Dam abnormal behavior early warning method and system based on neural network algorithm

Abstract

The invention discloses a dam abnormal behavior early warning method and system based on a neural network algorithm, and relates to the technical field of dam early warning, and the method comprises the following steps: S1, obtaining historical data, extracting the features of the historical data, simulating a scene result according to the features, and carrying out verification through back propagation; s2, collecting a video scene according to monitoring, performing stream extraction on the video scene to obtain image information, extracting features of the image information, and outputting a detection result; and S3, judging the condition of the detection result, and outputting the detection result to an early warning platform. According to the method, judgment is carried out according to the compared error degree data, the time of the stream taking interval can be adaptively adjusted based on the judgment result, different interval durations are adopted under the condition of different scene results, and compared with a regular detection mode in the past, the stream taking interval can be increased under the safe condition, so that the detection efficiency is improved. Occupation of a database is reduced, the flow taking interval is shortened under the dangerous condition, and the safety of the dam is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of dam early warning technology, and in particular to a method and system for early warning of abnormal dam behavior based on neural network algorithms. Background Technology

[0002] Water resources are a fundamental resource related to the national economy and people's livelihood. my country faces water problems such as water shortage, aggravated water pollution, and serious soil erosion. How to rationally and effectively protect and comprehensively utilize water resources has always been a key focus of water conservancy departments at all levels. Reservoirs and dams, as important water conservancy projects, play a significant role in engineering benefits and bear the heavy responsibility of supporting people's daily lives and industrial and agricultural production.

[0003] In actual inspections, due to factors such as the wide inspection scope, numerous equipment, heavy workload, and uneven experience and skill levels of inspection personnel, there are problems such as non-standard inspections, low inspection efficiency, missed inspections, and misjudgments. Therefore, by using AI algorithms, the shortcomings of manual monitoring can be effectively compensated for, and the intensity of personnel patrol work can be reduced. Summary of the Invention

[0004] In view of the problems existing in the current dam abnormal behavior early warning and system based on neural network algorithms, this invention is proposed.

[0005] Therefore, the problem that this invention aims to solve is the inability to effectively address the early warning of abnormal dam behavior.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: different calculation intervals are adopted according to different dam conditions, which reduces the calculation process while ensuring the safety of the dam and effectively improves the calculation speed.

[0007] In a first aspect, embodiments of the present invention provide a method for early warning of abnormal dam behavior based on a neural network algorithm, which includes the following steps.

[0008] Historical data is acquired, and features of the historical data are extracted. The results of the simulated scenario are then verified through backpropagation.

[0009] Based on the video scene captured by the monitoring, the video scene is streamed to obtain image information, the features of the image information are extracted, and the detection results are output.

[0010] The results of the tests are assessed and output to the early warning platform.

[0011] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, the historical data includes a training set and a corresponding validation set;

[0012] The training set includes photos of rivers, lakes, and mountains.

[0013] The validation set includes scene results corresponding to river photos, scene results corresponding to lake photos, and scene results corresponding to mountain photos.

[0014] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, the method for extracting features from the historical data includes:

[0015] The input images are classified, and features are extracted from different categories of images to obtain different detail images;

[0016] The detailed image is cropped to output different shape images;

[0017] Input the shape diagram into the loss function calculation result;

[0018] Backpropagation is performed to update the weight parameters.

[0019] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, wherein: 4. the loss function includes a bounding box regression loss function, a confidence prediction loss function, and a category prediction loss function;

[0020] The bounding box regression loss function is as follows:

[0021]

[0022] The confidence prediction loss function is:

[0023]

[0024] The category prediction loss function is:

[0025]

[0026] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, the video scene acquisition method includes:

[0027] A frame of the video scene is extracted and fed into the Backbone network to extract features and obtain a feature map.

[0028] Crop the feature map to output maps of different sizes;

[0029] The size diagram and the shape diagram are compared, and the next flow interval is determined based on the error rate of the comparison.

[0030] When the error rate of the comparison is less than 30% and less than or equal to 50%, the query validation set displays the scenario results. If the scenario result is abnormal, the next flow retrieval interval is 4 hours. If the scenario result is normal, the next flow retrieval interval is 20 hours.

