Method and system for monitoring water accumulation in a hydropower plant gallery

Abstract

The water power station corridor accumulated water monitoring method and system provided by the embodiment of the application relates to the artificial intelligence technical field. First, a video image in a water power station is acquired through a camera; then, a corridor image is obtained from the video image through detection, and whether the water cloth in the corridor accumulated water detection device is infected is detected from the corridor image; finally, when the water cloth in the corridor accumulated water detection device is infected, the corridor image is processed to obtain the accumulated water depth of the water power station corridor. The above scheme adopts a computer device to acquire the accumulated water depth of the water power station corridor through an image processing mode, and compared with a water level sensor or an ultrasonic water level monitoring mode, the above scheme is more stable and has higher measurement precision.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based method and system for monitoring water accumulation in hydropower station corridors. Background Technology

[0002] In the event of a malfunction in the powerhouse drainage system or a blockage in the outlet pipes of a hydropower station, tailwater will backflow into the corridor under the action of pressure differential, causing safety hazards. Accidents of "flooded corridors" in hydropower stations occur frequently, causing a series of problems such as the electromechanical equipment, powerhouse structure, personnel safety, and power generation of the hydropower station, seriously threatening the safe and stable operation of the hydropower station.

[0003] Existing technologies for monitoring water accumulation in corridors generally employ water level sensors or ultrasonic water level monitoring methods. However, these methods are limited by sensor accuracy, and in the early stages of water accumulation, they are prone to slow response, leading to delays in accident detection. Therefore, providing a highly accurate and stable corridor water accumulation monitoring solution is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In order to overcome the above-mentioned shortcomings in the prior art, the purpose of this application is to provide a method and system for monitoring water accumulation in the corridor of a hydropower station.

[0005] In a first aspect, embodiments of this application provide a method for monitoring water accumulation in a hydropower station corridor, applied to a hydropower station corridor water accumulation monitoring system. The hydropower station corridor water accumulation monitoring system includes a corridor water accumulation detection device, a camera, and computer equipment. The corridor water accumulation detection device includes a scale and a water-resistant cloth that changes color when exposed to water. The scale is flush with one end of the water-resistant cloth. The method includes:

[0006] The camera acquires video images;

[0007] The computer device detects the video image, and when it detects that the video image includes a corridor image of the corridor water accumulation detection device, it detects whether the water cloth in the corridor water accumulation detection device is contaminated by the corridor image.

[0008] When the water-absorbing cloth in the corridor water accumulation detection device is contaminated, the target image area where the corridor water accumulation detection device is located is extracted from the corridor image, and the target image area is processed to obtain the water accumulation depth of the hydropower station corridor.

[0009] In the above scheme, firstly, video images of the hydropower station are acquired using cameras; then, corridor images are obtained from the video images, and the water-absorbing cloth in the corridor's water accumulation detection device is detected as contaminated within these images; finally, if the water-absorbing cloth in the corridor's water accumulation detection device is contaminated, the corridor image is processed to obtain the water depth of the hydropower station's corridor. This scheme uses computer equipment to obtain the water depth of the hydropower station's corridor through image processing, which is more stable and has higher measurement accuracy compared to using water level sensors or ultrasonic water level monitoring methods.

[0010] In one possible implementation, the computer device detects the video image, and when it detects a corridor image including the corridor water accumulation detection device in the video image, the step of detecting whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated by the corridor image includes:

[0011] The video image is input into a trained deep learning network model. The trained deep learning network model detects whether the video image includes the corridor image of the corridor water accumulation detection device, and detects whether the water cloth in the corridor water accumulation detection device is contaminated by the corridor image.

[0012] In one possible implementation, the deep learning network model includes a feature extraction subnetwork, a feature fusion subnetwork, and a prediction subnetwork. The step of inputting the video image into the trained deep learning network model, detecting whether the video image includes an image of the corridor where the corridor water accumulation detection device is located, and detecting whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated by the corridor image, includes:

[0013] The video image is input into the feature extraction subnetwork to extract multidimensional deep features from the video image, resulting in a feature vector map.

[0014] The feature vector map is input into the feature fusion subnetwork to fuse the features in the feature vector map;

[0015] The fused features are input into the prediction sub-network. The fused features are used to detect whether there is a corridor image in the video image that includes the corridor water accumulation detection device, and to predict whether the water cloth in the corridor water accumulation detection device is contaminated.

