Water depth inversion method based on bird flock algorithm and dynamic segmentation ensemble learning

Abstract

The invention provides a water depth inversion method based on a bird flock algorithm and dynamic segmentation ensemble learning. The method comprises the following steps: S1, collecting water depth image data of a target area; s2, segmenting the water depth image data through a bird flock algorithm; s3, constructing a dynamic segmented integrated learning water depth inversion model; s4, predicting different segment data by using the dynamic segment integrated learning water depth inversion model to perform water depth prediction; and S5, performing integrated learning on the water depth prediction data of different segments through the integrated learning model to form a new integrated learning model. According to the method, the water depth inversion problem under different water body conditions can be effectively solved through dynamic segmented integrated learning of the water depth inversion model, an integration strategy of a machine learning algorithm and segmented dynamic adjustment of algorithm combination, parameter optimization is further carried out through the dynamic segmented integrated learning of the water depth inversion model in combination with a bird flock algorithm, and the water depth inversion efficiency is improved. And by using multispectral satellite image data, high-precision water depth estimation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of water depth inversion, and in particular to a water depth inversion method based on bird flock algorithm and dynamic segmentation integrated learning. Background Art

[0002] Water depth information is an indispensable part of surveying and managing the underwater environment, and is crucial to waterway safety, coastal construction and fishery development. Accurate and timely water depth information can help ships reduce the possibility of grounding and collision, and can provide necessary data for the planning and construction of coastal facilities. Water depth information also plays a key role in the sustainable management of fisheries and the protection of marine biodiversity. In addition, water depth measurement is also indispensable for monitoring the ecological health of oceans and inland waters, and assessing the environmental impact of human activities. Traditional water depth measurement requires manual measurement on board or physical measurement of water depth using instruments such as sonar, which not only consumes a lot of manpower and material resources, but also is affected by topography and is difficult to carry out in remote areas, which has great limitations. Remote sensing effectively provides a long-term and low-cost method for large-scale water depth measurement. Water depth inversion is an important research field in remote sensing technology research, which mainly estimates the depth of water bodies by the characteristics of reflected light. In recent years, researchers have proposed a variety of water depth inversion models, including empirical models and physical models. Although these technologies have alleviated the limitations of traditional methods to a certain extent, the current water depth inversion technology still faces many challenges, including the complexity of water absorption and scattering characteristics, the limitation of remote sensing data resolution, and the applicability of algorithms under different water conditions. With the development and widespread application of artificial intelligence technology, the application of machine learning in water depth inversion has gradually increased. Machine learning can learn features from data to make predictive classifications. Researchers have combined various machine learning algorithms with high-precision and multi-spectral remote sensing data, including algorithms based on random forests (RF) and support vector machines (SVM), to improve the accuracy and efficiency of inversion. The existing technology combines satellite images and bathymetric data to evaluate the accuracy of multi-factor linear regression and machine learning models in depth mapping of lakes on the Qinghai-Tibet Plateau. The highest accuracy was achieved using the random forest model, R 2The value is 0.9311, the MAE value is 1.13 meters, and the changes in water level and reserves are reliably estimated, providing a new method for the study of shallow lake depth and changes. However, this study only conducted an inversion experiment on a certain lake, and no other comparative experiments were conducted, so the generalization is poor. In the prior art, ETOPO2022, multi-beam bathymetry and SAR images were used to establish a depth inversion dataset containing 542,588 data points, and a random forest (RF) model was implemented to perform depth inversion from satellite images. The model achieved a mean relative error (MRE) of 3.72% and a root mean square error (RMSE) of 2.28 meters on an independent dataset. The overall trend is accurate when applied over a large range, with a spatial resolution of up to 20×22 meters, especially in shallow waters. It shows strong inversion capabilities. However, the model has a strong dependence on data, and the impact of human activities in the image will also cause large errors in the experimental results. In the prior art, a convolutional neural network (CNN-SLI) integrating spatial location information is proposed to address the problem of insufficient feature extraction of remote sensing data in nearshore water depth inversion. This method effectively extracts deep features of remote sensing data by adding spatial dimension information. Experiments were conducted using GF-6 image data in the waters near Nanshan Port in Sanya City. The results showed that CNN-SLI performed well in water depth inversion, with an RMSE of 1.34 meters, a MAE of 0.94 meters, and an R 2 It is 0.97, which is better than other models. This method not only improves the accuracy of water depth inversion, but also has stronger generalization ability, which is of great significance to nearshore water depth research, navigation management, coastal protection and disaster mitigation. In the prior art, the log band ratio algorithm is improved to solve the problem of nearshore turbid water terrain inversion, and the AdaBoost-GBDT model that integrates GBDT and AdaBoost algorithms is developed. Compared with the other four data sets, it is confirmed that it has high reliability and optimal data coverage and accuracy in all water depth intervals. The present invention proposes a water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning to solve the above problems. Summary of the invention

