An automatic identification method for siltation of a sluice front channel of a sluice by fusing unmanned aerial vehicle aerial photography

Abstract

This invention discloses an automatic identification method for siltation in the inlet channel of a sluice gate, integrating drone aerial photography. Relating to the field of image recognition, this invention automatically optimizes flight parameters through a dynamic feedback adjustment mechanism, avoiding the subjectivity and inefficiency of traditional manual adjustments, and significantly improving operational accuracy and stability. Simultaneously, it constructs dedicated segmentation models for different hydrological conditions, effectively addressing complex environmental challenges through a dual-model fusion processing strategy, greatly improving the accuracy and robustness of siltation identification. Furthermore, it integrates an intelligent time-consuming optimization mechanism, automatically selecting the optimal segmentation accuracy through experimental data-driven methods, minimizing processing time while ensuring monitoring quality. This provides reliable technical support for the efficient operation and maintenance management of hydraulic structures, truly realizing an intelligent and standardized operation mode from data acquisition to output.

Description

Technical Field

[0001] This invention belongs to the field of image recognition, and more specifically, it relates to an automatic identification method for siltation in the inlet channel of a sluice gate that integrates drone aerial photography. Background Technology

[0002] Traditional methods for monitoring siltation in sluice gate inlet channels suffer from numerous bottlenecks, including low efficiency, limited accuracy, poor security, inability to conduct large-scale rapid surveys, and insufficient data timeliness. These limitations make it difficult to meet the refined and intelligent operation and maintenance management needs of modern water conservancy projects. Specifically, the following key technical problems exist: 1. The dilemma of low measurement efficiency and long cycle: Traditional siltation monitoring mainly relies on manual boat surveys (such as using sounding rods, sounding hammers, etc.) or contact-based mobile surveys (such as ADCP-Acoustic Doppler Current Profiler). These methods require personnel or vessels to enter the water area for operation. Due to factors such as weather, water flow, and navigation conditions, the preparation time for operation is long and the time for a single measurement is long, making it impossible to achieve high-frequency and rapid response monitoring. Although ADCP is more efficient than purely manual measurement, it still requires operation on the water surface. For large areas such as sluice gates and canals of large rivers, a comprehensive measurement still requires a lot of time, manpower and resources. 2. The contradiction between data accuracy and coverage: Traditional methods often involve discrete point measurements or cross-sectional measurements, making it difficult to obtain continuous and complete siltation distribution data for the entire upstream diversion channel area. For example, some methods calculate the siltation area by measuring the water depth at a limited number of points and comparing it with the design cross-section, or by measuring the siltation thickness at a single point using a specific navigation device and then integrating the data to form the cross-sectional shape. This approach may result in the omission of severely silted areas, failing to accurately reflect the spatial distribution characteristics of siltation, thus affecting the accuracy of siltation volume calculation and the scientific nature of dredging decisions. 3. Safety hazards in high-risk areas: The water flow conditions in the diversion channel area in front of the sluice gate may be complex. Traditional manual boat surveying or some mobile surveying methods require staff to work on-site in the water, which poses certain safety risks, especially during the flood season or when the water flow is rapid. In addition, the steep slopes in some diversion channel areas also pose challenges to traditional surveying methods. 4. Lag in data processing and output: The raw data obtained by traditional methods usually requires a lot of manual processing, interpretation and calculation to generate siltation distribution maps or volume reports. This process is not only cumbersome, but also heavily dependent on the experience of operators, highly subjective, and has a long cycle from data collection to final output. It cannot meet the needs of real-time or near real-time monitoring of siltation status and timely sluice gate scheduling and dredging maintenance. 5. Limitations of existing technologies in high-precision, wide-range, non-contact measurement: Some underwater sonar methods (such as image sonar and multibeam sonar) or laser scanning methods mentioned in the "Technical Specification for Underwater Inspection of Water Conservancy and Hydropower Projects" are highly accurate, but they are often expensive and require highly specialized operation. They also face problems of efficiency, cost, or adaptability to water areas, making them difficult to use as a routine and universal means of siltation monitoring for sluice gates and canals. Summary of the Invention

[0003] To address the problems in related technologies, this invention proposes an automatic identification method for siltation in the inlet channel of a sluice gate that integrates drone aerial photography, in order to overcome the aforementioned technical problems existing in the existing related technologies.

[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to an automatic identification method for siltation in the inlet channel of a sluice gate, which integrates drone aerial photography and includes the following steps: S1. Construct flight path quality characterization indicators and image quality influencing factors for identifying siltation in the inlet channel of a sluice gate using UAV aerial photography, and set the range of each indicator. S2. Based on S1, collect multiple sets of historical flight route quality characterization indicators and image quality influencing factor data, and construct the final aerial flight route characterization indicator mapping model. S3. Input the current image quality influencing factor data into the mapping model in S2 for mapping; if the mapping result does not meet the range of each indicator in S1, adjust the current image quality influencing factor data until it meets the range; otherwise, no adjustment is needed. S4. Collect multiple sets of siltation image data and label them to train the turbulent water flow segmentation model and the calm water area segmentation model respectively; and carry out aerial photography of siltation based on the factor data adjusted in S3. After aerial photography, the collected images are stitched and corrected. S5. Based on the initial accuracy parameters, the turbulent water area is segmented, and the turbulent and calm areas are extracted. The turbulent and calm segmentation model in S4 is used to process different areas to obtain double segmentation results and calculate the siltation area and volume. S6. Conduct multiple rounds of experiments, repeating the aerial siltation operations in S3 and S4 as well as S5 in each experiment, collecting time consumption data under different segmentation accuracies, constructing the final siltation calculation time mapping model, and selecting the turbulent image segmentation accuracy with the minimum aerial siltation calculation time based on the model.

[0005] Preferably, step S1 includes the following steps: S11. Define several types of indicators to characterize the quality of UAV flight paths for aerial photography of siltation in the sluice gate's inlet channel, resulting in a set of siltation aerial photography flight path characterization indicator types. These include channel coverage, directional overlap, and lateral overlap. Then, define several types of factors that influence the image quality of aerial photography of siltation in the sluice gate's inlet channel, resulting in a set of siltation aerial photography image quality influencing factors. These include flight path coverage, UAV flight speed, camera shutter speed, camera focal length, camera image size, terrain complexity, flight path spacing, and flight platform stability. S12. Based on the set of siltation aerial photography route characterization index types, and according to the specific requirements of using drones to photograph siltation in the sluice gate's upstream channel, set the corresponding index range interval to obtain the current route characterization index interval set. By clearly distinguishing between route quality characterization indicators and image quality influencing factors, the planning stage can adaptively optimize various parameters, avoiding data redundancy or missing data caused by blind settings. By setting reasonable ranges for route characterization indicators, the number of flights and images can be effectively controlled while ensuring mapping accuracy, thereby shortening fieldwork time and reducing storage and processing costs.