[0016] In one possible implementation, the feature extraction subnetwork includes a slicing layer, a concatenation layer, a convolutional layer, a batch normalization layer, and an activation function layer. The step of inputting the video image into the feature extraction subnetwork to extract multidimensional depth features from the video image and obtain a feature vector map includes:

[0017] The video image is input into the slicing layer for slicing to obtain multiple slice features, wherein the video image includes 3D feature data;

[0018] The multiple slice features are input into the splicing layer for splicing processing to obtain the spliced ​​features;

[0019] The concatenated features are input into the convolutional layer for processing to obtain a multidimensional deep feature set.

[0020] The multidimensional depth feature set is input into the batch normalization layer for normalization processing to obtain the processed multidimensional depth feature set.

[0021] The multidimensional deep feature set is input into the activation function layer for activation to obtain the feature vector map, wherein the feature vector map includes 128-dimensional, 256-dimensional, or 512-dimensional feature data.

[0022] In one possible implementation, the feature fusion subnetwork includes a top-down feature fusion layer and a bottom-up feature fusion layer. The step of inputting the feature vector map into the feature fusion subnetwork and fusing the features in the feature vector map includes:

[0023] The sampled features obtained by downsampling the feature vector map are input into the top-down feature fusion layer, and the top-down feature fusion layer performs top-down feature fusion on the sampled features;

[0024] The features after top-down fusion are input into the bottom-up feature fusion layer, which then performs a second fusion on the features and outputs the second fused features.

[0025] In one possible implementation, the step of inputting the fused features into the prediction sub-network, detecting whether the video image contains a corridor image including the corridor water accumulation detection device, and predicting whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated, includes:

[0026] Based on the re-fused features, a bounding box for identifying the corridor water accumulation detection device is generated in the video image.

[0027] The category of the water-absorbing cloth in the corridor water accumulation detection device located in the boundary frame is predicted, and the category of the water-absorbing cloth includes two types: contaminated and uncontaminated.

[0028] In one possible implementation, before the step of inputting the video image into a trained deep learning network model and outputting whether the video image contains an image of a corridor including the corridor water accumulation detection device, the method further includes a step of training a deep learning network model, which includes:

[0029] Obtain a training image sample set, which includes image samples carrying annotation information, wherein the annotation information includes label information indicating whether the water cloth has been contaminated;

[0030] The image samples are sequentially input into a deep learning network model for processing, and the predicted information of whether the water cloth in the image samples is contaminated is output.

[0031] The loss function value of the deep learning network model is calculated based on the label information and the prediction information. The model parameters of the deep learning network model are iteratively optimized according to the loss function value until the loss function value of the deep learning network model is less than a preset loss function threshold, thus obtaining a trained deep learning network model.

[0032] In one possible implementation, the step of extracting the target image region where the water accumulation detection device is located from the corridor image and processing the target image region to obtain the water accumulation depth of the hydropower station corridor includes:

[0033] The corridor image is preprocessed;

[0034] The target image region is obtained by contour extraction from the preprocessed corridor image.

[0035] Obtain the relative positional relationship between the location of the water-soaked cloth in the target image region and the scale.

[0036] The water depth of the hydropower station corridor is determined based on the relative position of the water-soaked cloth and the scale.

[0037] In one possible implementation, the step of preprocessing the corridor image includes:

[0038] The corridor image is processed into grayscale to obtain a grayscale image of the corridor;

[0039] The grayscale image of the corridor is subjected to noise reduction processing to obtain a noise-reduced grayscale image of the corridor.

[0040] The denoised grayscale image of the corridor is binarized to obtain a binary image of the corridor, and then the binary image of the corridor is subjected to erosion and dilation processing.

[0041] Secondly, this application embodiment also provides a hydropower station corridor water accumulation monitoring system, which includes a corridor water accumulation detection device, a camera and computer equipment. The corridor water accumulation detection device includes a scale and a water-resistant cloth that can be dyed and changed color when exposed to water. The scale is flush with one end of the water-resistant cloth.

[0042] The camera is used to acquire video images;

[0043] The computer device is used to detect the video image, and when it detects that the video image includes a corridor image of the corridor water accumulation detection device, it uses the corridor image to detect whether the water cloth in the corridor water accumulation detection device is contaminated.

