Visual identification system and method for water passing monitoring of reservoir spillway

Abstract

The invention relates to the technical field of visual identification, and discloses a reservoir spillway water passing monitoring visual identification system and method, and the method comprises the steps: monitoring a spillway water passing image and a real-time image of a peripheral region in real time, and extracting key features in the spillway water passing image through a convolutional neural network; and the water level change trend is analyzed through a deep learning algorithm, the type and size of the floating object are identified, and the health condition of the structure is monitored, so that the flood discharge volume is predicted based on a water level prediction result, a floating object identification result and a structure anomaly detection result, the flood discharge volume prediction result is greatly improved, and the accuracy of flood discharge volume prediction is improved. A worker can make a quick response according to the prediction condition of the flood overflow volume, so that the flood overflow risk is effectively prevented, and the operation safety of a reservoir is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of visual recognition technology, and in particular to a visual recognition system and method for monitoring water flow through a reservoir spillway. Background Art

[0002] With the increasing complexity of reservoir and spillway management systems, traditional overflow monitoring methods can no longer meet the efficiency and accuracy requirements of modern reservoir management. Traditional overflow monitoring relies on manual observation and simple flow calculation models, ignoring the complexity of the spillway and its surrounding environment. In order to improve the accuracy of overflow prediction, in recent years, the use of drone monitoring technology to carry out intelligent monitoring and management of reservoir spillways has become a research hotspot. However, drone monitoring is easily affected by weather factors, and the monitoring data is unstable, which affects the accuracy and effectiveness of the final overflow prediction results. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a visual recognition system and method for monitoring water flow through a reservoir spillway, so as to solve the problem that the accuracy and effectiveness of current spillway prediction results are low.

[0004] The first aspect of the present invention discloses a reservoir spillway water flow monitoring visual recognition system, the system comprises a monitoring camera unit, a data transmission unit, and a visual recognition processing unit; wherein,

[0005] The monitoring camera unit includes a fixed wide-angle camera and a rotatable pan-tilt camera, which are installed in the spillway of the reservoir and the surrounding area, and are used to collect images of water flowing through the spillway and surrounding environment as first image data;

[0006] The data transmission unit is used to transmit the first image data to the visual recognition processing unit in real time;

[0007] The visual recognition processing unit is used to perform a flood overflow prediction operation based on the first image data; before performing the flood overflow prediction operation, the water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction result, the floating object recognition result and the structural anomaly detection result. Furthermore, the system also includes a water level monitoring unit to monitor the water level, and transmit the water level data to the visual recognition processing unit through the data transmission unit;

[0008] The water level prediction operation includes predicting the water level change trend through a water level prediction model based on spillway water flow image data, water level data and meteorological data.

[0009] Furthermore, the process of establishing the water level prediction model includes:

[0010] Obtain historical spillway water flow image data, water level data, and meteorological data, and extract water surface features from the spillway water flow images through a convolutional neural network; the water surface features include contours, waves, splashes, and flow states;

[0011] A water level prediction model is constructed through a deep learning network, and water surface characteristics, water level data, and meteorological data are used as training data sets to train the water level prediction model.

[0012] Furthermore, the deep learning network includes a fusion layer, and the fusion layer is provided with a time series modeling network, and the historical water level data and meteorological data are processed by the time series modeling network, and the processing process includes:

[0013] Performing data preprocessing operations on historical water level data and meteorological data to obtain first processed data;

[0014] Performing feature extraction operations on the first processed data through a gated recurrent unit GRU network to obtain time series features of historical water level data and meteorological data, and obtaining time series dependencies between the time series features through a gating structure in the GRU network, and learning dynamic change laws of historical water level data and meteorological data based on the obtained time series dependencies;

[0015] Outputting second processed data through a gated recurrent unit GRU network; the second processed data includes time series characteristics and dynamic change rules of historical water level data and meteorological data;

[0016] The time series features of the historical meteorological data in the second processed data are fused with the water surface features to obtain fused features;

[0017] The deep learning network also includes a prediction layer, which sets a regression network, inputs fusion features and dynamic change rules into the prediction layer, and outputs water level prediction values ​​through the prediction layer.