[0031] When the error rate of the comparison is ≤30%, the query validation set displays the scenario results. If the scenario result is abnormal, the next stream retrieval interval is 0.5h. If the scenario result is normal, the next stream retrieval interval is 24h.

[0032] When the error rate of the comparison is greater than 50%, the next sampling interval will be 12 hours.

[0033] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, the extraction of image features from the Backbone network includes the following steps:

[0034] Clean the image to remove inconsistencies caused by lighting, shadows, or weather, and ensure image quality;

[0035] Enhance the features in the image and annotate the features;

[0036] Feature annotation is divided into normal regions and abnormal regions.

[0037] As a preferred embodiment of the dam abnormal behavior early warning method based on neural network algorithm described in this invention, the early warning platform operation steps include periodic analysis requests and real-time analysis requests:

[0038] The specific steps of the real-time analysis request include: after the user submits the real-time analysis request on the client, the analysis image is inserted into the analysis module. After receiving the analysis request, n sets of images with similar features are extracted from the storage module for intelligent analysis. The analysis results are pushed to the message queue and displayed on the client.

[0039] The intelligent analysis method is as follows: determine the feature similarity between the analyzed image and n similar images, arrange the n images in order according to the decreasing trend of feature similarity, query the results displayed in the validation set of the n images, and output the analysis results;

[0040] The judgment method is as follows:

[0041] S=frac{sim_(i=1)^(n)w(i)*R(i)}{sim_(i=1)^(n)w(i)}

[0042] In the formula, w represents the weight, sim represents the similarity, R represents the result displayed on the image validation set, S represents the average result, and the weight and similarity are inversely proportional.

[0043] Among them, a threshold Z is set;

[0044] When S ≤ Z, the scene result is normal;

[0045] When S > Z, the scene result is abnormal;

[0046] When the judged scene result is normal and Z - S < Q, output the scene result of this part, and perform a secondary judgment on the scene result of this part. The flow interval duration of this part of the scene result is set to 2h, and finally mark the scene result of this part as an easily abnormal state;

[0047] When the judged scene result is normal and Z - S ≥ Q, the scene result is normal and no special operation is performed on it;

[0048] When the judged scene result is abnormal and S - Z ≥ Q, output the scene result of this part, and perform a secondary judgment on the scene result of this part. The flow interval duration of this part of the scene result is set to 0.5h and an alarm is issued;

[0049] When the judged scene result is abnormal and S - Z < Q, output the scene result of this part, and perform a secondary judgment on the scene result of this part. The flow interval duration of this part of the scene result is set to 1h and an alarm is issued;

[0050] Among them, Q represents the critical value;

[0051] The method of the secondary judgment includes:

[0052] Re - extract the features of the inserted pictures, compare them with the features of the extracted training set, and query the corresponding validation set of the training set.

[0053] In a second aspect, an embodiment of the present invention provides a dam abnormal behavior early warning system based on a neural network algorithm, which includes: a collection module, a storage module, and an analysis module,

[0054] The collection module is used to collect the live situation in real - time, obtain the full - process real - time video, and accurately extract the video stream;

[0055] The storage module is used to store historical data and classify and store pictures according to different features;

[0056] The analysis module is used to compare historical data with real - time video, calculate whether an abnormality will occur, and give an early warning. <Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described dam abnormal behavior early warning method based on a neural network algorithm.

[0058] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described dam abnormal behavior early warning method based on a neural network algorithm.

[0059] The beneficial effects of this invention are:

[0060] 1. The system makes judgments based on the error data and can adaptively adjust the flow sampling interval based on the judgment results. This allows for different interval durations to be used in different scenarios. Compared to the previous periodic detection method, this method increases the flow sampling interval under safe conditions to reduce database usage, and shortens the flow sampling interval under dangerous conditions to ensure dam safety.

[0061] 2. It enables the superposition of multiple detection methods to ensure the accuracy of numerical calculations. Furthermore, the multiple judgment method allows for the selection of different processing methods for different scenario results, making the entire process more complete. It can strengthen the detection of safety locations that are prone to danger and effectively solve the prevention role. Attached Figure Description

[0062] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0063] Figure 1 This is a flowchart of a method for early warning of abnormal dam behavior based on neural network algorithms.

[0064] Figure 2 This is a schematic diagram of the secondary judgment in the dam abnormal behavior early warning method based on neural network algorithm.