[0044] The computer equipment is also used to extract the target image area where the water-accumulation detection device is located from the corridor image when the water cloth in the corridor water accumulation detection device is contaminated, and to process the target image area to obtain the water accumulation depth of the hydropower station corridor.

[0045] Based on any of the above aspects, the hydropower station corridor water accumulation monitoring method and system provided in this application first acquires video images of the hydropower station using a camera; then, it detects corridor images obtained from the video images and checks whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated from the corridor images; finally, when the water-absorbing cloth in the corridor water accumulation detection device is contaminated, it processes the corridor image to obtain the water accumulation depth of the hydropower station corridor. The above solution uses computer equipment to obtain the water accumulation depth of the hydropower station corridor through image processing, which is more stable and has higher measurement accuracy compared to using water level sensors or ultrasonic water level monitoring methods. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings required in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A schematic block diagram of a hydropower station corridor water accumulation monitoring system provided in an embodiment of this application;

[0048] Figure 2 for Figure 1 A schematic diagram of a structure for a water accumulation detection device in a central corridor;

[0049] Figure 3 A schematic flowchart of a method for monitoring water accumulation in a hydropower station corridor provided in an embodiment of this application;

[0050] Figure 4 This is a schematic diagram of the structure of a deep learning network model provided in an embodiment of this application;

[0051] Figure 5 for Figure 3 A flowchart illustrating the sub-steps of step S102;

[0052] Figure 6 for Figure 3 A flowchart illustrating the sub-steps of step S103. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0054] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0055] Please refer to Figure 1 , Figure 1This document illustrates an application scenario diagram of the technical solution of this application embodiment, specifically a structural block diagram of a hydropower station corridor water accumulation monitoring system 10. As shown, the hydropower station corridor water accumulation monitoring system 10 may include a corridor water accumulation detection device 100, a camera 200, and a computer device 300. The camera 200 can be communicatively connected to the computer device 300. Specifically, the camera 200 and the computer device 300 can be connected via wired or wireless means. The shooting angle of the camera 200 can be varied; for example, the camera 200 can be mounted on a pan-tilt unit, and video images of various areas (including the corridor) within the hydropower station can be obtained by continuously adjusting the shooting angle. Alternatively, the shooting angle of the camera 200 can be fixed, ensuring that the camera 200 can capture video images of the corridor. The corridor water accumulation detection device 100 is installed inside the corridor; specifically, it can be installed on the side wall of the corridor. Please refer to [reference needed]. Figure 2 , Figure 2 The diagram illustrates the structure of a corridor water accumulation detection device 100. The corridor water accumulation detection device 100 may include a scale 110 and a water cloth 120. The scale 110 and the water cloth 120 are flush at one end. The water cloth 120 can be stained and discolored (e.g., turn black) when exposed to water.

[0056] Please refer to Figure 3 , Figure 3 This application provides a schematic flowchart illustrating a method for monitoring water accumulation in hydropower station corridors, which can be implemented by... Figure 1 The hydropower station corridor water accumulation monitoring system 10 is implemented. The order of some steps in the hydropower station corridor water accumulation monitoring method of this application embodiment can be interchanged according to actual needs, or some steps can be omitted or deleted. The following is in conjunction with... Figure 3 This paper introduces the detailed steps of the method for monitoring water accumulation in the corridors of hydropower stations.

[0057] Step S101: Camera 200 acquires video images.

[0058] As mentioned above, camera 200 can acquire video images of various areas within the hydropower station, or it can acquire video images of only the corridor.

[0059] In step S102, the computer device detects the video image, and when it detects that the video image includes a corridor image of the corridor water accumulation detection device, it detects whether the water cloth in the corridor water accumulation detection device is contaminated by the corridor image.

[0060] To reduce the computational load on the computer device 300, the computer device 300 can extract a portion of the video image from the video image through sampling, perform detection on the extracted video image, and adjust the size of the extracted video image.

[0061] In this step, video images can be input into a trained deep learning network model. The trained deep learning network model can detect whether the video images include images of the corridor water accumulation detection device, and then detect whether the water cloth in the corridor water accumulation detection device is contaminated by the corridor images.