[0006] Preferably, step S2 includes the following steps: S21. Based on the set of siltation aerial photography route characterization indexes and the set of siltation aerial photography image quality influencing factors, collect data on siltation aerial photography route characterization indexes and siltation aerial photography image quality influencing factors corresponding to several historical drone aerial photography processes of siltation in the sluice gate front channel. This will result in a historical aerial photography route characterization index dataset and a historical image quality influencing factor dataset. S22. Based on the historical aerial photography route characterization index dataset and the historical image quality influencing factor dataset, construct a mapping model with image quality influencing factor data as input and aerial photography route characterization index data as output, and obtain the final aerial photography route characterization index mapping model. By establishing a mapping relationship from influencing factors to characterization indicators, the model can automatically recommend the optimal combination of route parameters based on the environmental conditions of the current task, avoiding the inefficient process of repeated adjustments based on experience in traditional planning.

[0007] Preferably, step S3 includes the following steps: S31. Based on the set of influencing factors for the quality of siltation aerial images, before the current drone aerial photography of the sluice gate's inlet channel begins to silt up, obtain the corresponding data of various influencing factors for the quality of siltation aerial images, and obtain the current aerial image quality influencing factor dataset. S32. Input the current aerial image quality influencing factor dataset into the final aerial flight route characterization index mapping model for mapping to obtain the current aerial flight route characterization index dataset. S33. In conjunction with the current flight route characterization index interval set, if there are characterization index data in the current aerial photography flight route characterization index dataset that are not located within the corresponding characterization index interval, repeatedly adjust the various quality influencing factor data in the current aerial photography image quality influencing factor dataset; otherwise, no adjustment is required. In each repetition, the adjusted current aerial image quality influencing factor dataset is input into the final aerial flight route characterization index mapping model for mapping, and the current adjusted characterization index dataset is obtained; until no characterization index data in the current adjusted characterization index dataset is not located within the corresponding characterization index interval, the current final aerial image quality influencing factor dataset is obtained. When the generated flight path parameters are detected to deviate from the preset safe range, an iterative adjustment program can be initiated. Through cyclical feedback, the configuration of influencing factors is continuously optimized to ensure that the final output flight path indicators are always within the optimal working range. This effectively avoids fluctuations in aerial data quality caused by improper setting of a single parameter, enhances the adaptability to complex and ever-changing environments, reduces the need for manual intervention, and improves operational efficiency and reliability of results.

[0008] Preferably, step S4 includes the following steps: S41. Collect multiple images of siltation in the diversion channel when there is a flood season or turbulent water flow in front of the sluice gate to obtain a historical flood season-turbulent siltation image dataset; perform manual image segmentation on the turbulent water flow part in each image data in the historical flood season-turbulent siltation image dataset and obtain the corresponding segmentation accuracy data to obtain a historical turbulent siltation image dataset after segmentation, a historical turbulent siltation image dataset before segmentation, and a historical segmentation accuracy dataset. Then, select weather conditions with good lighting, no wind or light wind, and calm water surface (to reduce ripple interference) to collect multiple images of siltation in the diversion channel, and obtain a historical calm siltation image dataset; in the historical calm siltation image dataset, accurately label the pixel-level areas of water bodies and silt bodies on each image data, and obtain a historical labeled calm siltation image dataset. S42. Based on the historical sedimentation image dataset after turbulent segmentation, the historical sedimentation image dataset before turbulent segmentation, and the historical segmentation accuracy dataset, construct a mapping model with the sedimentation image data before turbulent segmentation and the segmentation accuracy data as inputs and the sedimentation image data after segmentation as outputs, and obtain the final turbulent sedimentation image segmentation mapping model. Based on the historical labeled quiescent siltation image dataset, a mapping model is constructed with quiescent siltation image data as input and siltation segmentation result map as output, to obtain the final quiescent siltation segmentation result map mapping model; S43. Based on the current final aerial image quality influencing factor dataset, initialize the drone, aerial flight path, and corresponding shooting equipment for the aerial photography of the siltation in front of the sluice gate; after the settings are completed, perform aerial photography to obtain the current siltation aerial image dataset; perform stitching and correction operations on all image data in the current siltation aerial image dataset to obtain the current processed siltation aerial image data. The trained mapping model can effectively learn the complex mapping relationship from the original image to the segmentation result. At the same time, it integrates the auxiliary information of segmentation accuracy, which makes the model more adaptable and robust when facing images under different water conditions. It significantly improves the automation level of siltation image processing, reduces the need for manual intervention, reduces errors caused by subjective judgment differences, and ensures the consistency and accuracy of segmentation results.

[0009] Preferably, step S5 includes the following steps: S51. Set the initial turbulent image segmentation accuracy; if there is a turbulent part in the currently processed siltation aerial image data, input the currently processed siltation aerial image data and the initial turbulent image segmentation accuracy into the final turbulent siltation image segmentation mapping model to perform image segmentation operation, and obtain the currently segmented turbulent image data and the currently segmented calm image data; otherwise, no segmentation operation is required, and the currently processed siltation aerial image data is used as the currently segmented calm image data. S52. Preprocess the current segmented turbulent image data to obtain the current processed turbulent image data; input the current processed turbulent image data and the current segmented calm image data into the final calm sedimentation segmentation result map mapping model for mapping to obtain the current first sedimentation segmentation result map and the current second sedimentation segmentation result map; S53. Calculate the siltation area and siltation volume based on the current first siltation segmentation result map and the current second siltation segmentation result map to obtain the current aerial siltation area data and the current aerial siltation volume data; By segmenting turbulent and calm areas separately, the system achieves accurate identification of the characteristics of different types of water bodies, avoiding the misjudgment problem of a single model in complex scenarios.

[0010] Preferably, step S6 includes the following steps: S61. Conduct multiple drone aerial photography experiments to identify siltation in the sluice gate's inlet channel. Each experiment repeats steps S31, S32, S33, S43, S51, S52, and S53, and the initial turbulent image segmentation accuracy is different in each experiment. When there is a turbulent water flow in the processed siltation aerial image data, obtain the initial turbulent image segmentation accuracy data set in each experiment and the total time consumed by repeating steps S51, S52, and S53 to obtain the turbulent image segmentation accuracy experimental dataset and the drone aerial siltation calculation time dataset. S62. Based on the experimental dataset of turbulent image segmentation accuracy and the dataset of UAV aerial photography sedimentation calculation time, construct a mapping model with turbulent image segmentation accuracy data as input and UAV aerial photography sedimentation calculation time data as output, and obtain the final sedimentation calculation time mapping model. S63. Set the turbulent image segmentation accuracy range according to the actual turbulent water flow segmentation requirements; randomly initialize multiple sets of turbulent image segmentation accuracy data according to the turbulent image segmentation accuracy range, and input them into the final sedimentation calculation time mapping model for mapping to obtain the current aerial sedimentation calculation time dataset. Select the turbulent image segmentation accuracy data corresponding to the aerial dust calculation time data with the smallest aerial dust calculation time data in the current aerial dust calculation time dataset as the final turbulent image segmentation accuracy data. By randomly initializing multiple sets of precision parameters and performing mapping predictions, the system can quickly select the optimal segmentation precision configuration, significantly improving overall operational efficiency and avoiding the inefficient mode of repeated manual trials.