[0018] Furthermore, the step of inputting the fusion features and the dynamic change rules into the prediction layer and outputting the water level prediction value through the prediction layer specifically includes:

[0019] Perform weighted processing on the fusion features based on the dynamic change rules to generate weighted features;

[0020] The weighted features are input into the regression network, and the weighted features are subjected to multi-layer nonlinear transformation through the regression network. The preliminary estimated water level prediction value is output in combination with the temporal dependency of the weighted features; wherein the preliminary estimated water level prediction value carries timestamp information;

[0021] The water level deviation between the preliminary estimated water level prediction value and the time series characteristics of the historical water level data in the second processed data is analyzed, the water level deviation is fed back to the water level prediction model, and the parameters of the water level prediction model are adjusted based on the water level deviation; the time series characteristics of the historical water level data in the second processed data include timestamp information and the corresponding water level; the timestamp information of the preliminary estimated water level prediction value is consistent with the timestamp information of the time series characteristics of the historical water level data in the second processed data.

[0022] Furthermore, the floating object identification operation includes:

[0023] Extracting floating object features from the spillway water flow image through a convolutional neural network; the floating object features include the floating object position, floating object shape, and floating object movement trajectory;

[0024] Identify the characteristics of floating objects through target monitoring algorithms to determine the type and size of floating objects;

[0025] The impact fraction of floating objects on the flood overflow is determined based on the type, size and location information of floating objects.

[0026] Further, the surrounding environment image includes image data of the spillway structure and the surrounding area;

[0027] The structural anomaly detection operation includes:

[0028] Extracting structural features from the surrounding environment image through a convolutional neural network; the structural features include crack location, crack size, corrosion spots, surface damage points, and deformation features;

[0029] The deep learning algorithm is used to identify the abnormal state of the structure based on the extracted structural features and generate corresponding abnormal judgment results;

[0030] The abnormal judgment results include the health score of the structure, and the impact score of the current structural state on the flood overflow is determined according to the health score.

[0031] Furthermore, the flood overflow prediction operation includes:

[0032] The preliminary flood overflow prediction value is calculated based on the predicted water level value and the relationship formula between water level and flood overflow;

[0033] Based on the impact fraction of floating objects on the overflow volume and the impact fraction of the current structural status on the overflow volume, the preliminary overflow volume prediction value is corrected to obtain the final overflow volume prediction value.

[0034] Furthermore, the system also includes an alarm unit;

[0035] Before executing the flood overflow prediction operation, it also includes setting an impact score safety threshold. When it is determined that the impact score of floating objects on the flood overflow and / or the impact score of the current structural state on the flood overflow is greater than the impact score safety threshold, an alarm message is generated, and an alarm is issued according to the alarm message through the alarm unit.

[0036] The second aspect of the present invention discloses a visual recognition method for monitoring water flow through a reservoir spillway, which is applied to the system disclosed in the first aspect, and comprises:

[0037] A fixed wide-angle camera and a rotatable pan-tilt camera are provided, and the fixed wide-angle camera and the rotatable pan-tilt camera are installed on the spillway of the reservoir and the surrounding area, and the water flow image of the spillway and the surrounding environment image are collected by the fixed wide-angle camera and the rotatable pan-tilt camera as the first image data;