[0062] In step S103, when the water cloth 120 in the corridor water accumulation detection device 100 is contaminated, the target image area where the corridor water accumulation detection device 100 is located is extracted from the corridor image, and the target image area is processed to obtain the water accumulation depth of the hydropower station corridor.

[0063] The solution provided in this application first acquires video images of the hydropower station using a camera 200; then, it detects corridor images obtained from the video images and checks whether the water-absorbing cloth in the corridor water accumulation detection device 100 is contaminated from the corridor images; finally, when the water-absorbing cloth in the corridor water accumulation detection device 100 is contaminated, it processes the corridor image to obtain the water depth of the hydropower station corridor. This solution uses a computer device 200 to obtain the water depth of the hydropower station corridor through image processing, which is more stable and has higher measurement accuracy compared to using water level sensors or ultrasonic water level monitoring.

[0064] Please refer to Figure 4 and Figure 5 , Figure 4 Example: A schematic diagram of the structure of a deep learning network model in an embodiment of this application. Figure 5 Example Figure 3 The flowchart of the sub-steps in step S102 is shown below. The deep learning network model 200 includes a feature extraction sub-network 210, a feature fusion sub-network 220, and a prediction sub-network 230. Further, step S102 can be implemented through the following sub-steps.

[0065] In sub-step S1021, the video image is input into the feature extraction sub-network 210 to extract multi-dimensional depth features from the video image and obtain a feature vector map.

[0066] Please refer to this again. Figure 4The feature extraction subnetwork 210 includes a slicing layer 2101, a splicing layer 2102, a convolutional layer 2103, a batch normalization layer 2104, and an activation function layer 2105. The input of the slicing layer 2101 is the input of the entire feature extraction subnetwork 210. The output of the slicing layer 2101 is connected to the input of the splicing layer 2102. The output of the splicing layer 2102 is connected to the input of the convolutional layer 2103. The output of the convolutional layer 2103 is connected to the input of the batch normalization layer 2104. The output of the batch normalization layer 2104 is connected to the input of the activation function layer 2105. The output of the activation function layer 2105 is the output of the feature extraction subnetwork 210.

[0067] Sub-step S1021 can be implemented in the following way.

[0068] First, the video image is input into the slicing layer 2101 for slicing operation to obtain multiple slice features, where the video image includes 3D feature data.

[0069] Next, multiple slice features are input into the splicing layer 2102 for splicing processing to obtain the spliced ​​features.

[0070] Next, the concatenated features are input into convolutional layer 2103 for processing to obtain a multidimensional deep feature set.

[0071] Then, the multidimensional deep feature set is input into the batch normalization layer 2104 for normalization processing to obtain the processed multidimensional deep feature set.

[0072] Finally, the multidimensional deep feature set is input into the activation function layer 2105 for activation to obtain a feature vector map, wherein the feature vector map includes 128-dimensional, 256-dimensional, or 512-dimensional feature data.

[0073] Sub-step S1022: Input the feature vector map into the feature fusion sub-network and fuse the features in the feature vector map.

[0074] Please refer to this again. Figure 4 The feature fusion subnetwork 220 may include a top-down feature fusion layer 2201 and a bottom-up feature fusion layer 2202.

[0075] Sub-step S1022 can be implemented in the following way.

[0076] First, the sampled features obtained by downsampling the feature vector map are input into a top-down feature fusion layer, which performs top-down feature fusion on the sampled features. Through top-down feature fusion, deep semantic features can be passed down, enhancing semantic information.

[0077] Then, the features obtained from the top-down fusion are input into the bottom-up feature fusion layer, which then fuses the top-down features again and outputs the re-fused features. Through bottom-up feature fusion, the localization features from lower layers can be passed up.

[0078] The above fusion can enhance both deep semantic features and low-level positioning features, so as to facilitate the subsequent positioning of the corridor water accumulation detection device 100 and the identification of whether the water cloth 120 is contaminated.

[0079] In sub-step S1023, the fused features are input into the prediction sub-network 230. The fused features are used to detect whether there is a corridor image including a corridor water accumulation detection device in the video image, and to predict whether the water cloth in the corridor water accumulation detection device is contaminated.