[0011] An automatic siltation identification system for sluice gate inlet channels integrating UAV aerial photography includes a siltation identification quality and influencing factor type setting module, an aerial photography route characterization index mapping model construction module, a current image quality influencing factor adjustment module, a siltation image segmentation model construction and current aerial image acquisition and processing module, a current siltation area and volume calculation module, and a turbulent image segmentation accuracy screening module.

[0012] The present invention has the following beneficial effects: 1. This invention automatically optimizes flight parameters through a dynamic feedback adjustment mechanism, avoiding the subjectivity and inefficiency of traditional manual adjustments, and significantly improving operational accuracy and stability. Simultaneously, it constructs dedicated segmentation models for different hydrological conditions (turbulent and calm), and effectively addresses complex environmental challenges through a dual-model fusion processing strategy, greatly improving the accuracy and robustness of sediment identification. Furthermore, it integrates an intelligent time-consuming optimization mechanism, automatically selecting the optimal segmentation accuracy through experimental data-driven methods, minimizing processing time while ensuring monitoring quality. This provides reliable technical support for the efficient operation and maintenance management of hydraulic structures, truly realizing an intelligent and standardized operation mode from data acquisition to output.

[0013] 2. In this invention, after segmenting the turbulent region, a specialized preprocessing procedure is used to optimize image quality, ensuring the stability and accuracy of the input data for subsequent segmentation models. This hierarchical processing strategy effectively improves the model's adaptability to different hydrological conditions. By segmenting the turbulent and calm regions independently, accurate identification of different types of water body characteristics is achieved, avoiding the misjudgment problem of a single model in complex scenarios.

[0014] 3. In this invention, by constructing an input-output mapping model, a mathematical relationship is established between segmentation accuracy and computation time, so that subsequent accuracy selection is no longer an empirical guess, but a data-driven scientific decision-making process; by randomly initializing multiple sets of accuracy parameters and performing mapping prediction, the system can quickly select the optimal segmentation accuracy configuration, significantly improving the overall operation efficiency and avoiding the inefficient mode of repeated manual trials.

[0015] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the 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 invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the automatic identification method for siltation in the inlet channel of a sluice gate that integrates drone aerial photography according to the present invention. Figure 2 A schematic diagram of the process for constructing the final flight path characterization index mapping model for this invention; Figure 3 This is a schematic diagram illustrating the process of adjusting current image quality influencing factors according to the present invention; Figure 4This is a schematic diagram illustrating the process of constructing a congestion image segmentation model to acquire and process the current aerial image. Figure 5 This is a schematic diagram of the process for calculating the current siltation area and volume according to the present invention; Figure 6 This is a line graph showing the trend of accumulation calculation time under different segmentation accuracies according to the present invention; Figure 7 This is a schematic diagram of the process for screening the segmentation accuracy of turbulent images according to the present invention; Figure 8 This is a schematic diagram of a module of an automatic identification system for siltation in the inlet channel of a sluice gate that integrates drone aerial photography according to the present invention. Detailed Implementation

[0018] 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.

[0019] 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.

[0020] Example 1 Please see Figure 1 This embodiment is a method for automatic identification of siltation in the inlet channel of a sluice gate by integrating drone aerial photography, including the following steps: S1. Construct flight path quality characterization indicators and image quality influencing factors for identifying siltation in the inlet channel of a sluice gate using UAV aerial photography, and set the range of each indicator. Please see Figure 2 S1 includes the following steps: S11. Define several types of indicators to characterize the quality of UAV flight paths for aerial photography of siltation in the canal in front of the sluice gate, resulting in a set of siltation aerial photography flight path characterization indicators. These indicators include canal coverage, directional overlap, and lateral overlap. Then, define several types of factors that influence the image quality of aerial photography of siltation in the canal in front of the sluice gate, resulting in a set of factors influencing the image quality of siltation aerial photography. These factors include flight path coverage, UAV flight speed, camera shutter speed, camera focal length, camera image size, terrain complexity, flight path spacing (smaller spacing results in higher overlap), and flight platform stability (UAV attitude stability affects actual overlap accuracy; a stable platform is required to ensure accurate parameters). S12. Based on the set of siltation aerial photography route characterization index types, and according to the specific requirements of using drones to photograph siltation in the sluice gate's upstream channel, set the corresponding index range interval to obtain the current route characterization index interval set. The flight path should cover the entire monitored diversion canal area, including slopes prone to siltation, and ensure directional and lateral overlap (generally recommended to be above 80%) to guarantee the quality of subsequent image stitching. A comprehensive evaluation system was constructed by pre-setting key indicators and factors affecting the quality of aerial photography of siltation in the diversion canal in front of the sluice gate, which helps improve the scientific rigor and accuracy of UAV aerial surveying operations. Specifically, by clearly distinguishing between flight path quality indicators (such as diversion canal coverage, directional and lateral overlap) and image quality influencing factors (such as flight speed, camera parameters, terrain conditions, etc.), the planning stage can adaptively optimize various parameters, avoiding data redundancy or missing data due to blind settings. By setting reasonable ranges for flight path indicators, the number of flights and images can be effectively controlled while ensuring surveying accuracy, thereby shortening fieldwork time and reducing storage and processing costs. Furthermore, by incorporating implicit factors such as UAV platform stability and flight path intervals, the system's adaptability to complex field environments is enhanced, reducing problems such as insufficient overlap caused by equipment vibration or attitude deviations, and improving the reliability and consistency of the results. S2. Based on S1, collect multiple sets of historical flight route quality characterization indicators and image quality influencing factor data, and construct the final aerial flight route characterization indicator mapping model. S2 includes the following steps: S21. Based on the set of siltation aerial photography route characterization indexes and the set of siltation aerial photography image quality influencing factors, collect data on siltation aerial photography route characterization indexes and siltation aerial photography image quality influencing factors corresponding to several historical drone aerial photography processes of siltation in the sluice gate front channel. This will result in a historical aerial photography route characterization index dataset and a historical image quality influencing factor dataset. Among them, the data can be exported through the log recording function of the drone flight control software, and combined with the quality inspection report of post-image processing software (such as Pix4D and ContextCapture) to obtain aerial flight route characterization index data such as coverage and overlap; preset parameters can be extracted from flight mission planning software (such as DJI Terra and Altizure), and hardware configuration information such as camera model and focal length (image quality influencing factor data) can be obtained through the corresponding equipment management system. For example, some data from the historical aerial flight route characterization index dataset and the historical image quality influencing factors dataset are shown in Table 1 below: Table 1: Example of Historical Aerial Photography Data

