. As Figure 3 shown, the construction method includes:

[0041] S301, the processor obtains a training image set.

[0042] The training image set is a temperature image set including a seepage temperature image and a normal temperature image of the area to be detected in a seepage state and a normal state respectively.

[0043] In this step, the area to be detected refers to a specific spatial range that needs to be detected, such as the wall or floor of a factory workshop.

[0044] In S302, the processor classifies the temperature images in the training image set according to the time stamp information of the temperature images, and obtains multiple sub-image training sets.

[0045] In this step, the temperature images in the training image set are classified according to the time stamp information, so that the temperature images are organized into different subsets, that is, sub-image training sets. Each sub-image training set corresponds to a specific time period. Among them, each sub-image training set includes seepage temperature images and normal temperature images.

[0046] In this step, classifying the temperature images in the training image set according to the time stamp information of the temperature images includes: classifying the temperature images in the training image set according to the time stamp information of the temperature images according to a preset classification rule. Among them, the preset classification rule is preset by the technician according to the actual geographical location or task requirements of the area to be detected, and is not specifically limited in this application.

[0047] Exemplarily, the time stamp (time mark) of a certain temperature image in the training image set is "XXXX03010459", then "XXXX03010459" represents 4:59 on March 1st, "XXXX" year; the time stamp of a certain temperature image is "XXXX11211859", then "XXXX11211859" represents 18:59 on November 21st, "XXXX" year. The preset classification rule is: first classify the temperature images into 4 major categories according to seasons (spring from March to May, summer from June to August, autumn from September to November, winter from December to February of the next year), and then continue to classify the 4 major categories into 3 small categories according to three time periods of morning, middle, and evening. Among them, the three time periods of morning, middle, and evening correspond to the time periods of 4:00 - 10:59, 11:00 - 19:59, and 20:00 - 3:59 of the next day respectively. Then, according to the preset classification rule, the temperature images in the training image set are divided into 12 groups, a total of 12 sub-image training sets.

[0048] In S303, the processor trains multiple pre-constructed sub-detection models based on multiple sub-image training sets respectively, and obtains multiple trained first detection models.

[0049] In S304, the processor fuses multiple trained first detection models to obtain a seepage detection model.

[0050] In this step, the fusion of multiple trained first detection models can be performed in a stacked or parallel manner to activate one or more first detection models for seepage detection if necessary.

[0051] The method for constructing a water seepage detection model provided by the embodiments of the present disclosure can obtain the temperature image of the area to be detected as the training image set, and classify the temperature images according to the time marking information to obtain the sub-image training sets, so that each sub-image training set corresponds to a specific time period. Using the classified multiple sub-image training sets, train multiple pre-constructed sub-detection models respectively, so that the corresponding sub-detection models can learn the water seepage characteristics of a specific time period. By training multiple sub-detection models and fusing them, the characteristic information of the sub-image training sets in different time periods can be fully utilized to improve the accuracy of water seepage detection of the water seepage detection model. Each sub-model can learn specific water seepage characteristics, and the fused water seepage detection model can capture the water seepage characteristics in different time periods more comprehensively, thereby improving the water seepage detection accuracy.

[0052] In the embodiments of the present disclosure, the temperature image is used as the training data. The temperature image shows the temperature distribution on the surface of the object through different colors or gray levels, and can intuitively show the temperature difference. Compared with the temperature parameters measured by the distributed temperature measurement optical fiber system in the related art, it is not affected by factors such as spatial resolution and measurement position. When the water seepage detection model in the embodiments of the present disclosure is applied to the process of water seepage detection, the stability and reliability are higher. In addition, since the water seepage detection model in the embodiments of the present disclosure includes multiple trained first detection models, and each first detection model learns the water seepage characteristics of a specific time period, the characteristic changes of the temperature images of the area to be detected in different time periods are considered, and the influence of the time period on the water seepage event is considered, so as to improve the ability of the water seepage detection model to capture the water seepage characteristics in different time periods, thereby improving the water seepage detection accuracy.

[0053] In some embodiments, the pre-constructed sub-detection model includes a convolutional neural network and / or a long short-term memory network.

[0054] Convolutional Neural Networks (CNN) are good at extracting local and global features from images. A convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is used to extract local features in the image, the pooling layer is used to reduce the dimension of the data, and the fully connected layer is used for classification tasks. The convolutional neural network can automatically learn the key features in the temperature image, such as temperature distribution, the shape of the water seepage area, etc. The trained convolutional neural network can classify the input water seepage temperature image and normal temperature image, so as to judge whether it belongs to the water seepage state.

[0055] In some embodiments, the pre - constructed sub - detection model includes a convolutional neural network. Training multiple pre - constructed sub - detection models respectively based on multiple sub - image training sets includes: pre - processing the temperature images in each sub - image training set; the pre - processing includes illumination correction and / or contrast enhancement; inputting the pre - processed temperature images into the convolutional neural network for training.