[0003] In view of this, the object of the present invention is to provide a water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning, so as to at least solve the above problems.

[0004] The technical solution adopted by the present invention is as follows:

[0005] A water depth inversion method based on bird flock algorithm and dynamic segmentation integrated learning, the method comprising the following steps:

[0006] S1, collect water depth image data of the target area;

[0007] S2, segmenting the water depth image data by using the bird flock algorithm;

[0008] S3, construct a dynamic segmented integrated learning water depth inversion model;

[0009] S4, using dynamic segmented ensemble learning water depth inversion model to predict different segmented data for water depth prediction;

[0010] S5. The water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model.

[0011] Furthermore, in step S1, collecting the water depth image data of the target area is specifically as follows:

[0012] Determine the scope and resolution of the target area, collect water depth data by using multi-beam echo sounder, sonar, GPS positioning system, and convert the collected water depth data into image format.

[0013] Furthermore, in step S2, the water depth image data is segmented by the bird flock algorithm as follows:

[0014] The water depth features of the water depth image data are extracted by the bird flock algorithm, and the water depth threshold is set as the parameter ε to segment the image data into different segments.

[0015] Furthermore, in step S3, the dynamic segmented integrated learning water depth inversion model is constructed as follows:

[0016] The random forest machine learning algorithm is selected to preprocess the data of each water depth threshold segment, and the preprocessed data is used to train the machine learning model to form a dynamic segmented integrated learning water depth inversion model for different water depth threshold segments.

[0017] Further, in step S4, the water depth prediction is performed by using the dynamic segmented integrated learning water depth inversion model to predict different segmented data:

[0018] The water depth image data to be predicted is input into the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments, and the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments is used for prediction to obtain the water depth prediction results of each different segment.

[0019] Further, in step S5, the water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model. Specifically, the model is:

[0020] The depth prediction results of the depth prediction results of each different segment are taken as input, and an ensemble learning method is used to perform ensemble learning fusion to obtain an ensemble learning model of different segment sets, and the final depth prediction result is obtained through the ensemble learning model of different segment sets.

[0021] Furthermore, when using the dynamic segmented integrated learning water depth inversion model with different water depth threshold segments for prediction, the water depth threshold parameter ε is adjusted to obtain the water depth prediction result of each different segment, and the water depth prediction result of each different segment is evaluated by the root mean square error RMSE. The RMSE evaluation is expressed as:

[0022]

[0023] Among them, α is an undetermined parameter, y' i Represents the predicted water depth, y i represents the actual value of water depth, and n is the total number of samples.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention discloses a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning. The present invention can effectively deal with the water depth inversion problem under different water conditions by using a dynamic segmented integrated learning water depth inversion model, adopting an integrated strategy of a machine learning algorithm, and dynamically adjusting the algorithm combination by segmentation. The water depth inversion model is further combined with the bird flock algorithm for parameter optimization through the dynamic segmented integrated learning, and multi-spectral satellite image data is used to achieve high-precision water depth estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 It is a schematic diagram of the overall process of a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention.

[0028] Figure 2 It is a schematic diagram comparing the correlation of prediction results of different models of the target area Nanshan Port in a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention.

[0029] Figure 3 It is a schematic diagram comparing the correlation of prediction results of different models for the target area Visakhapatnam Beach in a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention.

[0030] Figure 4It is a schematic diagram comparing the correlation of prediction results of different models of the target area Qilianyu Island in a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention.

[0031] Figure 5 It is a schematic diagram of profile water depth inversion results of Nanshan Port, Visakhapatnam Beach and Qilian Island by different models in a water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention.