[0065] Figure 3 This is a schematic diagram of the threshold Z for an early warning method for abnormal dam behavior based on a neural network algorithm. Detailed Implementation

[0066] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0067] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0068] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0069] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0070] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0071] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0072] Example 1

[0073] Reference Figure 1 and Figure 2This is the first embodiment of the present invention, which provides a method for early warning of abnormal dam behavior based on a neural network algorithm, including the following steps:

[0074] S1. Obtain historical data and extract the features of the historical data. Simulate the scenario results based on the features and then verify them through backpropagation.

[0075] Historical data includes the training set and the corresponding validation set;

[0076] The training set includes photos of rivers, lakes, and mountains.

[0077] The validation set includes scene results corresponding to river photos, scene results corresponding to lake photos, and scene results corresponding to mountain photos;

[0078] The provided historical data can be used to build the model in subsequent steps, and the historical data can be fed back into the model to verify the model's calculations.

[0079] Methods for extracting features from the historical data include:

[0080] The input images are classified, and features are extracted from different categories of images to obtain different detail images;

[0081] The detailed image is cropped to output different shape images;

[0082] Input the shape diagram into the loss function calculation result;

[0083] The loss functions include bounding box regression loss function, confidence prediction loss function, and category prediction loss function;

[0084] The bounding box regression loss function is as follows:

[0085]

[0086] The confidence prediction loss function is:

[0087]

[0088] The category prediction loss function is:

[0089]

[0090] Backpropagation is performed to update the weight parameters.

[0091] S2. Based on the video scene captured by the monitoring, extract the video scene stream to obtain image information, extract the features of the image information, and output the detection result.

[0092] Video scene capture methods include:

[0093] A frame of the video scene is extracted and fed into the Backbone network to extract features and obtain a feature map.

[0094] Extracting image features from the Backbone network includes the following steps:

[0095] Clean the image to remove inconsistencies caused by lighting, shadows, or weather, and ensure image quality;

[0096] Enhance the features in the image and annotate the features;

[0097] Feature annotation is divided into normal regions and abnormal regions.

[0098] Crop the feature map to output maps of different sizes;

[0099] The size diagram and the shape diagram are compared, and the next flow interval is determined based on the error rate of the comparison.

[0100] When the error rate of the comparison is less than 30% and less than or equal to 50%, the query validation set displays the scenario results. If the scenario result is abnormal, the next flow retrieval interval is 4 hours. If the scenario result is normal, the next flow retrieval interval is 20 hours.

[0101] When the error rate of the comparison is ≤30%, the query validation set displays the scenario results. If the scenario result is abnormal, the next stream retrieval interval is 0.5h. If the scenario result is normal, the next stream retrieval interval is 24h.

[0102] When the error rate of the comparison is >50%, the next sampling interval will be 12 hours.

[0103] When calculating the error, the feature maps of both are first extracted, and then any one of the following methods is used: mean square error, peak signal-to-noise ratio, and absolute error. The calculation includes the following steps:

[0104] In the method of calculating mean squared error, the average of the squared differences of each pixel value in the image is calculated:

[0105]

[0106] In the formula, MSE represents the error degree, N represents the height of the image, M represents the width of the image, and I(i,j) and K(i,j) represent the position pixels in the two images;

[0107] In the calculation of peak signal-to-noise ratio, the mean square error of the image is calculated and converted into a metric in decibels.

[0108]

[0109] In the formula, PSNR represents the mean square error of the image, and MAX represents the maximum possible value of the image pixel.

[0110] In the method of calculating absolute error, it is done by averaging the absolute differences of each pixel;

[0111]

[0112] In the formula, AE represents the average value, N represents the height of the image, M represents the width of the image, and I(i,j) and K(i,j) represent the position pixels in the two images;

[0113] In dangerous situations, the flow intake interval can be shortened, and the actual situation of the dam can be recalculated each time to reduce the occurrence of accidents. When the dam is in normal condition, the flow intake interval can be extended to reduce the use of the database. Furthermore, when the dam is at a critical value that is about to become dangerous, the flow intake interval can be shortened compared to normal conditions, which can effectively play an early warning role and prevent further deterioration.

