Method, system and device for identifying river channel silted foreign matters and medium

Abstract

The invention provides a method, a system, a device and a medium for identifying river channel silting foreign matters. The method for identifying the river channel silting foreign matters comprises the following steps: S1, determining a data acquisition position of the river channel silting foreign matters; s2, collecting river bottom data by using a side-scan sonar, and then supporting a river bottom map by using sonar data processing software; s3, determining sample categories in the river channel sediment foreign matter data set, and making the data set by using an image annotation tool; s4, in combination with a DeeplabV < 3 + > model improved by MobileNet-V3Marge, training, verifying and testing the model by using a data set to obtain a river channel siltation foreign matter recognition model; and S5, constructing a river channel siltation foreign matter identification system based on the river channel siltation foreign matter identification model. According to the method, the distribution rule of the river channel silting anisotropic substances is obtained in combination with the Klein 4000 side-scan sonar, the anisotropic substances can be removed in advance, and normal work of follow-up dredging of a dredger is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of river ecological environment technology, specifically relating to a method, system, device and medium for identifying heterogeneous substances deposited in riverbeds. Technical Background

[0002] River dredging is one of the main methods for solving river siltation. It can lower the riverbed height, increase the water surface area and regulate water storage, improve the river water environment, and enhance flood control, drainage, irrigation, and water supply capabilities. Traditional dredging work is generally carried out during the dry season, involving manual on-site dredging after draining the river water, which is time-consuming and inefficient. Using dredgers for river dredging can be done around the clock, without the need to drain the river, resulting in high efficiency. However, due to the presence of foreign objects on the riverbed, such as large rocks, aquatic plants, and fishing nets, and the inability to avoid these objects during underwater operations, the dredger's cutter head can be damaged, reducing operating time and efficiency. Therefore, determining the distribution pattern of foreign objects in the riverbed allows for their early removal, ensuring the normal operation of subsequent dredging.

[0003] Riverbed information is typically acquired using side-scan sonar. The intensity of the received scattered waves is used to distinguish riverbed targets, ultimately forming a map of the distribution of sedimentary anomalies in the river channel. Traditional manual interpretation classifies riverbed targets based on the color of pixels in the acoustic image: hard, raised riverbeds or targets produce strong echoes and are brightly colored; soft, concave riverbeds or targets produce weak echoes and are darkly colored; obscured riverbeds or targets do not produce echoes. This method requires a high level of expertise from interpreters and is inefficient. The introduction of machine learning has improved computational efficiency to some extent, but it still requires manual feature selection, and the computational models are prone to overfitting and have weak generalization ability, making them unsuitable for widespread practical applications. Summary of the Invention

[0004] The first objective of this invention is to provide a method for identifying foreign objects deposited in riverbeds, addressing the aforementioned problems. This method aims to accurately identify and detect foreign objects on the riverbed, determine their distribution patterns, and remove them in advance to ensure the normal operation of subsequent dredging operations.

[0005] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0006] A method for identifying heterogeneous sediment deposits in river channels, characterized by the following steps:

[0007] S1: Determine the data collection location for riverbed sedimentation;

[0008] S2: Use side-scan sonar to collect riverbed data, and then use sonar data processing software to support the riverbed map;

[0009] S3: Determine the sample categories in the riverbed sediment heterogeneous material dataset and create a dataset using an image annotation tool;

[0010] S4: Combine the DeeplabV3+ model improved with MobileNet-V3 Large, use the dataset to train, validate and test the model to obtain a river channel sedimentation heterogeneous object identification model;

[0011] S5: Construct a river sedimentation foreign object identification system based on the river sedimentation foreign object identification model.

[0012] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0013] As a preferred technical solution of the present invention: In step S1, the standard for selecting the location for collecting data on river siltation is as follows: (1) The riverbanks are urban residential areas, and there are residential siltation on the riverbed; (2) The width exceeds 5m, which is convenient for the operation of sonar boats.

[0014] As a preferred technical solution of the present invention: In step S2, the operating parameters of the side-scan sonar are as follows: frequency is 100kHz for low frequency and 400kHz for high frequency; beam angle is 0.7° horizontal and 50° vertical at 100kHz; horizontal and 0.3° vertical at 50° at 400kHz; range is 600m at 100kHz and 200m at 400kHz; length is 1.42m and weight is 27.3Kg.

