Deep learning-based river inlet discharge port identification system and deep learning-based river inlet discharge port identification method

Abstract

The invention discloses a deep learning-based river entrance discharge port identification system and method, and the system comprises an image collection module which employs a plurality of monitors to collect original image data of a plurality of monitoring regions of a target river channel within a target monitoring duration; the data transmission module stores the original image data in real time and uploads the original image data to a server; the data processing module is used for carrying out image processing on the original image data to obtain model input data; the automatic identification module is used for storing a pre-trained deep learning model, and the model input data are input into the deep learning model to obtain a river inlet discharge port identification result output by the deep learning model; wherein the deep learning model is obtained by training a pre-constructed deep learning network by using the image sample data of the target river channel. Therefore, dynamic identification of the river inlet is realized, and timely data support is provided for water environment management.

Description

Technical Field

[0001] The present invention relates to the field of environmental protection, and in particular to a river outlet identification system and method based on deep learning. Background Art

[0002] With the continuous acceleration of industrialization and urbanization, environmental problems have become increasingly prominent, especially water pollution has become an important restrictive factor for the sustainable development of rivers. By tracing the source of water pollution, it is possible to effectively assist the supervision and control of the water environment, thereby ensuring the safety of the water environment. The source of pollution mainly comes from the discharge of pollutants from river outlets, and the number of river outlets in cities is often very large, making investigations extremely difficult. At present, the tracing of river outlets is mainly based on manual investigations, investigating the drainage outlets along the river in towns one by one, collecting data such as location, attributes, pipe diameter, water flow, and then forming a drainage outlet and upstream tracing pipe network map.

[0003] In the traditional way, the investigation method of river outlets mainly includes several aspects: first, it is necessary to collect and integrate the planning, documents, drawings and other technical information related to the river outlets, clarify the scope of investigation and key investigation objects; then, in accordance with the requirements of "all outlets should be investigated, and all should be investigated", a three-level investigation of sewage outlets is carried out by comprehensively adopting remote sensing interpretation to identify suspected sewage outlets, manual on-foot investigation to confirm information, information review of problematic sewage outlets and detailed verification of hot spots; then, for areas where there is no condition to lower the water level of the investigated water body, the inspection wells connected to the drainage outlets on the shore are investigated, and the situation in the inspection wells is checked to determine whether there are underwater outlets and their types. Unmanned boats + side-scan sonar technology can also be used for comprehensive scanning to investigate unknown outlets; finally, according to the results of the sewage outlet investigation, the discharge source and type of the sewage outlet can be traced through data tracing, manual tracing, technical tracing and other methods.

[0004] However, the investigation of river outlets, which is mainly based on manual investigation, is difficult, has slow dynamic update capabilities, and a long update cycle. At the same time, due to the varying levels of technical personnel, the survey data has large errors, and the workload of later correction and verification is large. In addition, the existing automatic monitoring system for river outlets has poor functionality, weak automation capabilities, and lacks data analysis capabilities based on big data.

[0005] Deep learning is a subfield of machine learning. It is widely used in the fields of medicine, finance, etc. based on artificial neural networks, and is also partially involved in the field of environmental engineering. Deep learning models can automatically learn features from raw data, which reduces the dependence on professional domain knowledge and can process unstructured data such as images, sounds, and text. In many tasks, such as image recognition, deep learning models have reached or exceeded human-level accuracy. Well-trained deep learning models can generalize well to unseen data, which means they can handle new and unknown inputs. Deep learning allows the mapping from input to output to be learned directly, reducing the need for intermediate steps and simplifying the entire learning process. New network architectures continue to emerge in the field of deep learning, such as convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory networks (LSTM), etc. These architectures are optimized for specific types of data and tasks.

[0006] Although deep learning has many advantages, it also has some challenges, the most important of which is the large amount of data required. Neural network models require a large amount of raw data for model training and verification to ensure the high accuracy of deep learning models. In fact, the amount of survey data on river outlets in cities is often very large, which just provides sufficient basic data for deep learning models to train and learn.