[0056] In this embodiment, by pre - processing the temperature images, the quality of the temperature images is improved, making the temperature images clearer and easier to analyze, thereby reducing false alarms and missed detections and further improving the accuracy of water seepage detection. Among them, illumination correction is used to adjust the illumination conditions in the temperature images to make the illumination conditions more uniform and consistent. Through illumination correction, the non - uniformity of illumination is eliminated or reduced, making the water seepage features in the temperature images more obvious and easier to identify. Contrast enhancement is used to increase the brightness difference between different regions in the temperature images, thereby enhancing the details and clarity of the temperature images. Through contrast enhancement, the temperature difference between the water seepage area and the normal area can be magnified, making the water seepage features more prominent and easier to detect.

[0057] Optionally, the pre - constructed convolutional neural network includes a convolutional layer, a ReLU activation function, a pooling layer, and a fully - connected layer. Then inputting the pre - processed temperature images into the convolutional neural network for training includes: inputting the pre - processed temperature images into the convolutional layer for temperature distribution feature extraction: where I′ represents the pre - processed temperature image, is the convolutional kernel (weight), used to extract the temperature distribution features in the temperature image, is the bias, represents the feature map output by the convolutional layer, which can emphasize specific temperature distribution patterns on the temperature image; performing ReLU activation function processing on the feature map output by the convolutional layer: where, represents the output after ReLU activation function processing, represents for each element in if then the corresponding element in is itself, if then the corresponding element in is 0; inputting the feature map after ReLU activation function processing into the pooling layer for pooling: where P WDM represents the pooled feature map, and the pooled feature map reduces the number of parameters and the amount of computation; inputting the pooled feature map into the fully - connected layer for recognition: where, respectively represent the weights and biases of the fully connected layer; Y WDM represents the recognition result of the convolutional neural network.

[0058] The Long Short-Term Memory (LSTM) network is a variant of the Recurrent Neural Networks (RNN) and is particularly good at processing time series data. Recurrent neural networks can process sequential data, that is, there is a temporal or spatial dependence between data. Recurrent neural networks capture dynamic information in the sequence by using recurrent connections to take the output of the previous moment as the input of the next moment. The long short-term memory network controls the flow of information by introducing a gating mechanism (input gate, forget gate, and output gate), which can solve the deficiencies of recurrent neural networks in long-term dependence problems. In seepage detection, the long short-term memory network can utilize the time series information of temperature images to capture the evolution law of seepage events over time, so as to predict future seepage states.

[0059] In some embodiments, the pre-constructed sub-detection model includes a long short-term memory network. Training multiple pre-constructed sub-detection models respectively based on multiple sub-image training sets includes: extracting temperature features from the temperature images in each sub-image training set; sorting the extracted temperature features according to the time marking information of the temperature images to obtain a target temperature feature sequence; inputting the target temperature feature sequence into the long short-term memory network for training.

[0060] In this embodiment, by sorting the extracted temperature features according to the time marking information of the temperature images to form a target temperature feature sequence, it is ensured that the long short-term memory network can process the input data in chronological order, thereby capturing the change law of temperature features over time, and accurately predicting future states.

[0061] Optionally, extracting temperature features from the temperature images in each sub-image training set includes: extracting the temperature feature T in the temperature image based on Wien's law: where b is the Wien displacement constant, b = 2.898×10 -3 m·K, λ max represents the wavelength value corresponding to the pixel point of the collected temperature image, which can be obtained by the instrument for collecting the temperature image. In the temperature image, each pixel point corresponds to a radiation intensity. In the case of knowing the peak wavelength λ max of the radiation intensity, the temperature of this pixel point can be calculated using Wien's law to obtain the temperature distribution information in the temperature image.

[0062] The temperature feature extraction method based on Wien's law can accurately reflect the temperature distribution information in the temperature image, improving the sensitivity and accuracy of the water seepage detection model to temperature features. In this embodiment, by extracting accurate temperature features, the water seepage detection model can better learn the complex relationship between water seepage events and temperature distribution, enabling the water seepage detection model to have stronger generalization ability when facing different environmental conditions and different water seepage situations.

[0063] Optionally, sort the extracted temperature features according to the time stamp information of the temperature image to obtain a target temperature feature sequence, including: sorting the extracted temperature features according to the time stamp information of the temperature image to obtain an initial temperature feature sequence; performing an average process on the temperature features within the window according to a sliding window on the initial temperature feature sequence to obtain a target temperature feature sequence.

[0064] In this embodiment, first, according to the time stamp information of each temperature image, sort the extracted temperature features to generate an initial temperature feature sequence to ensure that the temperature feature sequence is arranged in chronological order, thus reflecting the real process of temperature feature change. Then, slide a window on the initial temperature feature sequence. The step size and size of the window can be preset and adjusted by technicians according to the analysis requirements. The temperature features within the window will be considered as a whole to capture the temperature change features within a local time period. By performing an average process on the temperature features within the window to smooth out some accidental temperature fluctuations, while generating a target temperature feature sequence, the accuracy of subsequent analysis is improved.