[0032] Figure 6 It is a comparative schematic diagram of the evaluation of water depth prediction results under the mean absolute error MAE, mean relative error MRE, root mean square error RMSE and determination coefficient R2 of the water depth threshold parameter ε in a water depth inversion method based on a bird flock algorithm and dynamic segmented integrated learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0034] Reference Figure 1-Figure 6 The present invention provides a water depth inversion method based on bird flock algorithm and dynamic segmentation integrated learning, the method comprising the following steps:

[0035] S1, collect water depth image data of the target area;

[0036] S2, segmenting the water depth image data by using the bird flock algorithm;

[0037] S3, construct a dynamic segmented integrated learning water depth inversion model;

[0038] S4, using dynamic segmented ensemble learning water depth inversion model to predict different segmented data for water depth prediction;

[0039] S5. The water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model.

[0040] Exemplarily, in order to improve the accuracy and stability of bathymetry under different water conditions, the accuracy and robustness of water depth prediction can be significantly improved by collecting data, segmenting the water depth image data through the bird flock algorithm, and constructing a dynamic segmented integrated learning water depth inversion model. Finally, the water depth prediction results of different segments are integrated by the integrated learning model, so as to further optimize the prediction results. The target areas are three geographical locations: Nanshan Port, Visakhapatnam Beach and Qilian Island. Nanshan Port is located in Sanya City, Hainan Province, China (18°18′30″N, 109°7′00″E). The water quality of the port has deteriorated due to human activities, and the water in some areas appears turbid. Visakhapatnam Beach is located in Visakhapatnam District, Andhra Pradesh, India. The beach is mainly composed of fine sand with an average particle size of 0.45 to 0.5 mm, with gentle waves and shallow water, clear sea water, and Qilian Island (16.95°N, 112.31°E).

[0041] In step S1, the water depth image data of the target area is collected as follows:

[0042] Determine the scope and resolution of the target area, collect water depth data by using multi-beam echo sounder, sonar, GPS positioning system, and convert the collected water depth data into image format.

[0043] For example, water depth data is collected by using a multi-beam echo sounder, sonar, and GPS positioning system. The multi-beam echo sounder is an advanced sounding device that can simultaneously transmit and receive multiple sound beams, thereby covering a wider area and providing more detailed water depth data. It is suitable for large-scale deep-sea measurements. Sonar measures water depth by transmitting sound waves and receiving reflected sound waves. It can be used for measurements in small waters or shallow waters. The GPS system is used to determine the precise location of the measurement point, which can ensure that the collected water depth data can be accurately associated with the geographic location. The collected water depth data is then converted into an image format, so that high-resolution water depth image data of the target area can be obtained.

[0044] In step S2, the water depth image data is segmented by the bird flock algorithm as follows:

[0045] The water depth features of the water depth image data are extracted by the bird flock algorithm, and the water depth threshold is set as the parameter ε to segment the image data into different segments.

[0046] Exemplarily, since different algorithms have different prediction accuracy in different water depth ranges, in order to improve the prediction accuracy of the model at different water depths, the water depth is divided into different ranges by setting a water depth threshold. The bird flock algorithm is used to determine the different water depth thresholds of the measured water depth data in the three target areas of Nanshan Port, Visakhapatnam Beach and Qilian Island, so as to dynamically distinguish the depth levels of each water area. For example, in Nanshan Port, the measured data water depth range is divided into four intervals of 0-3.44 meters, 3.44-6.34 meters, 6.34-8.73 meters and 8.73-12.97 meters. The water depth data is mainly concentrated in the range of 3.44-8.73 meters, and the peak of the data is observed at about 5 meters, indicating the main distribution characteristics of the water depth in this area. In the study of Visakhapatnam Beach, the water depth range is divided into four intervals of 0-2.45 meters, 2.45-5.08 meters, 5.08-7.8 meters and 7.8-11.79 meters. Among them, the sample points in the 0-2.45m interval have the highest frequency, while the sample points in the 50-100m, 100-160m and above 160m intervals are relatively evenly distributed. For the Qilianyu study area, the data in the three areas with a water depth below 29.62m are more and relatively evenly distributed. In the deep water area of ​​29.62m to 42.44m, the number of sample points is significantly reduced, indicating that the area of ​​water above 29.62m in the study area is small. The threshold ε is determined by the root mean square error RMSE. The predicted water depth range is expanded by an RMSE value, thereby reducing the difference between the predicted result and the actual water depth and improving the overall accuracy of the water depth inversion. The data with the expanded ε value can be input into the dynamic segmented integrated learning water depth inversion model DSEL model for prediction. According to the water depth interval provided by the bird flock algorithm and the reference water depth provided by the rough water depth prediction, the water depth of the point to be predicted is predicted using a suitable interval algorithm. After the range of the water depth area is expanded, there will be overlapping parts between adjacent areas. For example, when the threshold s=1, the water depth of 10-20 meters and 20-40 meters will be expanded to 9-21 meters and 19-41 meters respectively, and there is a two-meter overlap between the two areas. For the overlapping area in the middle, we use the algorithms of the two areas to make joint decisions. First, make predictions in the two areas respectively, and then take the average to get the final result. In this way, we can combine different models from different perspectives on the same problem, form a comprehensive and optimal prediction result based on multiple prediction results, reduce dependence on a single model, prevent poor prediction results due to model weaknesses, improve the accuracy and robustness of the overall prediction, and ultimately achieve the most comprehensive prediction effect.