[0038] A flood overflow prediction operation is performed based on the first image data; before performing the flood overflow prediction operation, water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction result, the floating object recognition result and the structural anomaly detection result.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention monitors the water flow images of the spillway and the surrounding areas in real time, uses a convolutional neural network to extract key features in the water flow images of the spillway, and uses a deep learning algorithm to analyze the water level change trend, identify the type and size of floating objects, and monitor the health status of the structure, so as to predict the overflow volume based on the water level prediction results, the floating object identification results and the structural abnormality detection results, which greatly improves the overflow volume prediction results and enables the staff to make a quick response according to the predicted overflow volume, thereby effectively preventing overflow risks and ensuring the safe operation of the reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the economic application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0042] Figure 1 It is a structural schematic diagram of a visual recognition system for monitoring water flow through a reservoir spillway disclosed in the first embodiment of the present invention;

[0043] Figure 2 A schematic flow chart of a visual recognition method for monitoring water flow through a reservoir spillway disclosed in yet another embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0045] Embodiment 1

[0046] The first aspect of the present invention discloses a reservoir spillway water flow monitoring visual recognition system, see Figure 1 , Figure 1 The present invention discloses a structure diagram of a visual recognition system for monitoring water flow through a reservoir spillway, the system comprising a monitoring camera unit, a data transmission unit, and a visual recognition processing unit; wherein:

[0047] The monitoring camera unit includes a fixed wide-angle camera and a rotatable pan-tilt camera, which are installed on the spillway of the reservoir and the surrounding area, and are used to collect images of water flowing through the spillway and surrounding environment as first image data;

[0048] The data transmission unit is used to transmit the first image data to the visual recognition processing unit in real time;

[0049] The visual recognition processing unit is used to perform a flood overflow prediction operation based on the first image data; before performing the flood overflow prediction operation, water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction results, floating object recognition results and structural anomaly detection results.

[0050] Furthermore, the system also includes setting a water level monitoring unit to monitor the water level, and transmitting the water level data to the visual recognition processing unit through the data transmission unit;

[0051] The water level prediction operation includes predicting the water level change trend through a water level prediction model based on spillway water flow image data, water level data and meteorological data.

[0052] Furthermore, the process of establishing the water level prediction model includes:

[0053] Historical spillway water flow image data, water level data and meteorological data are obtained, and water surface features are extracted from the spillway water flow images through a convolutional neural network; the water surface features include but are not limited to contours, waves, splashes, and flow states.

[0054] A water level prediction model is constructed through a deep learning network, and water surface characteristics, water level data, and meteorological data are used as training data sets to train the water level prediction model.

[0055] Specifically, in the embodiment of the present invention, the meteorological data is obtained based on the weather forecast or the set weather monitoring equipment, including but not limited to precipitation, wind speed, air pressure, wind direction, etc.

[0056] In an embodiment of the present invention, by extracting water surface features, visual information related to water level changes can be effectively obtained. At the same time, combined with meteorological data, various factors affecting water level changes can be more comprehensively reflected, and the complex relationship between water level and meteorological and image features can be learned, thereby improving the accuracy and reliability of water level prediction, and ultimately providing a scientific basis for the prediction and management of reservoir spillway water flow.

[0057] Furthermore, the deep learning network includes a fusion layer, and the fusion layer sets a time series modeling network. The historical water level data and meteorological data are processed through the time series modeling network. The processing process includes:

[0058] Performing data preprocessing operations on historical water level data and meteorological data to obtain first processed data;

[0059] Performing feature extraction operations on the first processed data through a gated recurrent unit GRU network to obtain time series features of historical water level data and meteorological data, and obtaining time series dependencies between the time series features through a gating structure in the GRU network, and learning dynamic change laws of historical water level data and meteorological data based on the obtained time series dependencies;

[0060] Outputting second processed data through a gated recurrent unit GRU network; the second processed data includes time series characteristics and dynamic change rules of historical water level data and meteorological data;

[0061] The time series features of the historical meteorological data in the second processed data are fused with the water surface features to obtain fused features;

[0062] The deep learning network also includes a prediction layer, which sets a regression network, inputs fusion features and dynamic change rules into the prediction layer, and outputs water level prediction values ​​through the prediction layer.