[0114] S3. Determine the status of the detection results and output them to the early warning platform.

[0115] The operation steps of the early warning platform include periodic analysis requests and real-time analysis requests:

[0116] The specific steps of the real-time analysis request include: after the user submits the real-time analysis request on the client, the analysis image is inserted into the analysis module. After receiving the analysis request, n sets of images with similar features are extracted from the storage module for intelligent analysis. The analysis results are pushed to the message queue and displayed on the client.

[0117] The intelligent analysis method is as follows: determine the feature similarity between the analyzed image and n similar images, arrange the n images in order according to the decreasing trend of feature similarity, query the results displayed in the validation set of the n images, and output the analysis results;

[0118] The judgment method is as follows:

[0119] S=frac{sim_(i=1)^(n)w(i)*R(i)}{sim_(i=1)^(n)w(i)}

[0120] In the formula, w represents the weight, sim represents the similarity, R represents the result displayed on the image validation set, S represents the average result, and the weight and similarity are inversely proportional.

[0121] After the user submits an on-demand analysis request on the client side, the analysis module analyzes the inserted pictures, generates various features of the pictures, and finds pictures with similar main features. In this embodiment, the optimal number is to find 5 groups of pictures, which specifically includes the following steps:

[0122] Table 1: Compare pictures with different quantities and judge the time consumption and accuracy

[0123] Number of images Time consumption Accuracy 2 groups 3.5s 74.7％ 3 groups 4.9s 83.3％ 4 groups 5.5s 90.1％ 5 groups 5.9s 97.6％ 6 groups 7.2s 97.9％ 7 groups 9.7s 98.0％

[0124] As can be seen from the above table, the time consumption of each calculation increases slowly. However, after more than 5 groups of pictures, for example, when changing from 5 groups of pictures to 6 groups of pictures, the total time consumption increases by 1.3 s, and when changing from 6 groups of pictures to 7 groups of pictures, the total time consumption increases by 1.5 s. When the number of pictures is within 5 groups, when changing from 4 groups of pictures to 5 groups of pictures, the total time consumption increases by 0.4 s. Therefore, when there are 5 groups of pictures, the time consumption is at the critical point of the increasing threshold. Similarly, after exceeding the 5th group of pictures, the increasing trend of the accuracy decreases significantly. Therefore, in this embodiment, the optimal number is to find 5 groups of pictures.

[0125] Among them, set a threshold Z and set the threshold Z to 0.63;

[0126] In this embodiment, when taking the value of the threshold Z, it includes the following steps:

[0127] Take 1000 groups of pictures, compare the calculated value S with each threshold Z once. The value range of the threshold Z is 0 - 1. Calculate the size of the calculated value S and the threshold Z, and output the scenario result according to their sizes. Then judge whether the output scenario structure is normal according to the actual scenario result until the maximum value of the threshold Z. The specific change graph of the threshold Z is as shown in the attached Figure 3 description.

[0128] When S ≤ Z, the scenario result is normal, and Z - S < Q. Output the scenario result of this part, and make a secondary judgment on the scenario result of this part. The flow interval duration of this part of the scenario result is set to 2 h, and finally mark the scenario result of this part as an easily abnormal state; when Z - S ≥ Q, the scenario result is normal and no special operation is performed on it.

[0129] When S > Z, the scenario result is abnormal, and S - Z ≥ Q. Output the scenario result of this part, and make a secondary judgment on the scenario result of this part. The flow interval duration of this part of the scenario result is set to 0.5 h and an alarm is issued; when the judged scenario result is abnormal and S - Z < Q, output the scenario result of this part, and make a secondary judgment on the scenario result of this part. The flow interval duration of this part of the scenario result is set to 1 h and an alarm is issued.

[0130] Among them, Q represents the critical value;

[0131] The critical value Q is determined using an adaptive thresholding method, which includes the following steps:

[0132] [T(x,y)=\mu_{G}(x,y)-C

[0133] In the formula, \mu_{G}(x,y) represents the neighborhood average of pixel (x,y) weighted by Gaussian. The overall calculation method is to multiply each pixel value in the local area by the Gaussian weight and sum them up. C represents a pre-set constant.