[0065] Exemplarily, the size of the window is 7, the step size of the window is 1, and the initial temperature feature sequence is [T0, T1, T2, T3,..., T 29 . By sliding the window, calculate the target temperature feature sequence: the first window is "T0, T1, T2, T3, T4, T5, T6", and perform an average process on the temperature features within the window: W1 = mean(T0, T1, T2, T3, T4, T5, T6); the second window is "T1, T2, T3, T4, T5, T6, T7", and perform an average process on the temperature features within the window: W2 = mean(T1, T2, T3, T4, T5, T6, T7); the third window is "T2, T3, T4, T5, T6, T7, T8", and perform an average process on the temperature features within the window: W3 = mean(T2, T3, T4, T5, T6, T7, T8)... and so on until the last window, generating the target temperature feature sequence as [W1, W2, W3,..., W 23 . Input the target temperature feature sequence [W1, W2, W3,..., W 23 into the long short-term memory network for training.

[0066] In a practical application, the long short-term memory network includes an input gate, a forget gate, and an output gate. The forget gate is used to determine which information will be discarded from the cell state. The input gate is used to determine which new information will be added to the cell state. The output gate is used to determine which information will be output from the cell state to the hidden state.

[0067] The specific process of inputting the target temperature feature sequence into the long short-term memory network for training is as follows: forget gate: Among them, represents the output of the forget gate, σ represents the sigmoid activation function, represents the input weight of the forget gate, represents the bias of the forget gate, represents the hidden state at the previous moment (t - 1), represents the input at the current moment t (in this embodiment, can be the temperature feature in the target temperature feature sequence), which together constitute the information received by the cell unit of the long short-term memory network; input gate: Among them, represents the output of the input gate, W i WPM and respectively represent the input weight and bias of the input gate; Among them, represents the new cell candidate value, that is, the new information that can be updated to the cell state, and respectively represent the input weight and bias of the cell candidate value; Among them, represents the cell state at the previous moment (t - 1), represents the cell state at the current moment t, which contains all the information meaningful for predicting the future state. * represents element-wise multiplication, which is used to combine the results of the forget gate and the input gate to update the cell state; output gate: Among them, represents the output of the output gate, and respectively represent the input weight and bias of the output gate; Among them, represents the hidden state at the current moment t, which can combine the output gate and the cell state to calculate the current hidden state, It contains all the information that needs to be passed to the next time step, as well as the information for predicting the final future state.

[0068] In some embodiments, the pre-built sub-detection model includes a convolutional neural network and a long short-term memory network. For the separate training processes of the convolutional neural network and the long short-term memory network, refer to the above embodiments and will not be elaborated here.

[0069] In some embodiments, when the pre-built sub-detection model includes a convolutional neural network and a long short-term memory network, training the multiple pre-built sub-detection models based on multiple sub-image training sets further includes: obtaining the prediction result of the region to be detected output by the long short-term memory network at the prediction time; obtaining the measured result of the region to be detected output by the convolutional neural network corresponding to the temperature image at the prediction time; when the correlation coefficient between the prediction result and the measured result is less than the coefficient threshold, determining the target sub-detection model from the multiple pre-built sub-detection models and optimizing the target sub-detection model.

[0070] In this embodiment, the target sub-detection model refers to the sub-detection model that needs to be optimized among the multiple pre-built sub-detection models, and the target sub-detection model may include a convolutional neural network and / or a long short-term memory network. This embodiment can mutually verify by combining the output results of the long short-term memory network and the convolutional neural network for model optimization, realizing the dynamic adjustment of the water seepage detection model and improving the accuracy of the water seepage detection model. The optimization process of the target sub-detection model includes but is not limited to adjusting model parameters, increasing training data, improving data preprocessing, or enhancing the model structure, etc.

[0071] Exemplarily, m > s, s > 0. Taking the temperature images in the time period from t m-s to t m as the input, obtaining the prediction result output by the long short-term memory network, and the prediction result represents the future state of the region to be detected predicted by the long short-term memory network at the prediction time t m+s . Taking the temperature image at time t m+s in the training image set as the input of the convolutional neural network, obtaining the measured result output by the convolutional neural network, and the measured result represents the t m+sThe real-time state at a moment. Calculate the correlation coefficient between the predicted result and the measured result to evaluate the consistency or similarity between the two. The correlation coefficient can be the Pearson correlation coefficient or the Spearman rank correlation coefficient. When the correlation coefficient is less than the preset coefficient threshold, it is considered that the difference between the predicted result and the measured result is large, and the target sub-detection model needs to be optimized. The target sub-detection model refers to the sub-detection model that is currently being evaluated or used and needs to be optimized. It can be a convolutional neural network, a long short-term memory network, or a combination of a convolutional neural network and a long short-term memory network.