[0007] In the existing technology, the method of automatically identifying the outlet of the river can be used by using drone remote sensing images, but the drone image data is relatively small, and the accuracy of the deep learning model cannot be guaranteed. Secondly, the data is often identified based on the influence of a certain point in time, and there is no automatic update function, and it cannot achieve automatic update and identification over time. In addition, a convenient identification system has not been formed. Summary of the invention

[0008] To this end, an embodiment of the present invention provides a river outlet identification system and method based on deep learning, in order to achieve dynamic identification of river outlets and provide timely data support for water environment management.

[0009] In order to achieve the above purpose, the embodiment of the present invention provides the following technical solutions:

[0010] The present invention provides a river outlet identification system based on deep learning, the system comprising:

[0011] An image acquisition module is used to respectively acquire original image data of multiple monitoring areas of a target river channel through multiple monitors within a target monitoring period;

[0012] A data transmission module, used for storing in real time and uploading the original image data to a server;

[0013] A data processing module, used for performing image processing on the original image data to obtain model input data;

[0014] An automatic identification module, used for storing a pre-trained deep learning model, wherein the model input data is input into the deep learning model to obtain a river outlet identification result output by the deep learning model;

[0015] The deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

[0016] The monitors are arranged at intervals in a zigzag structure along the target river channel.

[0017] In some embodiments, the data transmission module is further configured to:

[0018] intercepting the original image data at intervals of a preset time length;

[0019] Constructing a file name for the captured image data, wherein the file name at least includes the name of the target river, the number of the monitor, and the image capture time;

[0020] The image data with the file name is uploaded to the server.

[0021] In some embodiments, performing image processing on the original image data to obtain model input data specifically includes:

[0022] Using image edge detection algorithm to pre-process the acquired raw image data;

[0023] The preprocessed image is subjected to secondary denoising using a morphological algorithm, wherein the morphological algorithm includes dilation, erosion, opening operation and closing operation.

[0024] In some embodiments, the pre-built deep learning network is trained using the image sample data of the target river to obtain the deep learning model, specifically including:

[0025] Acquire image sample data of the target river channel, and perform preprocessing and data calibration on the image sample data to construct a river outlet image database;

[0026] Based on the image database, the pre-built deep learning network is trained, verified and tested in multiple training dimensions to obtain the deep learning model.

[0027] In some embodiments, the training dimensions specifically include:

[0028] Image feature extraction, including color, texture and edge feature extraction of the input image, as well as the feature extraction of the outlet shape, outlet structure and outlet pattern of the input image;

[0029] Classification task learning, including category recognition and multi-category classification;

[0030] Object detection learning, which involves determining the location and bounding box of objects in an image and identifying the category of the object;

[0031] Image generation learning, including generating images with specific styles or contents by learning the distribution and features of images; learning to repair and reconstruct partially damaged or missing images;

[0032] Transfer learning, which includes fine-tuning for a specific task, transferring knowledge learned in one domain to another related domain, and adjusting the number of times the model is trained based on the learning rate presented by the model.

[0033] In some embodiments, the system is integrated into a web page and / or an APP.

[0034] The present invention also provides a method for identifying a river outlet based on deep learning, the method comprising:

[0035] During the target monitoring period, collect original image data of multiple monitoring areas of the target river;

[0036] Performing image processing on the original image data to obtain model input data;

[0037] Inputting the model input data into the deep learning model to obtain the river outlet identification result output by the deep learning model;

[0038] The deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

[0039] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0040] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the above method when executed by a processor.

[0041] The river outlet recognition system and method based on deep learning provided by the present invention uses an image acquisition module to collect original image data of multiple monitoring areas of the target river channel using multiple monitors during the target monitoring period; the data transmission module stores the original image data in real time and uploads it to the server; the data processing module performs image processing on the original image data to obtain model input data; the automatic recognition module stores a pre-trained deep learning model, and the model input data is input into the deep learning model to obtain the river outlet recognition result output by the deep learning model; wherein the deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river channel. Thus, dynamic recognition of river outlets can be achieved, timely data support can be provided for water environment management, the regional water environment intelligent management capability can be improved, the water environment risk prevention and monitoring capability can be further enhanced, and the water environment management level can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0043] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0044] Figure 1 This is a structural block diagram of the automatic monitoring and management system for river outlets based on deep learning provided by the present invention;