[0021] S22. Based on the historical aerial photography route characterization index dataset and the historical image quality influencing factor dataset, construct a mapping model with image quality influencing factor data as input and aerial photography route characterization index data as output, and obtain the final aerial photography route characterization index mapping model. The final aerial flight route representation index mapping model adopts a fully connected feedforward neural network, which is a three-layer fully connected neural network (including an input layer, two hidden layers, and an output layer). The number of neurons in the input layer is equal to the number of image quality influencing factor types set. The first hidden layer has 128 neurons, the second hidden layer has 64 neurons, and the number of neurons in the output layer is equal to the dimension of the flight route representation index to be predicted (e.g., coverage, heading overlap, and lateral overlap are set to 3). ReLU (Linear Correction Unit) is used as the activation function between each layer to enhance the model's nonlinear fitting ability. The output layer can select a specific activation function according to the index characteristics (e.g., Sigmoid limits the output to the 0-1 range to represent a percentage). Mean squared error (MSE) is used as the loss function during training. Fully connected feedforward neural networks optimize network weights through backpropagation algorithms, enabling precise mapping from influencing factors such as flight speed and camera focal length to indicators such as route coverage and overlap. By collecting data on flight path characteristics (such as coverage and overlap) and image quality influencing factors (such as flight speed, camera parameters, and terrain conditions) from historical aerial photography missions, a statistically significant multi-dimensional dataset was formed, providing a solid foundation for model training. By establishing a mapping relationship from influencing factors to characteristics, the model can automatically recommend the optimal combination of flight path parameters (such as flight altitude, speed, and overlap settings) based on the current mission's environmental conditions (such as terrain complexity and equipment performance), avoiding the inefficient process of repeated adjustments based on experience in traditional planning. This data-driven approach not only enhances the scientific nature and adaptability of parameter settings but also maintains stable imaging quality under different geographical environments and equipment conditions, effectively reducing the risk of reshoots due to inappropriate parameters. It achieves optimal resource allocation while ensuring surveying accuracy, providing reliable technical support for the intelligent operation and maintenance monitoring of hydraulic structures. S3. Input the current image quality influencing factor data into the mapping model in S2 for mapping; if the mapping result does not meet the range of each indicator in S1, adjust the current image quality influencing factor data until it meets the range; otherwise, no adjustment is needed. Please see Figure 3 S3 includes the following steps: S31. Based on the set of influencing factors for the quality of siltation aerial images, before the current drone aerial photography of the sluice gate's inlet channel begins to silt up, obtain the corresponding data of various influencing factors for the quality of siltation aerial images, and obtain the current aerial image quality influencing factor dataset. S32. Input the current aerial image quality influencing factor dataset into the final aerial flight route characterization index mapping model for mapping to obtain the current aerial flight route characterization index dataset. S33. In conjunction with the current flight route characterization index interval set, if there are characterization index data in the current aerial photography flight route characterization index dataset that are not located within the corresponding characterization index interval, repeatedly adjust the various quality influencing factor data in the current aerial photography image quality influencing factor dataset; otherwise, no adjustment is required. In each repetition, the adjusted current aerial image quality influencing factor dataset is input into the final aerial flight route characterization index mapping model for mapping, and the current adjusted characterization index dataset is obtained; until no characterization index data in the current adjusted characterization index dataset is not located within the corresponding characterization index interval, the current final aerial image quality influencing factor dataset is obtained. By constructing an adaptive feedback adjustment mechanism, dynamic optimization and precise control of UAV aerial photography parameters were achieved, significantly improving the intelligence level and data quality stability of siltation monitoring operations in the sluice gate's upstream channel. Specifically, real-time environmental parameters are automatically collected and intelligently mapped and reasoned before operations. When the generated flight path parameters are detected to deviate from the preset safety range, an iterative adjustment program can be initiated. Through cyclical feedback, the configuration of influencing factors is continuously optimized to ensure that the final output flight path indicators are always within the optimal working range. This effectively avoids fluctuations in aerial photography data quality caused by improper settings of a single parameter, enhances adaptability to complex and changing environments, reduces the need for manual intervention, and improves operational efficiency and reliability. At the same time, by establishing dynamic correlations between parameters, flight path planning is no longer a static preset process, but an intelligent decision-making process with self-correction capabilities. This provides a solid technical foundation for achieving long-term automated monitoring of hydraulic structures and realizes a shift from "passive response" to "active optimization." S4. Collect multiple sets of siltation image data and label them to train the turbulent water flow segmentation model and the calm water area segmentation model respectively; and carry out aerial photography of siltation based on the factor data adjusted in S3. After aerial photography, the collected images are stitched and corrected. Please see Figure 4 S4 includes the following steps: S41. Collect multiple images of siltation in the diversion channel when there is a flood season or turbulent water flow in front of the sluice gate to obtain a historical flood season-turbulent siltation image dataset; perform manual image segmentation on the turbulent water flow part in each image data in the historical flood season-turbulent siltation image dataset and obtain the corresponding segmentation accuracy data (i.e., the ratio of the area of ​​the turbulent water flow part to the area of ​​the frame), to obtain a historical turbulent siltation image dataset after segmentation, a historical turbulent siltation image dataset before segmentation, and a historical segmentation accuracy dataset. Then, select weather conditions with good lighting, no wind or light wind, and calm water surface (to reduce ripple interference) to collect multiple images of siltation in the irrigation canal, and obtain a historical calm siltation image dataset; invite water conservancy experts or experienced practitioners to use image annotation tools to accurately annotate the pixel-level areas of water bodies and siltation bodies (or "non-water bodies", i.e., the beach exposed above the water surface) on each image in the historical calm siltation image dataset (the annotation should take into account various situations under different seasons, different lighting, different turbidity, and different siltation morphologies to increase the diversity of the dataset and the generalization ability of the model), and obtain a historical annotated calm siltation image dataset; S42. Based on the historical sedimentation image dataset after turbulent segmentation, the historical sedimentation image dataset before turbulent segmentation, and the historical segmentation accuracy dataset, construct a mapping model with the sedimentation image data before turbulent segmentation and the segmentation accuracy data as inputs and the sedimentation image data after segmentation as outputs, and obtain the final turbulent sedimentation image segmentation mapping model. Based on the historically labeled calm sedimentation image dataset, a mapping model is constructed that takes calm sedimentation image data as input and outputs a sedimentation segmentation result map (clearly indicating whether each pixel belongs to "water body" or "sediment body"), thus obtaining the final calm sedimentation segmentation result map mapping model. The final turbulent sedimentation image segmentation and mapping model adopts a generative adversarial network model, and an example structure is as follows: Input layer: Image size is uniformly normalized to [batch_size, channels=3, height=256px, width=256px] (color RGB image); segmentation accuracy data is expanded into a tensor with shape=[batch_size,1] and injected into the feature representation in the intermediate stage of the network; The encoder consists of a spatial compression path with progressive dimensionality reduction, formed by four convolutional blocks, as shown in Table 2 below. Table 2. Schematic diagram of encoder structure for image segmentation mapping model.

[0022] Each of the four convolutional blocks is followed by a max pooling layer (MaxPooling size=2×2, s=2) to further reduce dimensionality while preserving key semantic features; Intermediate fusion layer: The pre-calculated scaling factor (a floating-point number in the range [0,1]) is converted into a vector (e.g., a tensor of shape=[-1,512,1,1]) with a length equal to the number of E4 output channels. Then, it is added to or concatenated with the output of the last encoder to form a new latent representation. Decoder: Also contains four sets of deconvolution units to progressively restore the image resolution, as shown in Table 3 below: Table 3. Schematic diagram of the decoder structure for the image segmentation mapping model.