[0072] It should be noted that the specific value of the coefficient threshold is preset by the technical personnel according to the actual task requirements and is not specifically limited in this application.

[0073] Optionally, obtain the correlation coefficient R between the predicted result and the measured result in the following way: where, Y WDM represents the measured result, represents the average value of the measured results, Y WPM represents the predicted result, represents the average value of the predicted results. Through the calculation formula of the correlation coefficient R in this embodiment, the association between the predicted result and the measured result is realized to accurately evaluate the consistency or similarity between the predicted result and the measured result, thereby improving the accuracy of the water seepage detection model.

[0074] Optionally, determine the target sub-detection model from multiple pre-constructed sub-detection models, including: obtaining the status marking information of the temperature image at the prediction moment; in the case where the status marking information is the same as the measured result and different from the predicted result, use the long short-term memory network in the multiple pre-constructed sub-detection models as the target sub-detection model; in the case where the status marking information is different from the measured result and the same as the predicted result, use the convolutional neural network in the multiple pre-constructed sub-detection models as the target sub-detection model; in the case where the status marking information is different from both the measured and predicted results, use the long short-term memory network and the convolutional neural network in the multiple pre-constructed sub-detection models as the target sub-detection model.

[0075] In this embodiment, the status marking information may include the classification label of the temperature image (such as normal state or water seepage state), the description of the water seepage state (such as the water seepage location), etc. By retrieving the status marking information of the temperature image at the prediction moment, the target sub-detection model that needs to be optimized can be dynamically selected according to the actual situation, thereby improving the pertinence and efficiency of model optimization.

[0076] Optionally, when the target sub-detection model includes a long short-term memory network, optimizing the target sub-detection model includes: calculating the mean squared error value of the long short-term memory network; updating the model parameters of the long short-term memory network according to the mean squared error value, and recalculating the mean squared error value.

[0077] In this embodiment, by calculating the mean squared error value (MSE) of the long short-term memory network, the deviation degree between the prediction result of the long short-term memory network and the true result (the true state represented by the temperature image at the prediction moment, that is, the state label information) is measured. The smaller the mean squared error value, the better the prediction performance of the long short-term memory network. In the optimization iteration process, the model parameters of the long short-term memory network are continuously updated to minimize the mean squared error value. When the mean squared error value no longer decreases significantly, it can be considered that a better combination of model parameters has been found, and the optimization of the long short-term memory network is achieved.

[0078] In a practical application, the mean squared error value MSE of the long short-term memory network is calculated in the following manner: where Y i WSM represents the true result represented by the temperature image at the prediction moment, and Y i WPM represents the prediction result output by the long short-term memory network, and n represents the number of temperature images used to train the long short-term memory network.

[0079] In a practical application, updating the model parameters of the long short-term memory network according to the mean squared error value includes: where W t represents the model parameters in the current iteration step, and W t+1 represents the model parameters in the next iteration step, α represents the learning rate, which is a positive number used to control the step size of weight update in each iteration, represents the gradient of the mean squared error value with respect to the current weight W t That is, the loss amount. By continuously iteratively updating the model parameters, the prediction performance of the long short-term memory network will gradually improve. Because each update is based on the gradient of the current mean squared error value, it can ensure that the long short-term memory network moves in the direction of reducing the error, ensuring the continuity and effectiveness of the training process.

[0080] Optionally, training multiple pre-constructed sub-detection models based on multiple sub-image training sets respectively includes: preprocessing the temperature images in the sub-image training set; the preprocessing includes illumination correction and / or contrast enhancement; inputting the preprocessed temperature images into the pre-constructed sub-detection models for forward propagation and backward propagation.

[0081] In this embodiment, the temperature images in the sub-image training set are preprocessed first to improve the quality of the temperature images so that the information in the temperature images is clearer and more accurate, so that the sub-detection model can better capture the temperature features, thereby improving the model accuracy. During the training process, the model first calculates the detection results through forward propagation and calculates the gradient of the loss function through backward propagation, thereby updating the model parameters and performing iterative training of the model until the performance of the model reaches the preset standard or converges.

[0082] Combined with Figure 4 As shown, the present disclosure provides another method for constructing a water seepage detection model, including:

[0083] S401, the processor obtains the training image set.

[0084] The training image set is a temperature image set including water seepage temperature images and normal temperature images in which the area to be detected is in a water seepage state and a normal state respectively, and repaired temperature images after the water seepage repair of the area to be detected.

[0085] In this step, the repaired temperature images are temperature images of the area to be detected with different repair materials (such as polymer cement-based, cement-based grouting materials, polyurethane sealants, etc.) that are repaired well and not repaired well. The area to be detected being repaired well can be understood as the area to be detected where a water seepage event has occurred or is about to occur is successfully repaired to change the state of the ongoing or impending water seepage event. The area to be detected not being repaired well can be understood as the area to be detected where a water seepage event has occurred or is about to occur is repaired unsuccessfully and the state of the ongoing or impending water seepage event is not changed.