[0015] As a preferred technical solution of the present invention: In step S3, the sample categories in the riverbed sedimentation anomaly dataset are: rocks, bridge piers, riverbanks, and blank areas. The riverbed sedimentation anomaly geophysical dataset selects rocks, bridge piers, riverbanks, and blank areas, including the identified target and other interfering targets. Among them, rocks are the main anomaly identification target and also the main sedimentation obstacle anomaly that interferes with the normal operation of dredging vessels. Bridge piers and riverbanks are the main features of the riverbed. Blank areas are due to the inability of side-scan sonar to perform geophysical exploration on the vertical area below the sonar, resulting in blank areas in the generated riverbed geophysical map, which do not belong to the target of the riverbed itself.

[0016] As a preferred technical solution of the present invention: In step S3, the process of creating the dataset is as follows: the images of the relevant regions are annotated using an image annotation tool, each JPG format file corresponds to a JSON file, and the JPG format file is converted into a PNG format file. Finally, the dataset is created according to the VOC data format.

[0017] As a preferred technical solution of the present invention: In step S3, before the dataset is created, the images of the relevant regions are horizontally or vertically flipped, scaled, or rotated to achieve data augmentation. By performing horizontal / vertical flipping, scale transformation, and rotation operations on the images, the number of training samples and labels is increased, which can improve sample diversity and feature diversity, solve the shortcomings of small samples, enable the model to fit better, and enhance the robustness and richness of the heterogeneous object recognition model.

[0018] As a preferred technical solution of the present invention: In step S4, the feature extraction module in the DeeplabV3+ model is replaced with MobileNet-V3 Large, which greatly improves the speed of model feature inference and target segmentation, and increases the efficiency of practical applications. At the same time, it greatly reduces the recognition model's demand for computing resources, which is more conducive to the promotion and application of the present invention.

[0019] As a preferred embodiment of the present invention, the data collection and testing river areas are different. The data collection area is from (N30.73249°, E 120.73287°) to (N 30.73228°, E 120.74070°), with a length of 0.76 km, and from (N30.72906°, E120.73027°) to (N 30.72865°, E 120.74132°), with a length of 1.10 km. The testing river area is from (N30.73186°, E120.73687°) to (N 30.72883°, E 120.73768°), with a length of 0.38 km. Separating the collection area and the testing area can verify the accuracy and generalization ability of the river sedimentation anomaly identification model.

[0020] The second objective of this invention is to provide a system for identifying heterogeneous objects deposited in riverbeds.

[0021] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0022] A system for identifying heterogeneous sediment deposits in river channels, characterized in that the system comprises the following modules:

[0023] - Riverbed sedimentation material data collection location determination module, the riverbed sedimentation material data collection location module is used to determine the riverbed sedimentation material data collection location;

[0024] - Riverbed mapping module, which is used to construct a riverbed map based on riverbed data collected by side-scan sonar;

[0025] - Dataset creation module, which is used to determine the sample categories in the riverbed sediment heterogeneous material dataset and create the dataset using image annotation tools;

[0026] - Riverbed sedimentation heterogeneous object identification model construction module, which is used to combine the DeeplabV3+ model improved by MobileNet-V3 Large, and use the dataset to train, validate and test the model to obtain the riverbed sedimentation heterogeneous object identification model.

[0027] The third objective of this invention is to provide a device for identifying foreign objects deposited in riverbeds.

[0028] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0029] A device for identifying heterogeneous sediment deposits in river channels, the device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other via the communication bus, characterized in that:

[0030] The memory, which is used to store computer programs,

[0031] A processor for executing a computer program stored in a memory to implement the steps of the riverbed sediment foreign object identification method as described above.

[0032] Another object of the present invention is to provide a computer-readable storage medium.

[0033] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0034] A computer-readable storage medium, characterized in that: the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying heterogeneous objects in riverbed siltation as described above.

[0035] This invention provides a method, system, device, and medium for identifying heterogeneous sediment deposits in river channels. It identifies heterogeneous sediment deposits from riverbed maps acquired using a Klein 4000 side-scan sonar and obtains a distribution map of these deposits. This allows for the early removal of the deposits, ensuring the smooth operation of subsequent dredging. Specifically, the method includes the following steps: S1: Determining the location for collecting heterogeneous sediment data; S2: Collecting riverbed data using a side-scan sonar and then using Sonarwiz7 to support the riverbed map; S3: Determining the sample categories in the heterogeneous sediment dataset and creating a dataset using labelme; S4: Using a modified version of Deeplab-v3+ (MobileNet-V3 Large), training, validating, and testing the model with the dataset to obtain a heterogeneous sediment identification model; S5: Constructing a heterogeneous sediment identification system based on the identification model. To verify the practicality of this invention, the heterogeneous sediment identification model was applied to a test river, resulting in a distribution map of the heterogeneous sediment deposits. This invention combines the distribution pattern of foreign objects in riverbed sedimentation obtained by Klein4000 side-scan sonar, which can remove foreign objects in advance and ensure the normal operation of subsequent dredging.