[0023] The first three layers are equipped with Dropout regularization to prevent overfitting and improve generalization ability; the last layer determines the output image type (grayscale or color) and controls the pixel intensity range through Tanh / Sigmoid; in addition, feature maps from E3-E1 are copied from corresponding positions in D1-D3 and concatenated with them (jump connection mechanism) to help restore finer edge contours. Discriminant subnet: Used to evaluate the truth value of each N×N patch (referring to dividing a complete image into multiple square blocks with sides of N pixels), rather than the global truth value; as shown in Table 4 below: Table 4. Schematic diagram of the discriminant subnet structure of the image segmentation mapping model.

[0024] The Least Squares Error (LSEloss) loss function is chosen to help alleviate the gradient vanishing phenomenon and accelerate convergence. Here, batch_size, channels, height, width, Convolutional, Convolutional+BN, kernel, s, α, Transposed Conv.+BN, Final Transposed Conv., p, ConvLayer, BatchNorm, and out_channels represent the batch size, the number of channels in the convolutional kernel, the vertical pixel size of the image, the horizontal pixel size of the image, the convolutional layer, the convolutional layer + batch normalization, the convolutional kernel, the stride, the slope parameter of the negative interval in LeakyReLU, the transposed convolution + batch normalization, the final transposed convolution, the Dropout probability, the convolutional layer (same as "Convolutional"), the batch normalization (same as "BN"), and the number of channels in the feature map output of the convolutional layer. The final calm congestion segmentation result mapping model uses U-Net, which includes a shrinking path (encoder) and an expanding path (decoder). The shrinking path consists of four layers, each containing two convolutional layers (using 3×3 convolutional kernels), followed by batch normalization layers and ReLU activation functions (linear correction units, used to introduce non-linear characteristics). Each layer is finally downsampled using max pooling layers (2×2 pooling windows with a stride of 2), with the number of channels being 64, 128, 256, and 512 respectively. The expanding path also consists of four layers, using upsampling convolutions... The network performs upsampling on layers (transposed convolution, 2×2 kernel, stride of 2), with channel numbers of 1024, 512, 256, and 128 respectively. Each layer also contains two convolutional layers (3×3 kernel) and a ReLU activation function. After upsampling, the feature maps of the corresponding layers of the shrinking path are concatenated through skip connections. The network output layer is a 1×1 convolutional layer, which maps the number of feature map channels to the number of categories (2 categories in this scenario: water and silt). Finally, the Softmax activation function is used to output the probability distribution of each pixel belonging to each category, generating a binarized silt segmentation result map. S43. Based on the current final aerial image quality influencing factor dataset, initialize the drone, aerial flight path, and corresponding shooting equipment for the aerial photography of the siltation in front of the sluice gate; after the settings are completed, perform aerial photography to obtain the current siltation aerial image dataset; perform stitching and correction operations on all image data in the current siltation aerial image dataset to obtain the current processed siltation aerial image data. Specifically, the drone automatically flies along a predetermined route to collect high-resolution digital images of the sluice gate canal area. The flight altitude should be determined based on the required ground resolution to ensure that the texture features of the water body and the beach (silt) can be clearly distinguished to obtain high-quality images. For the stitching and correction operations, professional photogrammetry software (such as Pix4Dmapper, ContextCapture, etc.) is used to automatically stitch together hundreds or even thousands of individual images taken by the drone and perform aerial triangulation to generate a high-resolution digital orthophoto map of the entire sluice gate canal area. The generated DOM (digital orthophoto map) can accurately represent the planar position of ground features and eliminate distortions caused by camera tilt and terrain undulations. In addition, during the aerial photography, an appropriate number of ground control points are set up in the survey area, or a differential GPS system is used on the drone to ensure that the generated DOM has high planar positioning accuracy. By constructing an image segmentation mapping model driven by historical data, automated and precise monitoring of siltation in sluice gate canals was achieved. Specifically, historical siltation image data under flood season or turbulent water flow conditions was collected, and combined with manually labeled segmentation accuracy information, a training database containing input images and corresponding segmentation results was established. This provided high-quality supervision signals for subsequent model training. Based on this, the mapping model trained using this database can effectively learn the complex mapping relationship from the original image to the segmentation result. At the same time, by incorporating the auxiliary information of segmentation accuracy, the model has stronger adaptability and robustness when facing images under different water conditions. This significantly improves the automation level of siltation image processing, reduces the need for manual intervention, reduces errors caused by subjective judgment differences, and ensures the consistency and accuracy of segmentation results. Furthermore, it enables real-time monitoring and rapid response to siltation in sluice gate canals, effectively improving the safety and efficiency of sluice gate operation. S5. Based on the initial accuracy parameters, the turbulent water area is segmented, and the turbulent and calm areas are extracted. The turbulent and calm segmentation model in S4 is used to process different areas to obtain double segmentation results and calculate the siltation area and volume. Please see Figure 5 S5 includes the following steps: S51. Set the initial turbulent image segmentation accuracy (which can be adaptively set according to actual image processing requirements); if there is a turbulent water flow in the currently processed siltation aerial image data, input the currently processed siltation aerial image data and the initial turbulent image segmentation accuracy into the final turbulent siltation image segmentation mapping model for image segmentation operation to obtain the currently segmented turbulent image data and the currently segmented calm image data; otherwise, no segmentation operation is required, and the currently processed siltation aerial image data is used as the currently segmented calm image data. S52. Preprocess the current segmented turbulent image data to obtain the current processed turbulent image data; input the current processed turbulent image data and the current segmented calm image data into the final calm sedimentation segmentation result map mapping model for mapping to obtain the current first sedimentation segmentation result map and the current second sedimentation segmentation result map; The preprocessing includes the following steps: S521. Adaptive histogram equalization is performed on the current segmented turbulent image data to improve the grayscale difference between the siltation area and the background in the turbid water. Then, nonlocal mean filtering is used to eliminate high-frequency noise caused by water surface ripples, reflections, and suspended matter in the water, while preserving the edge structure. For areas with strong reflections, a polarization correction model (if the image contains polarization information) or color normalization based on illumination invariance is combined to reduce the interference of water surface specular reflection on the identification of siltation boundaries, resulting in a preprocessed grayscale image. S522. The preprocessed grayscale image is converted into a binary image, and the Otsu adaptive thresholding method is used to initially separate the water body and the siltation area. Then, the watershed algorithm is applied for fine segmentation. This algorithm regards the image grayscale value as the terrain elevation, simulates the "water accumulation" process, sets "drainage holes" at the local minimum value, and floods layer by layer. When the water levels of different areas are about to converge, the "watershed" boundary is automatically constructed, thereby realizing the single-pixel level contour extraction of complex siltation morphology, effectively distinguishing connected but irregularly shaped siltation patches, and obtaining the watershed segmentation result. S523. Perform morphological closing operation on the watershed segmentation results to fill the boundary breaks and internal holes caused by water flow disturbance; then, through contour filtering, remove noise areas that are too small (e.g., <100 pixels) or have abnormal aspect ratios, and retain continuous areas that conform to the spatial distribution characteristics of siltation to obtain the current processed turbulent image data. S53. Calculate the siltation area and siltation volume based on the current first siltation segmentation result map and the current second siltation segmentation result map to obtain the current aerial siltation area data and the current aerial siltation volume data; Specifically, as follows: (1) Calculation of siltation range and area 1) Baseline data: Select an orthophoto of the UAV at a certain historical moment (usually a state without siltation or as a baseline) and the water body boundary identified at that time (as the baseline boundary), or the original design cross-section of the diversion channel (which can be used as a baseline reference after being digitized). 2) Change detection: Spatial overlay analysis is performed between the currently identified water body boundary vector (which can be obtained based on the current first sedimentation segmentation result map and the current second sedimentation segmentation result map) and the selected historical baseline boundary vector; by calculating the difference area between the two boundaries, newly added sedimentation areas (areas where the current boundary shrinks into the water body relative to the historical baseline) or scour areas can be automatically identified. 