[0086] S402, the processor classifies the temperature images in the training image set according to the time stamp information of the temperature images to obtain multiple sub-image training sets.

[0087] S403, the processor trains multiple pre-constructed sub-detection models based on multiple sub-image training sets respectively to obtain multiple trained first detection models.

[0088] S404, the processor trains the pre-constructed sub-detection model based on the repaired temperature images to obtain a trained second detection model.

[0089] In this step, the pre-constructed sub-detection models include convolutional neural networks and / or long short-term memory networks.

[0090] S405, the processor fuses multiple trained first detection models and the second detection model to obtain a water seepage detection model.

[0091] In this step, the fusion of multiple trained first detection models and second detection models can be performed in a stacked or parallel manner to activate one or more first detection models and / or second detection models for water seepage detection when needed.

[0092] In the method for constructing the water seepage detection model provided by the embodiments of the present disclosure, in the original training image set, a repaired temperature image after water seepage repair in the area to be detected is added. The repaired temperature image provides temperature distribution information after water seepage repair, so that the model can learn the feature changes after water seepage repair, thus forming a complement with the first detection model. After the first detection model and the second detection model are fused, the water seepage detection model can more accurately identify water seepage repair events, reduce the probability of misidentifying water seepage repair events as water seepage events, and further improve the accuracy of the water seepage detection model. In addition, the repaired temperature image is a temperature image of the area to be detected including different repair materials with good repair and un-repaired areas. Based on the temperature images of the well-repaired and un-repaired areas to be detected, the second detection model is trained, so that the water seepage detection model can also determine the water seepage repair state, that is, whether the water seepage repair is successful or failed, improving the functionality of the water seepage detection model.

[0093] In the embodiments of the present disclosure, for the specific process of training a pre-constructed sub-detection model based on the repaired temperature image to obtain a trained second detection model, refer to the specific process of training multiple pre-constructed sub-detection models based on multiple sub-image training sets respectively to obtain multiple trained first detection models in the above embodiments, which will not be elaborated here.

[0094] In the embodiments of the present disclosure, during the process of training a pre-constructed sub-detection model based on the repaired temperature image, when the pre-constructed sub-detection model includes a convolutional neural network, based on the trained second detection model, the water seepage detection model can identify water seepage events and repair events. In practical applications, both water seepage events and repair events will cause temperature changes on the area to be detected (such as a wall surface), especially the temperature difference changes with the surrounding area. By accurately distinguishing water seepage events and repair events, the probability of misjudging repair events as water seepage events can be avoided, and the accuracy of the water seepage detection model can be further improved.

[0095] In the embodiments of the present disclosure, during the process of training a pre-constructed sub-detection model based on a patched temperature image, when the pre-constructed sub-detection model includes a long short-term memory network, based on the trained second detection model, the water seepage detection model can realize the pre-judgment of the patching state for identifying patching events. In practical applications, when a water seepage event occurs in the area to be detected, the temperature difference between the water seepage position and the surrounding area will gradually increase. After the water seepage at the water seepage position is patched and the patching is successful, the temperature at the patched position (water seepage position) will gradually tend to the temperature of the surrounding area, that is, the temperature difference between the patched position and the surrounding area will gradually decrease. By analyzing the temperature of the patched position (water seepage position) and the surrounding area through the second detection model, while accurately distinguishing between water seepage events and patching events, the pre-judgment of the patching state of the patching event is realized, and the functionality of the water seepage detection model is improved.

[0096] Combined with Figure 5 As shown, the embodiments of the present disclosure provide a water seepage detection method, including:

[0097] S501, the processor obtains a real-time temperature image of the area to be detected.

[0098] In this step, an infrared thermal imager or other temperature imaging device can be used to obtain a real-time temperature image of the area to be detected.

[0099] S502, the processor determines a target first detection model in the water seepage detection model according to the time information of the real-time temperature image.

[0100] Among them, the water seepage detection model is constructed by using the construction method of the water seepage detection model as described in the above embodiments. The time information of the real-time temperature image refers to the acquisition time of the real-time temperature image, and the time information of the real-time temperature image can be automatically obtained during the process of using an infrared thermal imager or other temperature imaging device to obtain a real-time temperature image of the area to be detected.

[0101] In this step, determining the target first detection model in the water seepage detection model according to the time information of the real-time temperature image includes: obtaining the time marking information of each temperature image in the sub-image training set for training the first detection model and determining the common time period of the time marking information; using the first detection model corresponding to the common time period that conforms to the time information as the target first detection model.

[0102] Exemplarily, the time information (acquisition time) of the real-time temperature image is XXXX11211859, that is, at 18:59 on November 21st, "XXXX" year. The common time period of the time marking information of each temperature image in the sub-image training set used by each first detection model refers to the above-mentioned preset classification rule example. The common time period that "XXXX11211859" conforms to is "from September to mid-November in autumn, 11:00 - 19:59 every day". Then, the first detection model corresponding to the sub-image training set whose time marking information of the temperature image is all within "from September to mid-November in autumn, 11:00 - 19:59 every day" is used as the target first detection model.