[0063] In this embodiment, the dynamic change law refers to the regular trends in the water level change process and the meteorological change process, which are usually affected by multiple factors such as seasonal changes, precipitation, temperature, and water volume in the basin. Through the learning of the time series modeling network, the model can not only identify these regular trends, but also dynamically adapt to changes in external factors. For example, under climate change or sudden weather conditions, the time series modeling network can capture these changes in time and adjust the output of the model, thereby improving the accuracy and adaptability of water level prediction.

[0064] It is understandable that the reason why the present embodiment uses a time series modeling network is that water level data and meteorological data have strong time series characteristics. The change in water level is not only affected by the current water surface conditions, but also by the long-term accumulation and change trend of factors such as historical data and meteorological conditions. By introducing a time series modeling network, especially a gated recurrent unit (GRU) network structure, historical data can be modeled and the time series dependency rules in the data can be mined.

[0065] Furthermore, the fusion features and dynamic change rules are input into the prediction layer, and the water level prediction values ​​output by the prediction layer include:

[0066] Perform weighted processing on the fusion features based on the dynamic change rules to generate weighted features;

[0067] The weighted features are input into the regression network, and the weighted features are subjected to multi-layer nonlinear transformation through the regression network. The preliminary estimated water level prediction value is output in combination with the temporal dependency of the weighted features; wherein the preliminary estimated water level prediction value carries timestamp information;

[0068] The water level deviation between the preliminary estimated water level prediction value and the time series characteristics of the historical water level data in the second processed data is analyzed, the water level deviation is fed back to the water level prediction model, and the parameters of the water level prediction model are adjusted based on the water level deviation; the time series characteristics of the historical water level data in the second processed data include timestamp information and the corresponding water level; the timestamp information of the preliminary estimated water level prediction value is consistent with the timestamp information of the time series characteristics of the historical water level data in the second processed data.

[0069] It should be explained that, in this embodiment, time series refers to the relationship and dependency between various time points in the data, and describes the dynamic connection between historical data and future data. For example, the water level prediction at a certain moment not only depends on the water level data at that moment, but also needs to refer to the water level changes at the past few moments, because the water level changes usually have a certain continuity and trend. In time series data, each piece of data usually has a timestamp, indicating the specific time when the data was collected or recorded. For example, water level data may include water level changes every hour, and each record will have a specific timestamp, indicating the time when the record occurred. Therefore, in the water level prediction model, it is necessary to compare the preliminary estimated water level prediction value with the timestamp of the historical water level data in the second processed data.

[0070] Furthermore, the floating object identification operation includes:

[0071] Extracting floating object features from the spillway water flow image through a convolutional neural network; the floating object features include the floating object position, floating object shape, and floating object movement trajectory;

[0072] Identify the characteristics of floating objects through target monitoring algorithms to determine the type and size of floating objects;

[0073] The impact fraction of floating objects on the flood overflow is determined based on the type, size and location information of floating objects.

[0074] Specifically, since floating objects of different types and sizes will produce different resistances and influences on the water flow, this embodiment classifies floating objects through a target monitoring algorithm, for example, determining whether they are wood blocks, branches, plastics, etc., and further determining their specific size information, such as the length, width, height and other parameters of the floating objects. Then, according to the type, size and location information of the floating objects, the impact score of the floating objects on the overflow is calculated. It can be understood that the impact score in this embodiment is a quantitative indicator used to reflect the potential interference degree of floating objects on the overflow. For example, larger floating objects or heavy objects may cause greater water flow obstruction in the spillway, resulting in a reduction in the overflow. The location information of the floating objects will also affect their effect on the overflow. For example, if the floating objects are concentrated in the key positions of the spillway, they may cause local blockage of the water flow or changes in the flow rate. Based on these characteristics of the floating objects, an impact score is calculated through preset rules and machine learning algorithms.

[0075] Furthermore, the surrounding environment image includes image data of the spillway structure and the surrounding area.