[0036] Specifically, the present invention has the following beneficial effects:

[0037] 1) Strong robustness: The geophysical dataset of river sedimentation anomalies not only includes sedimentation anomalies and other characteristic targets in the river, but also increases the sample diversity and feature diversity of the data through data augmentation methods such as horizontal / vertical image flipping, scale transformation and rotation operations. This can enhance the feature extraction capability of the model and avoid the interference of external factors such as distortion and similar targets on the stability and accuracy of model detection.

[0038] 2) Good generalization ability: On the one hand, river channel features can interfere with the identification of sedimentary debris. Adding relevant category targets to the dataset allows the identification model to learn the different features between them, effectively distinguishing sedimentary debris from interfering targets. On the other hand, separating the data collection area and the test area can better verify the model's practical application ability and directly verify the model's generalization ability.

[0039] 3) Strong Model Applicability: The model for identifying silt deposits uses MobileNet-V3 Large as its feature extraction module, reducing the number of model parameters and model size, and significantly improving the model's recognition speed. Although it may reduce accuracy somewhat, this has been verified and will not have a significant impact. In practical applications, the requirements for computing resources are reduced, facilitating model migration and multi-platform operation, thus increasing the model's applicability. Attached Figure Description

[0040] Figure 1This is a technical roadmap for the method of identifying heterogeneous sediment deposits in river channels provided by the present invention.

[0041] Figure 2 This is a schematic diagram showing the data collection locations for heterogeneous sediments in the river channel.

[0042] Figure 3 This is a diagram of a side-scan sonar device.

[0043] Figure 4 This is an example image of a dataset of heterogeneous sediment deposits in a river channel.

[0044] Figure 5 Illustration of a model for identifying heterogeneous objects deposited in river channels.

[0045] Figure 6 This is a map showing the distribution of heterogeneous sediments on the riverbed in the test area. Detailed Implementation

[0046] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0047] like Figures 1-6 As shown, a method for identifying heterogeneous sediment deposits in river channels includes the following steps:

[0048] (1) Location for collecting data on river channel sedimentation anisotropy

[0049] The data collection locations for riverbed sediment deposits were determined to be within the four points: (N 30.73249°, E 120.73287°), (N 30.73228°, E 120.74070°), (N 30.72906°, E 120.73027°), and (N 30.72865°, E 120.74132°). Figure 2 As shown.

[0050] (2) Side-scan sonar to collect riverbed data

[0051] The selected river area was scanned using a Klein 4000 side-scan sonar to create a riverbed map. The side-scan sonar system consists of three parts: a TPU, a computer host, and a transducer (towed fish). The operating parameters of the side-scan sonar are as follows: frequency: 100kHz (low frequency), 400kHz (high frequency); beam angle: 0.7° (horizontal), 50° (vertical) @ 100kHz; 0.3° (horizontal), 50° (vertical) @ 400kHz; range: 600m @ 100kHz, 200m @ 400kHz; length: 1.42m; weight: 27.3kg.

[0052] See relevant side-scan sonar detection instruments Figure 3 The data processing software used was Sonarwiz 7.

[0053] (3) Data creation of heterogeneous sediments in river channels

[0054] First, based on the target types in the river map, the dataset sample categories were selected as stones, bridge piers, riverbanks, and blank areas, and the relevant regions were extracted as images. Then, Labelme was installed and used to annotate the extracted images. Each JPG file's annotation result corresponds to a JSON file, which is then converted into a PNG image, as shown in the example below. Figure 4 As shown. Finally, the dataset was created according to the VOC data format.

[0055] (4) Riverbed sedimentation heterogeneous object identification model

[0056] First, based on the Deeplab-v3+ model, the feature extraction module was replaced with MobileNet-V3 Large, with the following structure: Figure 5 As shown. Then, the model is trained using a dataset of riverbed sedimentary anomalies, employing pre-trained model weights and setting a maximum of 500 iterations. Finally, the model is tested. The trained model is tested, and training continues if the test Accuray score is below 0.9. When the test mAP (mean Average Precision) is greater than 0.9, this model is selected as the riverbed sedimentary anomaly recognition model.