[0080] In this sub-step, the features input to the prediction sub-network 230 are the re-fused features output from the bottom-up feature fusion layer 2202 in the feature fusion sub-network 220. These re-fused features can be converted into specific coordinate information of the corridor water accumulation detection device 100 and label information indicating whether the water-covered sheet 120 is contaminated. Specifically, a bounding box for identifying the corridor water accumulation detection device can be generated in the video image based on the re-fused features, and the category of the water-covered sheet 120 within the corridor water accumulation detection device 100 located within the bounding box can be predicted. The categories of the water-covered sheet 120 include contaminated and uncontaminated.

[0081] Furthermore, in this embodiment of the application, before step S102, the method for monitoring water accumulation in the hydropower station corridor provided in this embodiment of the application may also include a step of training a deep learning network model, which can be implemented in the following ways.

[0082] First, a training image sample set is obtained, which includes image samples carrying annotation information, including labels indicating whether the tarpaulin has been contaminated.

[0083] Image samples in the image sample set can be obtained by expanding known images through rotation, translation, scaling, and mirroring. The annotation information includes two categories: uninfected and infected.

[0084] Next, the image samples are sequentially input into the deep learning network model for processing, and the output is a prediction of whether the water cloth in the image sample is contaminated.

[0085] Then, the loss function value of the deep learning network model is calculated based on the label information and prediction information. The model parameters of the deep learning network model are iteratively optimized according to the loss function value until the loss function value of the deep learning network model is less than the preset loss function threshold, thus obtaining the trained deep learning network model.

[0086] Further, please refer to Figure 6 , Figure 6 Example Figure 3 The flowchart of the sub-step S103 is shown below. Step S103 can be implemented in the following way.

[0087] Sub-step S1031: Preprocess the corridor image.

[0088] Specifically, the preprocessing can be done as follows.

[0089] First, the corridor image is processed into grayscale to obtain a grayscale image of the corridor;

[0090] Next, the grayscale image of the corridor is subjected to noise reduction processing to obtain a noise-reduced grayscale image of the corridor.

[0091] Then, the denoised grayscale image of the corridor is binarized to obtain a binary image of the corridor.

[0092] Finally, erosion and dilation processing is applied to the binary image of the corridor to make its features more refined and accurate.

[0093] Sub-step S1032 involves extracting the contour of the preprocessed corridor image to obtain the target image region.

[0094] In this step, the target image region can be obtained through morphological processing and contour edge detection. Specifically, the detected contours can be filtered according to pre-set contour filtering conditions (such as area and aspect ratio thresholds) to obtain the target image region corresponding to the filtered contours.

[0095] Sub-step S1033: Obtain the relative positional relationship between the location of the water-soaked cloth in the target image area and the scale.

[0096] Sub-step S1034: Determine the water depth of the hydropower station corridor based on the relative position of the water-soaked cloth and the scale.

[0097] The method for monitoring water accumulation in hydropower station corridors provided in this application first acquires video images of the hydropower station using a camera; then, it detects corridor images obtained from the video images and checks whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated; finally, when the water-absorbing cloth in the corridor water accumulation detection device is contaminated, the corridor image is processed to obtain the water depth of the hydropower station corridor. This method uses computer equipment to obtain the water depth of the hydropower station corridor through image processing, which is more stable and has higher measurement accuracy compared to using water level sensors or ultrasonic water level monitoring.

[0098] Furthermore, please refer to again Figure 1 and Figure 2 This application embodiment also provides a hydropower station corridor water accumulation monitoring system 10. The hydropower station corridor water accumulation monitoring system 10 may include a corridor water accumulation detection device 100, a camera 200 and a computer device 300. The corridor water accumulation detection device 100 includes a scale 110 and a water-resistant cloth 120 that can be dyed and changed color when exposed to water. The scale 110 and the water-resistant cloth 120 are set flush with one end.

[0099] Camera 200 is used to acquire video images.

[0100] Computer device 300 is used to detect video images, and when a corridor image including a corridor water accumulation detection device is detected in the video image, it detects whether the water cloth in the corridor water accumulation detection device is contaminated by the corridor image.

[0101] A deep learning network model can be pre-trained in computer device 300. The trained deep learning network model can be used to detect whether the video image includes the corridor image of the corridor water accumulation detection device, and the corridor image can be used to detect whether the water cloth in the corridor water accumulation detection device is contaminated.

[0102] The computer device 300 is also used to extract the target image area where the water-accumulation detection device is located from the corridor image when the water cloth in the corridor water accumulation detection device is contaminated, and to process the target image area to obtain the water accumulation depth of the hydropower station corridor.