[0076] Structural anomaly detection operations include:

[0077] Extracting structural features from the surrounding environment image through a convolutional neural network; the structural features include crack location, crack size, corrosion spots, surface damage points, and deformation features;

[0078] The deep learning algorithm is used to identify the abnormal state of the structure based on the extracted structural features and generate corresponding abnormal judgment results;

[0079] The abnormal judgment results include the health score of the structure, and the impact score of the current structural state on the flood overflow is determined according to the health score.

[0080] It is understandable that the structural health of the spillway and its surroundings directly affects its spillover function and ability to divert water flow. For example, cracks or damage may cause water flow obstruction and affect the amount of water flowing through, while corrosion spots and deformation may cause the structural strength to decrease and even endanger the stability of the entire spillway. Therefore, in this embodiment, by discovering these anomalies in advance and evaluating their impact on the amount of spillage, dangerous situations caused by structural problems in the spillway can be avoided.

[0081] Furthermore, the flood overflow prediction operation includes:

[0082] The preliminary flood overflow prediction value is calculated based on the predicted water level value and the relationship formula between water level and flood overflow;

[0083] Based on the impact fraction of floating objects on the overflow volume and the impact fraction of the current structural status on the overflow volume, the preliminary overflow volume prediction value is corrected to obtain the final overflow volume prediction value.

[0084] As a preferred embodiment, the relationship between water level and flood overflow is:

[0085]

[0086] Where Q(t) represents the initial overflow; α is the global impact coefficient; W(t) is the water level at time t; b(W(t)) is the width of the spillway, indicating how the width changes with the water level; γ 1 is the coefficient of relationship between water level and flow velocity; W(t) n is the nonlinear relationship between water level and flow velocity, and n is an exponent; is the weighted historical water level, W(ti) is the water level at the historical moment ti, and w i is the weighting coefficient of historical water level.

[0087] The flood overflow correction formula is:

[0088] Q final (t) = Q(t)·(1+γ float ·S float +γ struct ·S struct )

[0089] Among them, Q final (t) is the corrected overflow volume; γ float is the influence coefficient of floating objects on the flood overflow, and the flood overflow is adjusted according to the type, quantity and location of floating objects; S float is the impact fraction of floating objects on the flood overflow, which is calculated based on factors such as the location, type, and size of the floating objects and is a value between 0 and 1; struct S is the structural state influence coefficient; struct It is the impact score of the structural status on the flood overflow, which is a value between 0 and 1.

[0090] Further, the system also includes an alarm unit;

[0091] Before executing the flood overflow prediction operation, it also includes setting an impact score safety threshold. When it is determined that the impact score of floating objects on the flood overflow and / or the impact score of the current structural state on the flood overflow is greater than the impact score safety threshold, an alarm message is generated, and an alarm is issued according to the alarm message through the alarm unit.

[0092] Preferably, the system further comprises an adjustment unit for adjusting the monitoring strategy according to the water level prediction results, floating object identification results, structural anomaly detection results and flood overflow prediction results, and the monitoring camera unit performs monitoring operations according to the adjusted monitoring strategy.

[0093] During this operation, if the system detects that the overflow volume may exceed the set safety threshold or abnormal fluctuations occur, the adjustment unit will trigger the adjustment of the monitoring strategy. For example, the focus, angle, shooting frequency and other parameters of the monitoring camera unit may be reset according to the current water flow conditions to ensure that key areas (such as certain key locations or dangerous areas of the spillway) are monitored more closely. If the forecast shows that the overflow volume is small or the structure is in good condition, the monitoring strategy may be adjusted to a lower frequency of routine monitoring to save resources and improve the efficiency of the system. The core purpose of this process is to realize the intelligence of monitoring operations, adjust the monitoring method based on changes in real-time data, thereby improving the accuracy and response speed of spillway monitoring and ensuring the safe operation of the reservoir spillway.