[0103] S503, the processor adjusts the target first detection model to the active state and adjusts the non-target first detection model to the deactivated state.

[0104] S504, the processor inputs the real-time temperature image into the adjusted water seepage detection model to obtain a detection result.

[0105] The adjusted water seepage detection model in this step refers to the water seepage detection model adjusted through step S503, including the target first detection model in the active state and the non-target first detection model in the deactivated state.

[0106] The water seepage detection method provided by the embodiments of the present disclosure can construct a water seepage detection model using the construction method of the water seepage detection model as described in the above embodiments to perform water seepage detection to obtain a detection result. Therefore, the techniques possessed by the water seepage detection model in the construction method of the water seepage detection model described in the above embodiments are all possessed by this application, and will not be elaborated here.

[0107] In some embodiments, the detection result includes a normal state and a water seepage state. In the case where the detection result is the water seepage state, the water seepage detection method further includes: extracting feature points and brightness information from the real-time temperature image; performing transformation on the real-time temperature image according to the feature points and brightness information to obtain a transformed image; comparing the transformed image with the real-time temperature image to obtain and output an image of the area to be repaired.

[0108] In this embodiment, after detecting the water seepage state, it is possible to further process the real-time temperature image, extract feature points (such as positions with significant local features like edges and corner points in the image) and brightness information (the brightness information reflects the temperature difference in different regions of the image) in the image. Based on the extracted feature points and brightness information, perform transformation processing on the real-time temperature image to highlight the water seepage area and reduce background interference, obtain a clearer image of the water seepage area, that is, the transformed image, and thereby identify the area to be repaired and output the image of the area to be repaired for subsequent repair work.

[0109] In some embodiments, the BRISK (Binary Robust Independent Elementary Features) algorithm can be used to extract feature points from the real-time temperature image.

[0110] Optionally, the real-time temperature image is transformed according to the feature points and brightness information to obtain a transformed image, including: selecting the i-th feature point as the target feature point; the value of i is 1, 2, 3, …… m, where m is the number of feature points re-extracted from the real-time temperature image; the pixel points in the real-time temperature image with brightness greater than or equal to the target feature point are regarded as bright points and the pixel points with brightness less than the target feature point are regarded as dark points; randomly select a bright point and a dark point, and calculate the first distance from the target feature point to the dark point and the second distance from the target feature point to the bright point; if the first distance is less than the second distance, define the binary descriptor corresponding to the target feature point as 1, otherwise as 0; select the (i + 1)-th feature point as the target feature point... and so on until all feature points are traversed to obtain the transformed image.

[0111] In this embodiment, the binary descriptors of the feature points in the real-time temperature image are transformed according to the brightness information to obtain a transformed image, which can highlight the water seepage area, is insensitive to light changes, has stronger anti-noise ability, and the reliability of the image transformation is relatively high.

[0112] Optionally, the transformed image is compared with the real-time temperature image to obtain an image of the area to be repaired, including: comparing the binary descriptors of each pixel point in the transformed image and the real-time temperature image; determining the area composed of pixel points with different binary descriptors as the area to be repaired to obtain the image of the area to be repaired. In this embodiment, by comparing the changes in the binary descriptors of the transformed image and the real-time temperature image, the area to be repaired can be accurately identified, reducing the probability of false alarms and missed detections.

[0113] Combined Figure 6 As shown, an apparatus 60 for constructing a water seepage detection model according to an embodiment of the present disclosure includes an acquisition module 610, a classification module 620, a training module 630, and a fusion module 640. The acquisition module 610 is configured to acquire a training image set; the training image set is a temperature image set including water seepage temperature images and normal temperature images in which the area to be detected is in a water seepage state and a normal state respectively; the classification module 620 is configured to classify the temperature images in the training image set according to the time marking information of the temperature images to obtain a plurality of sub-image training sets; the training module 630 is configured to train a plurality of pre-constructed sub-detection models respectively based on the plurality of sub-image training sets to obtain a plurality of trained first detection models; the fusion module 640 is configured to fuse the plurality of trained first detection models to obtain a water seepage detection model.

[0114] The construction device 60 of the water seepage detection model provided by the embodiments of the present disclosure can execute the construction method of the water seepage detection model described in the above embodiments. Therefore, the technical effects possessed by the construction method of the water seepage detection model described in the above embodiments are all possessed by the embodiments of the present disclosure, and will not be elaborated here.

[0115] Optionally, the pre-constructed sub-detection model includes a convolutional neural network. The training module 630 is further configured to preprocess the temperature images in each sub-image training set; the preprocessing includes illumination correction and / or contrast enhancement; and input the preprocessed temperature images into the convolutional neural network for training.