[0103] In summary, the hydropower station corridor water accumulation monitoring method and system provided in this application first acquires video images of the hydropower station using a camera; then, it detects corridor images obtained from the video images and checks whether the water-absorbing cloth in the corridor water accumulation detection device is contaminated from the corridor images; finally, when the water-absorbing cloth in the corridor water accumulation detection device is contaminated, it processes the corridor image to obtain the water accumulation depth of the hydropower station corridor. The above solution uses computer equipment to obtain the water accumulation depth of the hydropower station corridor through image processing, which is more stable and has higher measurement accuracy compared to using water level sensors or ultrasonic water level monitoring methods.

[0104] The embodiments described above are merely some, not all, of the embodiments of this application. The components of the embodiments of this application typically described and illustrated in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of protection of this application, but merely to illustrate selected embodiments of this application. Based on this, the scope of protection of this application should be determined by the scope of the claims. Furthermore, all other embodiments that can be obtained by those skilled in the art based on the embodiments of this application without inventive effort should fall within the scope of protection of this application.

Claims

1. A method for monitoring water accumulation in a power plant corridor, characterized by, The application is applied to a hydropower station corridor water accumulation monitoring system, the hydropower station corridor water accumulation monitoring system comprises a corridor water accumulation detection device, a camera and a computer device, wherein the corridor water accumulation detection device comprises a ruler and a water cloth which can be infected and changed color when meeting water, the ruler is arranged flush with one end of the water cloth, and the method comprises: The camera acquires a video image; The computer device detects the video image, and when detecting that the video image comprises a corridor image of the corridor water accumulation detection device, detects whether the water cloth in the corridor water accumulation detection device is infected through the corridor image; When the water cloth in the corridor water accumulation detection device is infected, a target image area where the corridor water accumulation detection device is located is extracted from the corridor image, and the water depth of the hydropower station corridor is obtained by processing the target image area; The step that the computer device detects the video image, and when detecting that the video image comprises a corridor image of the corridor water accumulation detection device, detects whether the water cloth in the corridor water accumulation detection device is infected through the corridor image, comprises: inputting the video image into a feature extraction subnetwork to extract multi-dimensional depth features in the video image and obtain a feature vector graph; inputting the feature vector graph into a feature fusion subnetwork to fuse features in the feature vector graph; inputting the fused features into a prediction subnetwork to detect whether there is a corridor image comprising the corridor water accumulation detection device in the video image through the fused features, and to predict whether the water cloth in the corridor water accumulation detection device is infected, wherein the fused features comprise specific coordinate information of the corridor water accumulation detection device and label information of whether the water cloth is infected; When the water cloth in the corridor water accumulation detection device is infected, the step of extracting a target image area where the corridor water accumulation detection device is located from the corridor image, and obtaining the water depth of the hydropower station corridor by processing the target image area, comprises: preprocessing the corridor image; extracting a contour of the preprocessed corridor image to obtain the target image area; acquiring a relative position relationship between an infected position of the water cloth and the ruler in the target image area; determining the water depth of the hydropower station corridor according to the relative position relationship between the infected position of the water cloth and the ruler.

2. The method of claim 1, wherein the water accumulation in the power plant corridor is monitored by using a camera. The step that the computer device detects the video image, and when detecting that the video image comprises a corridor image of the corridor water accumulation detection device, detects whether the water cloth in the corridor water accumulation detection device is infected through the corridor image, comprises: inputting the video image into a trained deep learning network model to detect whether there is a corridor image comprising the corridor water accumulation detection device in the video image through the trained deep learning network model, and to detect whether the water cloth in the corridor water accumulation detection device is infected through the corridor image.

3. The method of claim 2, wherein the water level of the water is determined by a water level sensor. The feature extraction sub-network comprises a slice layer, a splicing layer, a convolution layer, a batch normalization layer and an activation function layer, the step of inputting the video image into the feature extraction sub-network to extract multi-dimensional deep features in the video image and obtain a feature vector map comprises: inputting the video image into the slice layer for slice operation to obtain a plurality of slice features, wherein the video image comprises 3-dimensional feature data; inputting the plurality of slice features into the splicing layer for splicing processing to obtain spliced splicing features; inputting the splicing features into the convolution layer for processing to obtain a multi-dimensional deep feature set; inputting the multi-dimensional deep feature set into the batch normalization layer for normalization processing to obtain a processed multi-dimensional deep feature set; inputting the multi-dimensional deep feature set into the activation function layer for activation to obtain the feature vector map, wherein the feature vector map comprises 128-dimensional, 256-dimensional or 512-dimensional feature data.