[0094] Embodiment 2

[0095] The second aspect of the present invention discloses a visual recognition method for monitoring water flow through a reservoir spillway. Figure 2 , Figure 2 1 is a flow chart of a visual recognition method for monitoring water flow through a reservoir spillway disclosed in another embodiment of the present invention, the method comprising:

[0096] A fixed wide-angle camera and a rotatable pan-tilt camera are provided, and the fixed wide-angle camera and the rotatable pan-tilt camera are installed on the spillway of the reservoir and the surrounding area, and the water flow image of the spillway and the surrounding environment image are collected by the fixed wide-angle camera and the rotatable pan-tilt camera as the first image data;

[0097] A flood overflow prediction operation is performed based on the first image data; before performing the flood overflow prediction operation, water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction result, the floating object recognition result and the structural anomaly detection result.

[0098] Furthermore, a water level monitoring unit is provided to monitor the water level, and the water level data is transmitted to the visual recognition processing unit through the data transmission unit;

[0099] The water level prediction operation includes predicting the water level change trend through a water level prediction model based on spillway water flow image data, water level data and meteorological data.

[0100] Furthermore, the process of establishing the water level prediction model includes:

[0101] Obtain historical spillway water flow image data, water level data, and meteorological data, and extract water surface features from the spillway water flow images through a convolutional neural network; the water surface features include contours, waves, splashes, and flow states;

[0102] A water level prediction model is constructed through a deep learning network, and water surface characteristics, water level data, and meteorological data are used as training data sets to train the water level prediction model.

[0103] Furthermore, the deep learning network includes a fusion layer, and the fusion layer sets a time series modeling network. The historical water level data and meteorological data are processed through the time series modeling network. The processing process includes:

[0104] Performing data preprocessing operations on historical water level data and meteorological data to obtain first processed data;

[0105] Performing feature extraction operations on the first processed data through a gated recurrent unit GRU network to obtain time series features of historical water level data and meteorological data, and obtaining time series dependencies between the time series features through a gating structure in the GRU network, and learning dynamic change laws of historical water level data and meteorological data based on the obtained time series dependencies;

[0106] Outputting second processed data through a gated recurrent unit GRU network; the second processed data includes time series characteristics and dynamic change rules of historical water level data and meteorological data;

[0107] The time series features of the historical meteorological data in the second processed data are fused with the water surface features to obtain fused features;

[0108] The deep learning network also includes a prediction layer, which sets a regression network, inputs fusion features and dynamic change rules into the prediction layer, and outputs water level prediction values ​​through the prediction layer.

[0109] Furthermore, the fusion features and dynamic change rules are input into the prediction layer, and the water level prediction values ​​output by the prediction layer include:

[0110] Perform weighted processing on the fusion features based on the dynamic change rules to generate weighted features;

[0111] The weighted features are input into the regression network, and the weighted features are subjected to multi-layer nonlinear transformation through the regression network. The preliminary estimated water level prediction value is output in combination with the temporal dependency of the weighted features; wherein the preliminary estimated water level prediction value carries timestamp information;

[0112] The water level deviation between the preliminary estimated water level prediction value and the time series characteristics of the historical water level data in the second processed data is analyzed, the water level deviation is fed back to the water level prediction model, and the parameters of the water level prediction model are adjusted based on the water level deviation; the time series characteristics of the historical water level data in the second processed data include timestamp information and the corresponding water level; the timestamp information of the preliminary estimated water level prediction value is consistent with the timestamp information of the time series characteristics of the historical water level data in the second processed data.

[0113] Furthermore, the floating object identification operation includes:

[0114] Extracting floating object features from the spillway water flow image through a convolutional neural network; the floating object features include the floating object position, floating object shape, and floating object movement trajectory;

[0115] Identify the characteristics of floating objects through target monitoring algorithms to determine the type and size of floating objects;

[0116] The impact fraction of floating objects on the flood overflow is determined based on the type, size and location information of floating objects.