[0116] Optionally, the pre-constructed sub-detection model includes a long short-term memory network. The training module 630 is further configured to extract temperature features from the temperature images in each sub-image training set; sort the extracted temperature features according to the time marker information of the temperature images to obtain a target temperature feature sequence; and input the target temperature feature sequence into the long short-term memory network for training.

[0117] Optionally, the pre-constructed sub-detection model includes a convolutional neural network and a long short-term memory network. The training module 630 is further configured to obtain the prediction result of the area to be detected at the prediction moment output by the long short-term memory network; obtain the measured result of the area to be detected output by the convolutional neural network corresponding to the temperature image at the prediction moment; when the correlation coefficient between the prediction result and the measured result is less than the coefficient threshold, determine the target sub-detection model from multiple pre-constructed sub-detection models and optimize the target sub-detection model.

[0118] Optionally, when the target sub-detection model includes a long short-term memory network, the training module 630 is further configured to calculate the mean square error value of the long short-term memory network; update the model parameters of the long short-term memory network according to the mean square error value, and recalculate the mean square error value.

[0119] Optionally, the training image set further includes the repaired temperature image after the water seepage repair of the area to be detected. The fusion module 640 is further configured to train the pre-constructed sub-detection model based on the repaired temperature image to obtain a trained second detection model; and fuse multiple trained first detection models and the second detection model to obtain the water seepage detection model.

[0120] Combined with Figure 7As shown in the figure, an embodiment of the present disclosure provides a water seepage detection device 70, which includes an acquisition module 710, an analysis module 720, an adjustment module 730, and a detection module 740. The acquisition module 710 is configured to obtain a real-time temperature image of the area to be detected; the analysis module 720 is configured to determine a target first detection model in the water seepage detection model according to the time information of the real-time temperature image; wherein, the water seepage detection model is constructed by using the construction method of the water seepage detection model described in the above embodiment; the adjustment module 730 is configured to adjust the target first detection model to an active state and adjust the non-target first detection models to a deactivated state; the detection module 740 is configured to input the real-time temperature image into the adjusted water seepage detection model to obtain a detection result.

[0121] The water seepage detection device 70 provided by the embodiment of the present disclosure can execute the water seepage detection method described in the above embodiment. Therefore, the technical effects of the water seepage detection method described in the above embodiment are all possessed by the embodiment of the present disclosure, and will not be elaborated here.

[0122] Optionally, the water seepage detection device 70 further includes a positioning module 750. In the case where the detection result is a water seepage state, the positioning module 750 is configured to extract feature points and brightness information from the real-time temperature image; transform the real-time temperature image according to the feature points and brightness information to obtain a transformed image; compare the transformed image with the real-time temperature image to obtain and output an image of the area to be repaired.

[0123] In some embodiments, in combination with Figure 2 As shown in the figure, the controller 20 includes a processor 200 and a memory 201. Optionally, the controller 20 may further include a communication interface 202 and a bus 203. Among them, the processor 200, the communication interface 202, and the memory 201 can communicate with each other through the bus 203. The communication interface 202 can be used for information transmission. The processor 200 can call the logical instructions in the memory 201 to execute the construction method of the water seepage detection model and / or the water seepage detection method of the above embodiment.

[0124] In addition, when the logical instructions in the above-mentioned memory 201 are implemented in the form of a software functional unit and sold or used as an independent product, they can be stored in a computer-readable storage medium.

[0125] The memory 201, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of the present disclosure. The processor 200 executes functional applications and data processing by running the program instructions / modules stored in the memory 201, that is, implements the method for constructing the water seepage detection model and / or the water seepage detection method in the above embodiments.

[0126] The memory 201 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the terminal device and the like. In addition, the memory 201 may include a high-speed random access memory and may also include a non-volatile memory.

[0127] The embodiments of the present disclosure provide a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are configured to execute the method for constructing the water seepage detection model and / or the water seepage detection method described above.

[0128] The technical solution of the embodiments of the present disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present disclosure. The foregoing storage medium may be a non-transitory storage medium, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc and other media that can store program codes.

[0129] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure, enabling those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, process, and other changes. Embodiments merely represent possible variations. Unless explicitly required, individual components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terms used in this application are only for describing embodiments and do not limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to also include the plural forms. Similarly, as used in this application, the term "and / or" refers to any and all possible combinations of one or more of the associated listed items. Additionally, when used in this application, the term "comprise" and its variants "comprises" and / or "comprising" etc. mean the presence of the stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups of these. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, or apparatus comprising the element. Herein, what each embodiment focuses on can be the differences from other embodiments, and the same or similar parts among the embodiments can be referred to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, the relevant parts can refer to the description of the method part.

[0130] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner can depend on the specific application and design constraints of the technical solution. The skilled person can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the embodiments of the present disclosure. The skilled person 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 foregoing method embodiments and will not be elaborated herein.