[0045] Figure 2 This is a layout effect diagram of the monitor provided by the present invention;

[0046] Figure 3 This is one of the flow charts of the method for automatic monitoring and management of river outlets based on deep learning provided by the present invention;

[0047] Figure 4 The second flowchart of the method for automatic monitoring and management of river outlets based on deep learning provided by the present invention;

[0048] Figure 5This is a structural block diagram of a computer device provided by the present invention. DETAILED DESCRIPTION

[0049] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] In view of the problems existing in the prior art, the present invention provides an automatic monitoring and management system for river outlets based on deep learning, which mainly includes an image acquisition module, a data transmission module, a data processing module, and an automatic identification module. The system also has functions such as information query, platform management, and new river outlet prompts, and finally forms a web client and a mobile phone APP to complete the dynamic identification of river outlets. It provides timely data support for water environment management, improves the regional water environment intelligent management capabilities, further enhances the water environment risk prevention and monitoring capabilities, and improves the water environment management level.

[0051] In a specific embodiment, Figure 1 As shown, the river outlet identification system based on deep learning provided by the present invention includes the following modules:

[0052] The image acquisition module 110 is used to collect raw image data of multiple monitoring areas of the target river through multiple monitors during the target monitoring period; the target monitoring period is, for example, one month, two months or half a year. The specific target monitoring period can be determined according to the usage scenario. This article is only an example and does not make specific limitations; the target river refers to the river that needs to be detected. In the specific implementation process, Figure 2 As shown, monitors are arranged on both sides of the river to conduct all-round monitoring of the river. The positions of the monitors (such as cameras) should be arranged in a zigzag shape, and the intervals should be arranged according to the actual shooting effects. In principle, the monitors in the same river should be arranged to cover the entire river range; the monitor pixels should not be less than 10 million. When arranging the monitors, attention should be paid to ensure that the shooting screen is unobstructed and the height is moderate.

[0053] The data transmission module 120 is used to store the original image data in real time and upload it to the server. For example, the monitor has its own data storage module to store the monitoring data in real time, or an additional configured memory can be used for data storage and transmission. The server can be a local server or a cloud server.

[0054] The data processing module 130 is used to perform image processing on the original image data to obtain model input data. Since there is a lot of interference in the original image data, in order to improve the effect of model training and recognition, the input data needs to be processed during model application and model training.

[0055] The automatic identification module 140 is used to store a pre-trained deep learning model. The model input data is input into the deep learning model to obtain the river outlet identification result output by the deep learning model; wherein the deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

[0056] The automatic identification module 140 of data includes the establishment and verification of the deep learning model. When establishing the deep learning model, this embodiment adopts the Interception V3 model. The Interception V3 model is the third generation model in the Google Interception model. Compared with other neural network models, its biggest feature is that it expands the convolution operation between neural network layers and adopts convolution kernels of different sizes, so that there are receptive fields of different sizes, and finally achieves splicing to achieve the fusion of features of different scales. The processed raw data is calibrated, the position of the river outlet is manually identified, and then input into the Interception V3 model for machine learning. According to the learning rate presented by the model, the number of model training times is adjusted. In the verification process, according to the model learning rate, when the number of model training times reaches a certain level, the recognition accuracy of the river outlet will reach a relatively high value and will not increase. Select part of the original image data for verification. When the verification accuracy reaches the recognition accuracy of the training set, it can be considered that the model training is completed, and then the model is built into the cloud server. The image data uploaded subsequently can be automatically input into the neural network for automatic analysis. When the model determines that there is a new outlet into the river, a system prompt will be generated.

[0057] Furthermore, the data transmission module is also used for:

[0058] The original image data is intercepted at each preset time interval, and the preset time interval may be, for example, a desired time interval of 1 hour, 2 hours, 4 hours, 8 hours, etc. The intercepted original image data may be the entire image or a portion of the image;

[0059] Constructing a file name for the captured image data, wherein the file name at least includes the name of the target river, the number of the monitor, and the image capture time;

[0060] The image data with the file name is uploaded to the server; when uploading, a single image can be uploaded, or multiple files can be merged into a folder and uploaded.