[0117] Further, the surrounding environment image includes image data of the spillway structure and the surrounding area;

[0118] Structural anomaly detection operations include:

[0119] Extracting structural features from the surrounding environment image through a convolutional neural network; the structural features include crack location, crack size, corrosion spots, surface damage points, and deformation features;

[0120] The deep learning algorithm is used to identify the abnormal state of the structure based on the extracted structural features and generate corresponding abnormal judgment results;

[0121] The abnormal judgment results include the health score of the structure, and the impact score of the current structural state on the flood overflow is determined according to the health score.

[0122] Furthermore, the flood overflow prediction operation includes:

[0123] The preliminary flood overflow prediction value is calculated based on the predicted water level value and the relationship formula between water level and flood overflow;

[0124] Based on the impact fraction of floating objects on the overflow volume and the impact fraction of the current structural status on the overflow volume, the preliminary overflow volume prediction value is corrected to obtain the final overflow volume prediction value.

[0125] Furthermore, before executing the flood overflow prediction operation, it also includes setting an impact score safety threshold. When it is determined that the impact score of floating objects on the flood overflow and / or the impact score of the current structural state on the flood overflow is greater than the impact score safety threshold, an alarm message is generated and an alarm operation is performed based on the alarm message.

[0126] It should be noted that the specific implementation process of Example 5 is similar to that of Examples 1 and 2, and will not be repeated in this example.

[0127] Finally, it should be noted that the visual recognition system and method for monitoring water flow through a reservoir spillway disclosed in the embodiment of the present invention only discloses a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features therein may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A reservoir spillway water monitoring visual recognition system, characterized in that: The system includes a monitoring camera unit, a data transmission unit, and a visual recognition processing unit; wherein, The monitoring camera unit includes a fixed wide-angle camera and a rotatable pan-tilt camera, which are installed in the spillway of the reservoir and the surrounding area, and are used to collect images of water flowing through the spillway and surrounding environment as first image data; The data transmission unit is used to transmit the first image data to the visual recognition processing unit in real time; The visual recognition processing unit is used to perform a flood overflow prediction operation based on the first image data; before performing the flood overflow prediction operation, water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction results, floating object recognition results and structural anomaly detection results.

2. The reservoir spillway water flow monitoring visual recognition system according to claim 1 is characterized in that: The system further comprises a water level monitoring unit for monitoring the water level, and transmitting the water level data to the visual recognition processing unit through the data transmission unit; The water level prediction operation includes predicting the water level change trend through a water level prediction model based on spillway water flow image data, water level data and meteorological data.

3. The reservoir spillway water flow monitoring visual recognition system according to claim 2 is characterized in that: The process of establishing the water level prediction model includes: Obtain historical spillway water flow image data, water level data, and meteorological data, and extract water surface features from the spillway water flow images through a convolutional neural network; the water surface features include contours, waves, splashes, and flow states; A water level prediction model is constructed through a deep learning network, and water surface characteristics, water level data, and meteorological data are used as training data sets to train the water level prediction model.

4. The reservoir spillway water flow monitoring visual recognition system according to claim 3 is characterized in that: The deep learning network includes a fusion layer, and the fusion layer is provided with a time series modeling network. The historical water level data and meteorological data are processed by the time series modeling network. The processing process includes: Performing data preprocessing operations on historical water level data and meteorological data to obtain first processed data; Performing feature extraction operations on the first processed data through a gated recurrent unit GRU network to obtain time series features of historical water level data and meteorological data, and obtaining time series dependencies between the time series features through a gating structure in the GRU network, and learning dynamic change laws of historical water level data and meteorological data based on the obtained time series dependencies; Outputting second processed data through a gated recurrent unit GRU network; the second processed data includes time series characteristics and dynamic change rules of historical water level data and meteorological data; The time series features of the historical meteorological data in the second processed data are fused with the water surface features to obtain fused features; The deep learning network also includes a prediction layer, which sets a regression network, inputs fusion features and dynamic change rules into the prediction layer, and outputs water level prediction values ​​through the prediction layer.