[0057] (5) Practical application testing

[0058] The riverbed sedimentation anomaly identification model obtained in (4) was applied to the test river channel (N 30.73186°, E120.73687°) to (N30.72883°, E 120.73768°), with a length of 0.38 km. A Klein 4000 side-scan sonar was used to obtain the riverbed map, as shown below. Figure 6 As shown in (a). Then, the data is input into the river channel sedimentation anomaly identification system to obtain the final river channel anomaly distribution map, as shown in (a). Figure 6 As shown in (b).

[0059] The present invention also provides a system for identifying foreign objects deposited in river channels, comprising the following modules:

[0060] - Riverbed sedimentation material data collection location determination module, the riverbed sedimentation material data collection location module is used to determine the riverbed sedimentation material data collection location;

[0061] - Riverbed mapping module, which is used to construct a riverbed map based on riverbed data collected by side-scan sonar;

[0062] - Dataset creation module, which is used to determine the sample categories in the riverbed sediment heterogeneous material dataset and create the dataset using image annotation tools;

[0063] - Riverbed sedimentation heterogeneous object identification model construction module, which is used to combine the DeeplabV3+ model improved by MobileNet-V3 Large, and use the dataset to train, validate and test the model to obtain the riverbed sedimentation heterogeneous object identification model.

[0064] The present invention also provides a device for identifying foreign objects deposited in river channels. The device includes a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The device is characterized by:

[0065] The memory, which is used to store computer programs,

[0066] A processor for executing a computer program stored in a memory to implement the steps of the riverbed sediment foreign object identification method as described above.

[0067] The present invention also provides a computer-readable storage medium, characterized in that: the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying heterogeneous objects in riverbed siltation as described above.

[0068] The above specific embodiments are used to explain and illustrate the present invention, and are only preferred embodiments of the present invention, not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for identifying heterogeneous sediment deposits in river channels, characterized in that: The method includes the following steps: S1: Determine the data collection location for riverbed sedimentation; S2: Use side-scan sonar to collect riverbed data, and then use sonar data processing software to support the riverbed map; S3: Determine the sample categories in the riverbed sediment heterogeneous material dataset and create a dataset using an image annotation tool; S4: Combine the DeeplabV3+ model improved with MobileNet-V3 Large, use the dataset to train, validate and test the model to obtain a river channel sedimentation heterogeneous object identification model; S5: Construct a river sedimentation foreign object identification system based on the river sedimentation foreign object identification model.

2. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S1, the criteria for selecting the location for collecting data on riverbed sediment are: (1) the riverbanks are urban residential areas with sediments of domestic use on the riverbed; (2) the width exceeds 5m to facilitate the operation of sonar vessels.

3. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S2, the operating parameters of the side-scan sonar are as follows: frequency: 100kHz for low frequency and 400kHz for high frequency; beam angle: 0.7° horizontal and 50° vertical at 100kHz; 0.3° horizontal and 50° vertical at 400kHz; range: 600m at 100kHz and 200m at 400kHz; length: 1.42m; weight: 27.3Kg.

4. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S3, the sample categories in the riverbed sediment heterogeneous material dataset are: stones, bridge piers, riverbanks, and blank areas.

5. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S3, the dataset is created as follows: images of relevant regions are annotated using an image annotation tool, each JPG file corresponds to a JSON file, and the JPG files are converted to PNG files. Finally, the dataset is created according to the VOC data format.

6. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S3, before creating the dataset, the images of the relevant regions are horizontally or vertically flipped, scaled, or rotated to achieve data augmentation.

7. The method for identifying heterogeneous sediment deposits in river channels according to claim 1, characterized in that: In step S4, the feature extraction module in the DeeplabV3+ model is replaced with MobileNet-V3 Large.

8. A system for identifying foreign objects deposited in river channels, characterized in that: The system includes the following modules: - Riverbed sedimentation material data collection location determination module, the riverbed sedimentation material data collection location module is used to determine the riverbed sedimentation material data collection location; - Riverbed mapping module, which is used to construct a riverbed map based on riverbed data collected by side-scan sonar; - Dataset creation module, which is used to determine the sample categories in the riverbed sediment heterogeneous material dataset and create the dataset using image annotation tools; - Riverbed sedimentation heterogeneous object identification model construction module, which is used to combine the DeeplabV3+ model improved by MobileNet-V3 Large, and use the dataset to train, validate and test the model to obtain the riverbed sedimentation heterogeneous object identification model.

9. A device for identifying foreign objects deposited in river channels, the device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other via the communication bus, characterized in that: The memory, which is used to store computer programs, A processor for executing a computer program stored in a memory to implement the steps of the method for identifying heterogeneous sediment deposits in a riverbed as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying heterogeneous sediment deposits in river channels as described in any one of claims 1-7.

Images