[0061] In a specific usage scenario, for example, surveillance images can be captured every 4 hours, uploaded to the cloud server, and saved in folders on the cloud server according to the monitor number. The photo names are formed according to the principle of "river name-monitor number-image acquisition time".

[0062] In the data processing module 130, the original image data is processed to obtain model input data, specifically including:

[0063] The acquired raw image data is preprocessed using an image edge detection algorithm; the image edge information defines the basic information of the object, and different edge combinations convey complex image information. Computer processing of images begins with edge recognition. The edge information in the image can reflect the characteristic differences between local areas, thereby forming the boundary demarcation lines of the areas and segmenting the shapes of each area. The present invention uses an image edge detection algorithm to preprocess the acquired image. After being processed by the edge detection algorithm, the edge information of the image is effectively extracted, but there is still a lot of isolated edge information;

[0064] The preprocessed image is subjected to secondary denoising using a morphological algorithm, which includes dilation, erosion, opening and closing operations. Furthermore, in order to remove invalid edges while retaining the original target edge information, morphology is selected to achieve secondary denoising of the image. The main operations in morphology include dilation, erosion, opening and closing operations, which ultimately form the original data for deep model learning.

[0065] In some embodiments, the pre-built deep learning network is trained using the image sample data of the target river to obtain the deep learning model, specifically including:

[0066] Acquire image sample data of the target river channel, and perform preprocessing and data calibration on the image sample data to construct a river outlet image database;

[0067] Based on the image database, the pre-built deep learning network is trained, verified and tested in multiple training dimensions to obtain the deep learning model.

[0068] The training dimensions specifically include:

[0069] Image feature extraction, including color, texture and edge feature extraction of the input image, as well as the feature extraction of the outlet shape, outlet structure and outlet pattern of the input image;

[0070] Classification task learning, including category recognition and multi-category classification;

[0071] Object detection learning, which involves determining the location and bounding box of objects in an image and identifying the category of the object;

[0072] Image generation learning, including generating images with specific styles or contents by learning the distribution and features of images; learning to repair and reconstruct partially damaged or missing images;

[0073] Transfer learning, which includes fine-tuning for a specific task, transferring knowledge learned in one domain to another related domain, and adjusting the number of times the model is trained based on the learning rate presented by the model.

[0074] In some embodiments, the system is integrated on a web page and / or an APP. That is to say, all module functions are integrated on the web page and mobile APP platform, and the platform should have the following functions: 1) Data management function, which can extract data according to demand classification and calculate the number of river outlets identified; 2) Human-computer interaction function, which allows manual confirmation and adjustment for automatically identified river outlets to prevent errors in river outlet identification due to image blur and other reasons; 3) Newly added river outlet early warning function, for newly identified river outlets, pop-up windows and messages are sent in the system to facilitate verification of the site at the first time. Taking the cloud server as an example, except for the image acquisition module containing the monitor, which needs to be installed and implemented on site, the remaining modules are installed on the cloud server, the on-site image data is uploaded to the cloud, and the processed data is retrieved from the cloud.

[0075] In the above specific implementation, the river outlet recognition system based on deep learning provided by the present invention uses multiple monitors to respectively collect original image data of multiple monitoring areas of the target river during the target monitoring time through the image acquisition module; the data transmission module stores the original image data in real time and uploads it to the server; the data processing module performs image processing on the original image data to obtain model input data; the automatic recognition module stores the pre-trained deep learning model, and the model input data is input into the deep learning model to obtain the river outlet recognition result output by the deep learning model; wherein the deep learning model is obtained by training the pre-constructed deep learning network using the image sample data of the target river. Thereby, the dynamic recognition of the river outlet is realized, timely data support is provided for water environment management, the regional water environment intelligent management capability is improved, the water environment risk prevention and monitoring capability is further enhanced, and the water environment management level is improved.