[0047] In step S3, the dynamic segmented integrated learning water depth inversion model is constructed as follows:

[0048] The random forest machine learning algorithm is selected to preprocess the data of each water depth threshold segment, and the preprocessed data is used to train the machine learning model to form a dynamic segmented integrated learning water depth inversion model for different water depth threshold segments.

[0049] For example, for each water depth threshold segment, a random forest model is trained. These models together constitute a part of the dynamic segmented integrated learning water depth inversion model. In practical applications, the water depth threshold can be dynamically adjusted as needed, so as to select different segments and corresponding models for water depth inversion. This dynamic adjustment capability enables the model to adapt to different water environments and measurement requirements.

[0050] In step S4, the water depth prediction is performed by using the dynamic segmented integrated learning water depth inversion model to predict different segmented data:

[0051] The water depth image data to be predicted is input into the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments, and the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments is used for prediction to obtain the water depth prediction results of each different segment.

[0052] For example, the water depth image data of the area to be predicted is collected. These data have similar formats and characteristics to the data used in training the model. According to the water depth characteristics and requirements of the area to be predicted, appropriate water depth threshold segments are selected. These segments are consistent with the segments used in training the model. The water depth image data assigned to different segments are input into the corresponding model to perform the prediction operation. Each model will generate a water depth prediction result based on the input data, and the water depth prediction results are collected from each model.

[0053] In step S5, the water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model. Specifically, the following is the process:

[0054] The depth prediction results of the depth prediction results of each different segment are taken as input, and an ensemble learning method is used to perform ensemble learning fusion to obtain an ensemble learning model of different segment sets, and the final depth prediction result is obtained through the ensemble learning model of different segment sets.

[0055] Exemplarily, before this step, the dynamic segmented integrated learning water depth inversion model has been used to predict the water depth image data of each segment, and the corresponding water depth prediction results have been obtained. The water depth prediction results of these segments are used as new input data for integrated learning, and the water depth prediction results of each segment are fused to form a new integrated learning model. Through integrated learning fusion, a new integrated learning model can be obtained so as to integrate the advantages of the prediction results of each segment, improve the accuracy and stability of the prediction, and use the new integrated learning model to predict the water depth of the target area.

[0056] When using the dynamic segmented integrated learning water depth inversion model with different water depth threshold segments for prediction, the water depth threshold parameter ε is adjusted to obtain the water depth prediction result of each different segment, and the water depth prediction result of each different segment is evaluated by the root mean square error RMSE. The RMSE evaluation represents:

[0057]

[0058] Among them, α is an undetermined parameter, y' i Represents the predicted water depth, y i represents the actual value of water depth, and n is the total number of samples.