3) Area statistics: Using GIS (Geographic Information System) software or custom scripts, the projected area of ​​these disparate areas can be automatically calculated, that is, the total area of ​​the siltation range, to obtain the current aerial siltation area data; (2) Estimate of siltation volume 1) Introducing Elevation Information: Calculation of siltation volume requires elevation information. Several approaches can be used: Option 1 (based on historical topographic data): If an early high-precision digital elevation model (DEM) of the sluice gate diversion channel area is available (possibly from earlier airborne lidar measurements or earlier UAV photogrammetry), the currently identified siltation range (which can be obtained based on the current first siltation segmentation result map and the current second siltation segmentation result map) can be superimposed onto this historical DEM; assuming that the elevation of the siltation surface is basically consistent with the elevation of the corresponding area on the historical DEM (or corrected based on a small number of verification points), the siltation volume can be estimated by calculating the integral of the elevation of the historical DEM within the siltation range with the channel bottom design elevation (or another reference elevation), thus obtaining the current aerial siltation volume data; Option 2 (Estimation based on average silt thickness): If an accurate historical DEM is unavailable, but the design cross-sectional shape of the diversion channel (e.g., trapezoidal cross-section) and design water depth are known, or the water flow area without siltation can be calculated based on historical baseline images, the average thickness of the silt mass can be estimated by combining the current boundary line position (which can be obtained based on the current first and second silt segmentation results) with certain cross-sectional shape assumptions (e.g., the silt surface is planar or extends according to the design slope). Then, the average thickness is multiplied by the silt area to obtain the current aerial silt volume data. The above-mentioned calculation process of siltation area and volume can be automated through programming; simply input the vector file of the current boundary and the benchmark data (historical boundary vector, historical DEM or design parameters), and the program can automatically complete spatial analysis, thickness estimation (if needed) and volume calculation; Scheme 1 is suitable for situations with high-precision historical topographic data, while Scheme 2, although slightly less accurate, is easy to implement and still has high value for trend judgment and periodic comparison. By constructing a multi-stage image processing and segmentation fusion mechanism, accurate analysis and quantitative evaluation of sedimentation monitoring data under complex hydrological conditions were achieved. Specifically, through adaptive accuracy setting and intelligent judgment mechanism, the system can automatically identify turbulent water flow areas in the image and perform targeted segmentation processing. This intelligent judgment logic avoids the accuracy loss caused by applying a uniform processing standard to all images in traditional methods. After segmenting the turbulent areas, a dedicated preprocessing workflow optimizes image quality, ensuring the stability and accuracy of the input data for subsequent segmentation models. This hierarchical processing strategy effectively improves the model's adaptability to different hydrological conditions. By performing independent segmentation processing on turbulent and calm areas, accurate identification of different types of water body characteristics was achieved, avoiding the misjudgment problem of a single model in complex scenarios. The final area and volume calculations are based on the comprehensive analysis of two independent segmentation results. This dual verification mechanism significantly improves the reliability and accuracy of sedimentation calculation, providing more accurate and reliable data support for the refined management and maintenance decisions of hydraulic structures. S6. Conduct multiple rounds of experiments, repeating the aerial photography operations for siltation in S3 and S4 as well as S5 in each experiment, collecting time data under different segmentation accuracies, constructing the final siltation calculation time mapping model; and selecting the turbulent image segmentation accuracy with the minimum aerial siltation calculation time based on the model. Please see Figure 6 , Figure 7 S6 includes the following steps: S61. Conduct multiple drone aerial photography experiments to identify siltation in the sluice gate's inlet channel. Each experiment repeats steps S31, S32, S33, S43, S51, S52, and S53, and the initial turbulent image segmentation accuracy is different in each experiment. When there is a turbulent water flow in the processed siltation aerial image data, obtain the initial turbulent image segmentation accuracy data set in each experiment and the total time consumed by repeating steps S51, S52, and S53 to obtain the turbulent image segmentation accuracy experimental dataset and the drone aerial siltation calculation time dataset. S62. Based on the experimental dataset of turbulent image segmentation accuracy and the dataset of UAV aerial photography sedimentation calculation time, construct a mapping model with turbulent image segmentation accuracy data as input and UAV aerial photography sedimentation calculation time data as output, and obtain the final sedimentation calculation time mapping model. S62 includes the following steps: S621. Construct an initial congestion calculation time mapping model and set a first training data ratio (e.g., 8:2 or 7:3, which can be adjusted adaptively according to the actual training situation); divide the turbulent image segmentation accuracy experimental dataset and the UAV aerial photography congestion calculation time dataset according to the first training data ratio to obtain the first training dataset and the first test dataset. S622. Set a first training error threshold (10%~15%, which can be adjusted adaptively according to the actual training situation); input the first training dataset into the initial congestion calculation time-consuming mapping model for training; during the training process, if the training error is less than the first training error threshold, stop training and obtain the trained congestion calculation time-consuming mapping model; otherwise, continue training until the training error is less than the first training error threshold. S623. Set a first test accuracy threshold (90%~95%, which can be adjusted adaptively according to the actual test situation); input the first test dataset into the trained congestion computation time-consuming mapping model for testing; after the test is completed, obtain the first test accuracy data; if the first test accuracy data is greater than or equal to the first test accuracy threshold, use the trained congestion computation time-consuming mapping model as the final congestion computation time-consuming mapping model; otherwise, return to S622 to continue training the trained congestion computation time-consuming mapping model and repeat S623 until the first test accuracy data is greater than or equal to the first test accuracy threshold. The initial siltation calculation time mapping model adopts a multinomial regression model, which includes an input layer, a hidden layer, and an output layer; The input layer has one neuron, corresponding to a single input feature "turbulent image segmentation accuracy". The hidden layer has 64 neurons and uses the ReLU activation function (linear correction unit) to introduce nonlinear transformation capabilities, which can effectively fit the complex nonlinear relationship between segmentation accuracy and computation time. The output layer contains one neuron to output the predicted computation time of drone aerial image congestion. Before network training, the input segmentation accuracy data is expanded using a third-order polynomial feature expansion (that is, the original feature x is expanded into a feature vector of [x, x², x³]) to enhance the model's ability to fit nonlinear trends. The model uses mean squared error (MSE) as the loss function and optimizes the parameters using the Adam optimizer with a learning rate of 0.001. This allows for rapid convergence and accurate prediction of computation time under different segmentation accuracy settings, providing a reliable performance evaluation basis for subsequent accuracy optimization decisions. S63. Set the turbulent image segmentation accuracy range according to the actual turbulent water flow segmentation requirements; randomly initialize multiple sets of turbulent image segmentation accuracy data according to the turbulent image segmentation accuracy range (the number of data sets can be adaptively set according to the actual situation), and input them into the final sedimentation calculation time mapping model for mapping to obtain the current aerial sedimentation calculation time dataset. Select the turbulent image segmentation accuracy data corresponding to the aerial dust calculation time data with the smallest aerial dust calculation time data in the current aerial dust calculation time dataset as the final turbulent image segmentation accuracy data. By repeatedly executing the complete aerial photography processing workflow and recording the computation time under different segmentation accuracy settings, a precise performance-accuracy trade-off database was established. This optimization method based on measured data avoids the uncertainties of theoretical derivation and ensures reliability in practical applications. By constructing an input-output mapping model, a mathematical relationship is established between segmentation accuracy and computation time, making subsequent accuracy selection no longer an empirical guess but a data-driven scientific decision-making process. By randomly initializing multiple sets of accuracy parameters and performing mapping predictions, the system can quickly select the optimal segmentation accuracy configuration, significantly improving overall operational efficiency and avoiding the inefficient mode of repeated manual trials. Thus, it not only achieves optimal allocation of computing resources but also provides quantifiable and predictable performance assurance for sedimentation monitoring under complex hydrological conditions, and provides an efficient and reliable solution for long-term automated monitoring of hydraulic structures.