[0076] In addition to the above system, the present invention also provides a method for identifying river outlets based on deep learning. The method is implemented based on the above system. Figure 3 As shown, the method comprises the following steps:

[0077] S310: collecting original image data of multiple monitoring areas of the target river during the target monitoring time;

[0078] S320: performing image processing on the original image data to obtain model input data;

[0079] S330: inputting the model input data into the deep learning model to obtain the river outlet identification result output by the deep learning model;

[0080] The deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

[0081] In a specific usage scenario, such as Figure 4 As shown, the method comprises the following steps:

[0082] Step 1: Obtain monitoring image information. Specifically, monitors are arranged on both sides of the river to monitor the river in all directions. The camera positions should be arranged in a zigzag pattern, and the intervals should be arranged according to the actual shooting effect. In principle, the monitors in the same river should be arranged to cover the entire river. The image acquisition time interval is once every 4 hours, and a total of 6 times a day. After acquisition, it is automatically stored in the acquisition module.

[0083] Step 2: Image data transmission, using 5G network for data transmission, the image data stored in the acquisition module is automatically uploaded every 2 hours. The photo name is formed according to the principle of "river name-monitor number-image acquisition time", and each monitoring device is stored independently in a folder.

[0084] Step 3: Image data preprocessing, standardizing the image data collected to the cloud. Image edge information defines the basic information of the object. Different edge combinations convey complex image information. Computer processing of images begins with edge recognition. Edge information can reflect the characteristic differences between local areas, thereby forming the boundary demarcation lines of the area and segmenting the shapes of each area. Based on this principle, the canny operator is used to denoise the image and reduce the impact of invalid edge information on edge extraction.

[0085] After edge detection algorithm processing, the edge information of the image is effectively extracted, but there are still many isolated edge information. In order to remove invalid edges while retaining the original target edge information, morphology is used to achieve secondary denoising of the image. The main operations in morphology include: dilation, erosion, opening and closing operations, which ultimately complete the preprocessing of the image and reduce the influence of factors such as lighting, shooting angle, and obstacles.

[0086] Step 4: Establish a deep learning model. As the latest convolutional neural network model, the InceptionV3 model has unique advantages in processing images. The processed raw data is calibrated according to the supervised calibration method, and the information content such as the category, characteristics and location of the known river outlets is annotated. Establish a river outlet image database. The final data set is used for model training (80%), verification (10%) and testing (10%) images, and then input into the InceptionV3 model for machine learning. The learning content mainly includes the following aspects:

[0087] (1) Image feature extraction:

[0088] Low-level features: such as color, texture, edge, etc. InceptionV3 can automatically learn image features at different levels, from simple pixel-level features to more abstract semantic features;

[0089] Advanced features: shape, structure, pattern, etc. of the outlet; through multi-layer convolution and pooling operations, the model can capture the complex structure and pattern information in the image.

[0090] (2) Classification task learning:

[0091] Category recognition: learning to distinguish different image categories. The model establishes a mapping relationship from image features to category labels by learning from a large amount of labeled data.

[0092] Multi-category classification: For images containing multiple categories, learning to recognize multiple categories simultaneously requires the model to be able to handle complex category relationships and overlaps.

[0093] (3) Object detection learning:

[0094] Object localization: Determine the position and bounding box of objects in the image. InceptionV3 can be combined with target detection algorithms to learn to accurately locate different objects in the image.

[0095] Object recognition: While locating the object, identify the category of the object, which requires the model to learn the appearance features and category information of the object at the same time.

[0096] (4) Image generation learning:

[0097] Image synthesis: learning to generate new images; by learning the distribution and features of images, the model can generate images with a specific style or content.

[0098] Image restoration: Learning to repair and reconstruct partially damaged or missing images requires the model to understand the structure and content of the image and to perform reasonable filling and restoration based on surrounding information.

[0099] (5) Transfer learning:

[0100] Fine-tuning of pre-trained models: Using the InceptionV3 model pre-trained on a large-scale dataset, fine-tuning for specific tasks can speed up learning and improve the performance of the model on new tasks. ;

[0101] Cross-domain learning: transfer knowledge learned in one field to another related field;

[0102] According to the learning rate presented by the model, the number of model training times is adjusted to form a deep learning model of how to arrange the outlets based on the convolutional neural network.