4. The method of claim 3, wherein the water level of the water is determined by using a water level sensor. The feature fusion sub-network comprises a top-down feature fusion layer and a bottom-up feature fusion layer, the step of inputting the feature vector map into the feature fusion sub-network to fuse features in the feature vector map comprises: inputting sampling features obtained by downsampling the feature vector map into the top-down feature fusion layer to perform top-down feature fusion on the sampling features by the top-down feature fusion layer; inputting the top-down fused features into the bottom-up feature fusion layer to perform re-fusion on the top-down fused features by the bottom-up feature fusion layer and output re-fused features.

5. The method of claim 4, wherein the water level of the water is determined by the water level sensor. The step of inputting the fused features into the prediction sub-network, detecting whether a gallery image comprising the gallery water detection device exists in the video image through the fused features, and predicting whether the water cloth in the gallery water detection device is infected comprises: generating a bounding box for identifying the gallery water detection device in the video image according to the re-fused features; predicting the category of the water cloth in the gallery water detection device located in the bounding box, the category of the water cloth comprising infected and uninfected.

6. The method of monitoring the accumulation of water in the corridors of a hydroelectric plant according to any of claims 2 to 5, characterized in that, Before the step of inputting the video image into the trained deep learning network model and outputting whether a gallery image comprising the gallery water detection device exists in the video image through the trained deep learning network model, the method further comprises a step of training a deep learning network model, which comprises: obtaining a training image sample set comprising image samples carrying annotation information, wherein the annotation information comprises label information annotated with whether the water cloth is infected; sequentially inputting the image samples into the deep learning network model for processing to output prediction information of whether the water cloth is infected in the image samples; Calculate a loss function value of the deep learning network model based on the label information and the prediction information, and iteratively optimize model parameters of the deep learning network model according to the loss function value until the loss function value of the deep learning network model is less than a preset loss function threshold, to obtain a trained deep learning network model.

7. The method of monitoring the accumulation of water in the gallery of a hydroelectric power plant according to claim 6, characterized in that, The step of preprocessing the gallery image comprises: gray processing the gallery image to obtain a gallery gray image; noise reduction processing the gallery gray image to obtain a noise-reduced gallery gray image; binarization processing the noise-reduced gallery gray image to obtain a gallery binary image, and performing erosion and expansion processing on the gallery binary image.

8. A hydropower plant gallery ponding monitoring system, characterized by, The water power station gallery water accumulation monitoring system comprises a gallery water accumulation detection device, a camera and a computer device, wherein the gallery water accumulation detection device comprises a ruler and a water cloth that can be infected and changed in color when meeting water, and the ruler is arranged flush with one end of the water cloth; The camera is configured to acquire a video image. The computer device is configured to detect the video image, and when detecting that the video image comprises a gallery image of the gallery water accumulation detection device, detect whether the water cloth in the gallery water accumulation detection device is infected through the gallery image. The computer device is further configured to, when the water cloth in the gallery water accumulation detection device is infected, extract a target image region where the gallery water accumulation detection device is located from the gallery image, and obtain the water accumulation depth of the water power station gallery by processing the target image region. The computer device is further configured to input the video image into a feature extraction sub-network to extract multi-dimensional depth features in the video image, to obtain a feature vector graph; input the feature vector graph into a feature fusion sub-network to fuse features in the feature vector graph; and input the fused features into a prediction sub-network to detect whether the gallery image comprising the gallery water accumulation detection device exists in the video image through the fused features, and to predict whether the water cloth in the gallery water accumulation detection device is infected, wherein the fused features comprise specific coordinate information of the gallery water accumulation detection device and label information of whether the water cloth is infected. The computer device is further configured to preprocess the gallery image, extract a contour of the preprocessed gallery image to obtain the target image region, acquire a relative position relationship between an infected position of the water cloth and the ruler in the target image region, and determine the water accumulation depth of the water power station gallery according to the relative position relationship.

Images