[0025] Example 2 Please see Figure 8 This embodiment discloses an automatic siltation identification system for the sluice gate inlet channel that integrates UAV aerial photography. The system can implement the method of the above embodiment, including a siltation identification quality and influencing factor type setting module, an aerial photography route characterization index mapping model construction module, a current image quality influencing factor adjustment module, a siltation image segmentation model construction and current aerial photography image acquisition and processing module, a current siltation area and volume calculation module, and a turbulent image segmentation accuracy screening module. The siltation identification quality and influencing factor type setting module constructs flight path quality characterization indicators and image quality influencing factors for siltation identification in the sluice gate's inlet channel by UAV aerial photography, and sets the range range of each indicator; The aerial flight route characterization index mapping model construction module collects multiple sets of historical flight route quality characterization indexes and image quality influencing factor data based on the siltation identification quality and influencing factor type setting module, and constructs the final aerial flight route characterization index mapping model based on this. The current image quality influencing factor adjustment module collects current image quality influencing factor data and inputs it into the mapping model in the aerial flight route characterization index mapping model construction module for mapping. If the mapping result does not meet the range of each index in the siltation identification quality and influencing factor type setting module, the current image quality influencing factor data is adjusted until the range is met; otherwise, no adjustment is required. The siltation image segmentation model construction and current aerial image acquisition and processing module collects multiple sets of siltation image data under turbulent and calm conditions during the flood season and annotates them to train the turbulent water flow segmentation model and the calm water area segmentation model respectively; and performs siltation aerial photography based on the factor data adjusted in S3, and stitches and corrects the collected images after aerial photography. The current siltation area and volume calculation module segments the turbulent water area based on the initial accuracy parameters, extracts turbulent and calm areas, and uses the siltation image segmentation model construction and the turbulent and calm segmentation model in the current aerial image acquisition and processing module to process different areas, obtain double segmentation results, and calculate the siltation area and volume. The turbulent image segmentation accuracy screening module conducts multiple rounds of experiments. In each experiment, it repeats the current image quality influencing factor adjustment module, the siltation image segmentation model construction and the siltation aerial photography operation in the current aerial image acquisition and processing module, as well as the current siltation area and volume calculation module. It collects time consumption data under different segmentation accuracies, constructs the final siltation calculation time mapping model, and selects the turbulent image segmentation accuracy with the minimum aerial siltation calculation time based on the model.

[0026] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0027] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present invention.

Claims

1. A method for automatic identification of siltation in the inlet channel of a sluice gate, integrating drone aerial photography, characterized in that, Includes the following steps: S1. Construct flight path quality characterization indicators and image quality influencing factors for identifying siltation in the inlet channel of a sluice gate using UAV aerial photography, and set the range of each indicator. S2. Based on the data collected in S1, multiple sets of flight route quality characterization indicators and image quality influencing factors in history are collected, and a final aerial flight route characterization indicator mapping model is constructed. S3. Input the current image quality influencing factor data into the mapping model in S2 for mapping; if the mapping result does not meet the range of each indicator in S1, adjust the current image quality influencing factor data until it meets the range; Otherwise, no adjustment is needed; S4. Collect multiple sets of siltation image data and label them to train the turbulent water flow segmentation model and the calm water area segmentation model respectively; and carry out aerial photography of siltation based on the factor data adjusted in S3. After aerial photography, the collected images are stitched and corrected. S5. Based on the initial accuracy parameters, the turbulent water area is segmented, and the turbulent and calm areas are extracted. The turbulent and calm segmentation model in S4 is used to process different areas to obtain double segmentation results and calculate the siltation area and volume. S6. Conduct multiple rounds of experiments, repeating the aerial siltation operations in S3 and S4 as well as S5 in each experiment, collecting time consumption data under different segmentation accuracies, constructing the final siltation calculation time mapping model, and selecting the turbulent image segmentation accuracy with the minimum aerial siltation calculation time based on the model.

2. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating UAV aerial photography as described in claim 1, characterized in that, S1 includes the following steps: S11. Define several types of indicators to characterize the quality of UAV flight paths for aerial photography of siltation in the diversion channel in front of the sluice gate, including diversion channel coverage, forward overlap, and lateral overlap; then define several types of factors that affect the image quality of aerial photography of siltation in the diversion channel in front of the sluice gate, including flight path coverage, UAV flight speed, camera shutter speed, camera focal length, camera image size, terrain complexity, flight path spacing, and flight platform stability. S12. Set the index range interval corresponding to each type of siltation aerial photography route characterization index.

3. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating UAV aerial photography as described in claim 1, characterized in that, S2 includes the following steps: S21. Collect data on the characteristics of siltation aerial photography routes and the factors affecting the quality of siltation aerial images during several historical drone aerial photography sessions of the siltation process in front of the sluice gate. Construct a mapping model with image quality influencing factor data as input and aerial photography route characteristics data as output, and obtain the final aerial photography route characteristics mapping model.

4. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating drone aerial photography as described in any one of claims 3, characterized in that, S3 includes the following steps: S31. Before the current drone aerial photography of the sluice gate's inlet channel begins to collect data on various factors affecting the quality of aerial images of siltation, and input these data into the final aerial photography route characterization index mapping model to obtain the current aerial photography route characterization index dataset.

5. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating drone aerial photography according to any one of claims 4, characterized in that, S3 further includes: S33. Based on S12, if there are any indicator data in the current aerial flight route indicator dataset that are not within the corresponding indicator range, the various quality influencing factor data in the current aerial image quality influencing factor dataset shall be repeatedly adjusted; otherwise, no adjustment is required. In each repetition, the adjusted dataset of current aerial image quality influencing factors is input again into the final aerial flight route characterization index mapping model for mapping, until no characterization index data is found that is not within the corresponding characterization index range in the mapping result, thus obtaining the current final aerial image quality influencing factor dataset.

6. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating drone aerial photography according to any one of claims 5, characterized in that, S4 includes the following steps: S41. Collect multiple images of siltation in the diversion channel when there is flood season or turbulent water flow in front of the sluice gate, perform manual image segmentation and obtain the corresponding segmentation accuracy data to obtain a historical siltation image dataset after turbulent segmentation, a historical siltation image dataset before turbulent segmentation and a historical segmentation accuracy dataset. Multiple images of siltation in the irrigation canal under calm water conditions were collected to obtain a historical calm siltation image dataset; then, pixel-level regions of water bodies and siltation bodies were marked on each image to obtain a historical annotated calm siltation image dataset. S42. Based on the data collected in S41, construct a mapping model with the input of sedimentation image data before turbulent segmentation and segmentation accuracy data, and the output of sedimentation image data after segmentation, to obtain the final turbulent sedimentation image segmentation mapping model. Then, a mapping model is constructed with input calm siltation image data and output siltation segmentation result map to obtain the final calm siltation segmentation result map mapping model; S43. Based on the current final aerial image quality influencing factor dataset, initialize the drone, aerial flight path, and corresponding shooting equipment for the aerial photography of the siltation in front of the sluice gate; after the settings are completed, perform aerial photography to obtain the current siltation aerial image dataset; perform stitching and correction operations on all image data in the current siltation aerial image dataset to obtain the current processed siltation aerial image data.

7. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating UAV aerial photography as described in claim 6, characterized in that, S5 includes the following steps: S51. Set the initial turbulent image segmentation accuracy; if there is a turbulent part in the currently processed siltation aerial image data, input the currently processed siltation aerial image data and the initial turbulent image segmentation accuracy into the final turbulent siltation image segmentation mapping model to perform image segmentation operation, and obtain the currently segmented turbulent image data and the currently segmented calm image data; otherwise, no segmentation operation is required, and the currently processed siltation aerial image data is used as the currently segmented calm image data. S52. Preprocess the current segmented turbulent image data to obtain the current processed turbulent image data; input both the current processed turbulent image data and the current segmented calm image data into the final calm sedimentation segmentation result map mapping model for mapping to obtain the current first sedimentation segmentation result map and the current second sedimentation segmentation result map.

8. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating UAV aerial photography as described in claim 7, characterized in that, The S5 also includes: S53. Calculate the siltation area and siltation volume based on the current first siltation segmentation result map and the current second siltation segmentation result map to obtain the current aerial siltation area data and the current aerial siltation volume data.

9. The method for automatic identification of siltation in the inlet channel of a sluice gate by integrating drone aerial photography according to any one of claims 8, characterized in that, S6 includes the following steps: S61. Conduct multiple drone aerial photography experiments to identify siltation in the sluice gate's inlet channel. Each experiment repeats steps S31, S32, S33, S43, S51, S52, and S53, and the initial turbulent image segmentation accuracy is different in each experiment. When there is a turbulent water flow in the processed siltation aerial image data, obtain the initial turbulent image segmentation accuracy data set in each experiment and the total time consumed by repeating steps S51, S52, and S53 to obtain the turbulent image segmentation accuracy experimental dataset and the drone aerial siltation calculation time dataset. S62. Based on the experimental dataset of turbulent image segmentation accuracy and the dataset of UAV aerial photography sedimentation calculation time, construct a mapping model with turbulent image segmentation accuracy data as input and UAV aerial photography sedimentation calculation time data as output, and obtain the final sedimentation calculation time mapping model. S63. Set the turbulent image segmentation accuracy range according to the actual turbulent water flow segmentation requirements; randomly initialize multiple sets of turbulent image segmentation accuracy data according to the turbulent image segmentation accuracy range, and input them into the final sedimentation calculation time mapping model for mapping to obtain the current aerial sedimentation calculation time dataset. The turbulent image segmentation accuracy data corresponding to the aerial sedimentation calculation time data with the smallest calculation time in the current aerial sedimentation calculation time dataset is selected as the final turbulent image segmentation accuracy data.

10. A system for implementing the automatic identification method for siltation in the inlet channel of a sluice gate based on any one of claims 1-9, characterized in that: It includes modules for setting the quality and influencing factors of siltation identification, constructing a mapping model for aerial flight route characterization indicators, adjusting the influencing factors of current image quality, constructing a siltation image segmentation model and acquiring and processing the current aerial image, calculating the current siltation area and volume, and filtering the segmentation accuracy of turbulent images. The siltation identification quality and influencing factor type setting module constructs flight path quality characterization indicators and image quality influencing factors for siltation identification in the sluice gate's inlet channel by UAV aerial photography, and sets the range range of each indicator; The aerial flight route characterization index mapping model construction module collects multiple sets of historical flight route quality characterization indexes and image quality influencing factor data based on the siltation identification quality and influencing factor type setting module, and constructs the final aerial flight route characterization index mapping model based on this. The current image quality influencing factor adjustment module collects the current image quality influencing factor data and inputs it into the mapping model in the aerial flight route characterization index mapping model construction module for mapping; if the mapping result does not meet the range of each index in the siltation identification quality and influencing factor type setting module, the current image quality influencing factor data is adjusted until the range is met; Otherwise, no adjustment is needed; The siltation image segmentation model construction and current aerial image acquisition and processing module collects multiple sets of siltation image data under turbulent and calm conditions during the flood season and annotates them to train the turbulent water flow segmentation model and the calm water area segmentation model respectively; and performs siltation aerial photography based on the factor data adjusted in S3, and stitches and corrects the collected images after aerial photography. The current siltation area and volume calculation module segments the turbulent water area based on the initial accuracy parameters, extracts turbulent and calm areas, and uses the siltation image segmentation model construction and the turbulent and calm segmentation model in the current aerial image acquisition and processing module to process different areas, obtain double segmentation results, and calculate the siltation area and volume. The turbulent image segmentation accuracy screening module conducts multiple rounds of experiments. In each experiment, it repeats the current image quality influencing factor adjustment module, the siltation image segmentation model construction and the siltation aerial photography operation in the current aerial image acquisition and processing module, as well as the current siltation area and volume calculation module. It collects time consumption data under different segmentation accuracies, constructs the final siltation calculation time mapping model, and selects the turbulent image segmentation accuracy with the minimum aerial siltation calculation time based on the model.

Images