[0134] The Gaussian weights are related to the Euclidean distance from the center, calculated using a two-dimensional Gaussian function:

[0135] W(i,j)=\frac{1}{2\pi\sigma^2}e^[-\frac{i^2+j^2}{2\sigma^2}]

[0136] In the formula, (i,j) represents the coordinates relative to the neighborhood center, and sigma represents the standard deviation of the Gaussian kernel, which is used to control the width of the Gaussian distribution;

[0137] By using a two-step judgment method, calculation errors can be effectively avoided. During the second judgment, the time of the inserted image is obtained. Based on the result of the second judgment, the duration of the stream interval is adjusted. Then, the stream interval is regenerated for the captured video scene. According to the calculation method in the above steps, the subsequent judgment result is generated. By comparing the two judgment results, it is convenient for operators to make timely adjustments, reduce the occurrence of accidents, and provide a certain degree of protection for the dam.

[0138] The secondary determination method includes:

[0139] Re-extract the features of the inserted image, compare them with the features extracted from the training set, and query the validation set corresponding to the training set.

[0140] In summary, based on the comparative error data, and based on the judgment results, the flow sampling interval can be adaptively adjusted. This allows for different interval lengths to be used under different scenario conditions. Compared to the previous periodic detection method, this method increases the flow sampling interval under safe conditions to reduce database usage, and shortens the flow sampling interval under dangerous conditions to ensure dam safety. The superposition of multiple detection methods ensures the accuracy of numerical calculations, and the multiple judgment method allows for the selection of different processing methods for different scenario results. The entire process is more complete, enabling enhanced detection of potentially dangerous locations and effectively addressing preventative measures.

[0141] Example 2

[0142] Building upon the first embodiment, this embodiment further provides a dam abnormal behavior early warning system based on a neural network algorithm, including a data acquisition module, a storage module, and an analysis module.

[0143] The acquisition module is used to acquire real-time data, obtain full real-time video recording, and accurately extract the stream from the real-time video recording;

[0144] The storage module is used to store historical data and classify and store images according to different characteristics;

[0145] The analysis module is used to compare historical data and real-time video recordings, calculate whether any anomalies will occur, and issue an early warning.

[0146] This embodiment also provides a computer device applicable to the dam abnormal behavior early warning method based on neural network algorithm, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the dam abnormal behavior early warning method based on neural network algorithm proposed in the above embodiment.

[0147] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0148] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the method for early warning of abnormal dam behavior based on a neural network algorithm as proposed in the above embodiments.

[0149] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0150] Example 3

[0151] Based on the previous two embodiments, this embodiment provides a method for early warning of abnormal dam behavior based on a neural network algorithm. To verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0152] Existing technology 1: Using inspection personnel to collect information on the condition of the dam and make judgments;

[0153] Prior art 2: It adopts the threshold Z calculation method in embodiment 1, but does not use secondary judgment;

[0154] Verification method: Input 100 video clips, collect information from these 100 video clips, calculate the judgment result, compare it with the actual result, and calculate the accuracy.

[0155] Time consumption accuracy Early warning efficiency Prior Art 1 3.5h 79.6％ 68.6％ Existing technology 2 12s 92.1％ 87.6％ Example 1 14s 98.4％ 98.9％

[0156] As can be seen from the table above, because a secondary judgment method is set, the time period of the inserted image can be obtained, and subsequent videos can be found in the data. Based on the calculation result of the secondary judgment, a new streaming interval is generated. The new streaming interval is compared with the initial streaming interval result. When the results are consistent, no special processing is required. However, when the results are different, the video needs to be marked and a warning reminder needs to be given to the staff, thereby increasing the efficiency of the warning and further improving the accuracy to avoid accidents at the dam.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of abnormal dam behavior based on neural network algorithms, characterized in that: Includes the following steps, Historical data is acquired, and features of the historical data are extracted. The results of the simulated scenario are then verified through backpropagation. Based on the video scene captured by the monitoring, the video scene is streamed to obtain image information, the features of the image information are extracted, and the detection results are output. The results of the tests are assessed and output to the early warning platform.

2. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 1, characterized in that: The historical data includes the training set and the corresponding validation set; The training set includes photos of rivers, lakes, and mountains. The validation set includes scene results corresponding to river photos, scene results corresponding to lake photos, and scene results corresponding to mountain photos.

3. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 2, characterized in that: Methods for extracting features from the historical data include: The input images are classified, and features are extracted from different categories of images to obtain different detail images; The detailed image is cropped to output different shape images; Input the shape diagram into the loss function calculation result; Backpropagation is performed to update the weight parameters.

4. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 3, characterized in that: The loss functions include bounding box regression loss function, confidence prediction loss function, and category prediction loss function; The bounding box regression loss function is as follows: The confidence prediction loss function is: The category prediction loss function is:

5. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 4, characterized in that: The video scene capture method includes: A frame of the video scene is extracted and fed into the Backbone network to extract features and obtain a feature map. Crop the feature map to output maps of different sizes; The size diagram and the shape diagram are compared, and the next flow interval is determined based on the error rate of the comparison. When the error rate of the comparison is less than 30% and less than or equal to 50%, the query validation set displays the scenario results. If the scenario result is abnormal, the next flow retrieval interval is 4 hours. If the scenario result is normal, the next flow retrieval interval is 20 hours. When the error rate of the comparison is ≤30%, the query validation set displays the scenario results. If the scenario result is abnormal, the next stream retrieval interval is 0.5h. If the scenario result is normal, the next stream retrieval interval is 24h. When the error rate of the comparison is greater than 50%, the next sampling interval will be 12 hours.

6. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 5, characterized in that: Extracting image features from the Backbone network includes the following steps: Clean the image to remove inconsistencies caused by lighting, shadows, or weather, and ensure image quality; Enhance the features in the image and annotate the features; Feature annotation is divided into normal regions and abnormal regions.

7. The dam abnormal behavior early warning method based on neural network algorithm as described in claim 6, characterized in that: The operation steps of the early warning platform include periodic analysis requests and real-time analysis requests: The specific steps of the real-time analysis request include: after the user submits the real-time analysis request on the client, the analysis image is inserted into the analysis module. After receiving the analysis request, n sets of images with similar features are extracted from the storage module for intelligent analysis. The analysis results are pushed to the message queue and displayed on the client. The intelligent analysis method is as follows: determine the feature similarity between the analyzed image and n similar images, arrange the n images in order according to the decreasing trend of feature similarity, query the results displayed in the validation set of the n images, and output the analysis results; The judgment method is as follows: S = \frac{\sum_{i = 1}^{n} w(i) * R(i)}{\sum_{i = 1}^{n} w(i)} In the formula, w represents the weight, sim represents the similarity, R represents the result shown in the picture verification set, S represents the average result, and the weight and similarity are set inversely; Among them, a threshold Z is set; When S ≤ Z, the scenario result is normal; When S > Z, the scenario result is abnormal; When the judged scenario result is normal and Z - S < Q, output the scenario result of this part, and perform a secondary determination on the scenario result of this part. The flow interval duration of the scenario result of this part is set to 2h, and finally mark the scenario result of this part as an easily abnormal state; When the judged scenario result is normal and Z - S ≥ Q, the scenario result is normal and no special operation is performed on it; When the judged scenario result is abnormal and S - Z ≥ Q, output the scenario result of this part, and perform a secondary determination on the scenario result of this part. The flow interval duration of the scenario result of this part is set to 0.5h and an alarm is issued; When the judged scenario result is abnormal and S - Z < Q, output the scenario result of this part, and perform a secondary determination on the scenario result of this part. The flow interval duration of the scenario result of this part is set to 1h and an alarm is issued; Among them, Q represents the critical value; The method of the secondary determination includes: Re - extract the features of the inserted pictures, compare them with the features of the extracted training set, and query the corresponding verification set of the training set.

8. A dam abnormal behavior early warning system based on a neural network algorithm, based on the dam abnormal behavior early warning method based on a neural network algorithm according to any one of claims 1 to 7, characterized in that: It includes a collection module, a storage module, and an analysis module, The collection module is used to collect the live situation in real - time, obtain the full - process real - time video, and accurately extract the video stream; The storage module is used to store historical data and classify and store pictures according to different features; The analysis module is used to compare historical data and real - time video, calculate whether an abnormality will occur, and give an early warning.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the dam abnormal behavior early - warning method based on the neural network algorithm according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the dam abnormal behavior early - warning method based on the neural network algorithm according to any one of claims 1 to 7.

Images