[0059] For example, the parameter threshold setting of the dynamic segmented integrated learning water depth inversion model DSEL has a great influence on the model prediction effect. The selection of different thresholds will result in different prediction effects, and the segmentation threshold will change due to the influence of different water body data characteristics. In order to deeply explore the optimal parameters of the dynamic segmented integrated learning water depth inversion model DSEL and analyze the impact of parameter changes on model performance, a control experiment can be designed when setting the threshold, that is, four evaluation indicators are selected during prediction, namely, mean absolute error MAE, mean relative error MRE, root mean square error RMSE and determination coefficient R 2 , in the experiment, the number of training and test samples in each experimental area is kept constant, and only the segmentation threshold parameter is adjusted. The formal mathematical expression of the evaluation standard is:

[0060]

[0061] exist Figure 6 In the figure, the three parameters of MAE, MRE and RMSE are all around ε. The lower the value, the smaller the prediction error of the dynamic segmented integrated learning water depth inversion model DSEL is, while R 2 The closer the value is to 1, the smaller the model error is. 2 The higher the value at the ε value, the smaller the prediction error of the model. In the study of Nanshan Port, RMSE performed best at ε = 0.4, while MAE, MRE and R 2The three parameters also achieve the best results at ε = 0.4. In the Visakhapatnam beach area, RMSE performs relatively well in the range of ε = 0.4-0.5, and the other three parameters also perform well at the corresponding thresholds. However, when MAE reaches its lowest value at ε = 0.8, the value of MRE is relatively high, and the model effect is not ideal. In the study of Qilianyu Island, the four parameters performed well in the range of ε = 0.4-0.5, but at ε = 0.6, where the MRE effect is the best, although RMSE reaches the lowest value, MAE is too high. By comparing the experimental results of the three study areas, we found that the correlation between the parameters and RMSE is more significant. When the RMSE value is good, the values ​​of other parameters are also relatively good, and the model prediction is more accurate. In addition, in the turbid water environment, when the ε value is half of the RMSE, the model prediction effect is best.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A water depth inversion method based on bird flock algorithm and dynamic segmentation ensemble learning, characterized in that: The method comprises the following steps: S1, collect water depth image data of the target area; S2, segmenting the water depth image data by using the bird flock algorithm; S3, construct a dynamic segmented integrated learning water depth inversion model; S4, using dynamic segmented ensemble learning water depth inversion model to predict different segmented data for water depth prediction; S5. The water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model.

2. The method for water depth inversion based on bird flock algorithm and dynamic segmented integrated learning according to claim 1 is characterized in that: In step S1, the water depth image data of the target area is collected as follows: Determine the scope and resolution of the target area, collect water depth data by using multi-beam echo sounder, sonar, GPS positioning system, and convert the collected water depth data into image format.

3. The water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning according to claim 2 is characterized in that: In step S2, the water depth image data is segmented by the bird flock algorithm as follows: The water depth features of the water depth image data are extracted by the bird flock algorithm, and the water depth threshold is set as the parameter ε to segment the image data into different segments.

4. The water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning according to claim 3 is characterized in that: In step S3, the dynamic segmented integrated learning water depth inversion model is constructed as follows: The random forest machine learning algorithm is selected to preprocess the data of each water depth threshold segment, and the preprocessed data is used to train the machine learning model to form a dynamic segmented integrated learning water depth inversion model for different water depth threshold segments.

5. The water depth inversion method based on bird flock algorithm and dynamic segmented integrated learning according to claim 4 is characterized in that: In step S4, the water depth prediction is performed by using the dynamic segmented integrated learning water depth inversion model to predict different segmented data: The water depth image data to be predicted is input into the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments, and the dynamic segmented integrated learning water depth inversion model of different water depth threshold segments is used for prediction to obtain the water depth prediction results of each different segment.

6. The method for water depth inversion based on bird flock algorithm and dynamic segmented integrated learning according to claim 5 is characterized in that: In step S5, the water depth prediction data of different segments are integrated and learned through an integrated learning model to form a new integrated learning model. Specifically, the following is the process: The depth prediction results of the depth prediction results of each different segment are taken as input, and an ensemble learning method is used to perform ensemble learning fusion to obtain an ensemble learning model of different segment sets, and the final depth prediction result is obtained through the ensemble learning model of different segment sets.

7. The method for water depth inversion based on bird flock algorithm and dynamic segmented integrated learning according to claim 4 is characterized in that: When using the dynamic segmented integrated learning water depth inversion model with different water depth threshold segments for prediction, the water depth threshold parameter ε is adjusted to obtain the water depth prediction result of each different segment, and the water depth prediction result of each different segment is evaluated by the root mean square error RMSE. The RMSE evaluation represents: Among them, α is an undetermined parameter, y′ i Represents the predicted water depth, y i represents the actual value of water depth, and n is the total number of samples.

Images