[0131] In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units can be merely a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms. The units described as separate components can be or can not be physically separated. The components displayed as units can be or can not be physical units, that is, they can be located in one place or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to implement this embodiment. Additionally, in the embodiments of the present disclosure, the various functional units can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the blocks can occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, which can depend on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks can also occur in a different order than that disclosed in the description. Sometimes, there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, which can depend on the functions involved. Each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

Claims

1. A method for constructing a water seepage detection model, characterized in that: include: Get a training image set; The training image set is a temperature image set including water seepage temperature images of the area to be detected in a water seepage state and normal temperature images of the area in a normal state; Classifying the temperature images in the training image set according to the time tag information of the temperature images to obtain multiple sub-image training sets; Based on the multiple sub-image training sets, the multiple pre-built sub-detection models are trained respectively to obtain multiple trained first detection models; A plurality of trained first detection models are fused to obtain a water seepage detection model.

2. The construction method according to claim 1, characterized in that: The pre-built sub-detection model includes a convolutional neural network; and the pre-built sub-detection models are trained based on the multiple sub-image training sets respectively, including: Preprocessing the temperature image in each sub-image training set; the preprocessing includes illumination correction and / or contrast enhancement; The preprocessed temperature image is input into the convolutional neural network for training.

3. The construction method according to claim 1, characterized in that: The pre-built sub-detection models include a long short-term memory network; and the pre-built sub-detection models are trained based on the multiple sub-image training sets, including: Extract temperature features from the temperature images in each sub-image training set; The extracted temperature features are sorted according to the time stamp information of the temperature image to obtain a target temperature feature sequence; The target temperature feature sequence is input into the long short-term memory network for training.

4. The construction method according to claim 1, characterized in that: The pre-built sub-detection models include a convolutional neural network and a long short-term memory network; multiple pre-built sub-detection models are trained based on multiple sub-image training sets, including: Obtain the prediction result of the area to be detected output by the long short-term memory network at the prediction time; Obtain the actual measurement result of the area to be detected output by the convolutional neural network corresponding to the temperature image at the prediction time; When the correlation coefficient between the prediction result and the measured result is less than a coefficient threshold, a target sub-detection model is determined from a plurality of pre-built sub-detection models and the target sub-detection model is optimized.

5. The construction method according to claim 4, characterized in that: In the case where the target sub-detection model includes a long short-term memory network, the target sub-detection model is optimized, including: Calculate the mean square error of the long short-term memory network; Update the model parameters of the long short-term memory network according to the mean square error value, and recalculate the mean square error value.

6. The construction method according to any one of claims 1 to 5, characterized in that: The training image set also includes a repair temperature image of the area to be detected after the water seepage repair is completed; a plurality of trained first detection models are fused to obtain a water seepage detection model, including: Training the pre-built sub-detection model based on the patched temperature image to obtain a trained second detection model; A plurality of trained first detection models and second detection models are fused to obtain a water seepage detection model.

7. A water seepage detection method, characterized in that: include: Obtain real-time temperature images of the area to be detected; Determine a target first detection model in the water seepage detection model according to the time information of the real-time temperature image; wherein the water seepage detection model is constructed by the water seepage detection model construction method according to any one of claims 1 to 6; Adjusting the target first detection model to be in an activated state, and adjusting the non-target first detection model to be in a deactivated state; The real-time temperature image is input into the adjusted water seepage detection model to obtain the detection result.

8. The water seepage detection method according to claim 7, characterized in that: When the detection result is a water seepage state, the water seepage detection method further includes: Extract feature points and brightness information from real-time temperature images; The real-time temperature image is transformed according to the feature points and brightness information to obtain a transformed image; The transformed image is compared with the real-time temperature image to obtain the image of the area to be repaired and output it.

9. A device for constructing a water seepage detection model, characterized in that: include: an acquisition module, configured to acquire a training image set; The training image set is a temperature image set including water seepage temperature images of the area to be detected in a water seepage state and normal temperature images of the area in a normal state; A classification module is configured to classify the temperature images in the training image set according to the time tag information of the temperature images to obtain multiple sub-image training sets; A training module is configured to train a plurality of pre-built sub-detection models based on a plurality of sub-image training sets to obtain a plurality of trained first detection models; The fusion module is configured to fuse multiple trained first detection models to obtain a water seepage detection model.

10. A water seepage detection device, characterized in that: include: An acquisition module is configured to obtain a real-time temperature image of the area to be detected; An analysis module is configured to determine a target first detection model in a water seepage detection model according to time information of the real-time temperature image; wherein the water seepage detection model is constructed by the water seepage detection model construction method according to any one of claims 1 to 6; an adjustment module, configured to adjust the target first detection model to be in an activated state, and adjust the non-target first detection model to be in a deactivated state; The detection module is configured to input the real-time temperature image into the adjusted water seepage detection model to obtain the detection result.

11. An electronic device, characterized in that: include: A memory storing program instructions; The processor is configured to execute the method for constructing a water seepage detection model as described in any one of claims 1 to 6, and / or the water seepage detection method as described in claim 7 or 8 when running the program instructions.

Images