5. The reservoir spillway water flow monitoring visual recognition system according to claim 4 is characterized in that: The step of inputting the fusion features and the dynamic change rules into the prediction layer and outputting the water level prediction value through the prediction layer specifically includes: Perform weighted processing on the fusion features based on the dynamic change rules to generate weighted features; The weighted features are input into the regression network, and the weighted features are subjected to multi-layer nonlinear transformation through the regression network. The preliminary estimated water level prediction value is output in combination with the temporal dependency of the weighted features; wherein the preliminary estimated water level prediction value carries timestamp information; The water level deviation between the preliminary estimated water level prediction value and the time series characteristics of the historical water level data in the second processed data is analyzed, the water level deviation is fed back to the water level prediction model, and the parameters of the water level prediction model are adjusted based on the water level deviation; the time series characteristics of the historical water level data in the second processed data include timestamp information and the corresponding water level; the timestamp information of the preliminary estimated water level prediction value is consistent with the timestamp information of the time series characteristics of the historical water level data in the second processed data.

6. The reservoir spillway water flow monitoring visual recognition system according to claim 2 is characterized in that: The floating object identification operation includes: Extracting floating object features from the spillway water flow image through a convolutional neural network; the floating object features include the floating object position, floating object shape, and floating object movement trajectory; Identify the characteristics of floating objects through target monitoring algorithms to determine the type and size of floating objects; The impact fraction of floating objects on the flood overflow is determined based on the type, size and location information of floating objects.

7. The reservoir spillway water flow monitoring visual recognition system according to claim 6 is characterized in that: The surrounding environment image includes image data of the spillway structure and the surrounding area; The structural anomaly detection operation includes: Extracting structural features from the surrounding environment image through a convolutional neural network; the structural features include crack location, crack size, corrosion spots, surface damage points, and deformation features; The deep learning algorithm is used to identify the abnormal state of the structure based on the extracted structural features and generate corresponding abnormal judgment results; The abnormal judgment results include the health score of the structure, and the impact score of the current structural state on the flood overflow is determined according to the health score.

8. The reservoir spillway water flow monitoring visual recognition system according to claim 7 is characterized in that: The flood overflow prediction operation includes: The preliminary flood overflow prediction value is calculated based on the predicted water level value and the relationship formula between water level and flood overflow; Based on the impact fraction of floating objects on the overflow volume and the impact fraction of the current structural state on the overflow volume, the preliminary overflow volume prediction value is corrected to obtain the final overflow volume prediction value.

9. The reservoir spillway water flow monitoring visual recognition system according to claim 8, characterized in that: The system also includes an alarm unit; Before executing the flood overflow prediction operation, it also includes setting an impact score safety threshold. When it is determined that the impact score of floating objects on the flood overflow and / or the impact score of the current structural state on the flood overflow is greater than the impact score safety threshold, an alarm message is generated, and an alarm is issued according to the alarm message through the alarm unit.

10. A visual recognition method for monitoring water flow through a reservoir spillway, the method being applied to the system according to any one of claims 1 to 9, characterized in that: The method comprises: A fixed wide-angle camera and a rotatable pan-tilt camera are provided, and the fixed wide-angle camera and the rotatable pan-tilt camera are installed on the spillway of the reservoir and the surrounding area, and the water flow image of the spillway and the surrounding environment image are collected by the fixed wide-angle camera and the rotatable pan-tilt camera as the first image data; A flood overflow prediction operation is performed based on the first image data; before performing the flood overflow prediction operation, water level prediction, floating object recognition and structural anomaly detection operations are performed respectively, and the flood overflow prediction is performed based on the water level prediction result, the floating object recognition result and the structural anomaly detection result.

Images