[0103] Step 5: Model verification. According to the model learning rate, when the number of model training times reaches a certain level, the recognition accuracy of the river outlet will reach a relatively high value and will not increase any further. Select some original image data for verification. When the verification accuracy reaches the recognition accuracy of the training set, the model training can be considered complete, and then the model is built into the cloud server. The image data uploaded later can be automatically input into the neural network for automatic analysis. When the model determines that there is a new river outlet in the river, a system prompt will be generated.

[0104] Step 6: Finally, the module functions are integrated into the web page and mobile APP platform. The platform should have 1) data management function, which can extract data according to demand and calculate the number of river outlets identified; 2) human-computer interaction function, which allows manual confirmation and adjustment for automatically identified river outlets to prevent errors in river outlet identification due to image blur and other reasons; 3) a new river outlet early warning function, which pops up windows and sends messages in the system for newly identified river outlets to facilitate on-site verification at the first time.

[0105] The implementation method of web page integration is as follows:

[0106] (1) Backend development:

[0107] Choose a suitable backend technology stack, such as Python's Flask or Django framework;

[0108] Build a server environment to receive image data from the front end and input it into the trained InceptionV3 model for prediction;

[0109] Process the output of the model and return it to the front end for display.

[0110] (2) Front-end development:

[0111] Build user interfaces using HTML, CSS, and JavaScript;

[0112] Provides image upload function, you can upload images by file selection or drag and drop;

[0113] Send image data to the backend server and receive the prediction results returned by the server;

[0114] Display prediction results in an intuitive way, such as displaying image category, confidence level, and other information.

[0115] (3) Model deployment:

[0116] Save the trained InceptionV3 model to a suitable format, such as TensorFlow's Saved Model format;

[0117] Load the model on the server and ensure that the model can respond quickly to front-end requests;

[0118] You can consider using server clusters or cloud computing platforms to improve the concurrent processing capabilities of the model.

[0119] The implementation method of mobile APP integration is as follows:

[0120] (1) Select the development platform:

[0121] For Android, you can use Java or Kotlin for development; for iOS, you can use Objective-C or Swift for development;

[0122] You can also consider using a cross-platform development framework, such as Flutter or ReactNative, to reduce development workload.

[0123] (2) Model integration:

[0124] Convert the trained InceptionV3 model to a format suitable for running on mobile devices, such as TensorFlow Lite;

[0125] Load the model in the app and perform image prediction locally to reduce network latency and improve response speed;

[0126] Alternatively, the image data can be sent to a server for prediction, but network connectivity and data security issues need to be considered.

[0127] (3) User interface design:

[0128] Design a simple and easy-to-use user interface to facilitate users to upload images and view prediction results;

[0129] It can provide multiple image acquisition methods such as taking photos and selecting images from the album;

[0130] Display the prediction results in a clear manner, such as pop-up notifications, displaying them on the interface, etc.

[0131] (4) Performance optimization:

[0132] Optimize the model to reduce its size and computational complexity to improve the performance and responsiveness of the app;

[0133] Reasonably process image data to avoid excessive memory usage and lag;

[0134] Conduct sufficient testing to ensure the stability and reliability of the APP on different devices and network environments.

[0135] In the above specific implementation manner, the river outlet identification method based on deep learning provided by the present invention can achieve automatic identification of river outlets through monitoring images and the InceptionV3 convolutional neural network model without the need for human intervention; it provides an effective automation algorithm and a computer program for automatic extraction of river outlets for the interpretation of river outlets. Based on the existing image target recognition algorithm, this method adds a data processing link, greatly increases the authenticity and reliability of the original data, and compared with manual visual interpretation, it can improve the efficiency by about 10 times while ensuring a completeness rate of about 80%. In addition, the automatic river outlet monitoring system formed by the present invention successfully realizes unmanned and intelligent river outlet investigation work, and the formed set of software results is conducive to the regional management of river outlets, with great social and economic benefits.

[0136] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a model prediction. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The model prediction of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0137] Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0138] Corresponding to the above embodiment, the embodiment of the present invention further provides a computer storage medium, which contains one or more program instructions, wherein the one or more program instructions are used to execute the above method.

[0139] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the above method.

[0140] In the embodiment of the present invention, the processor may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0141] The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware.

[0142] The storage medium may be a memory, which may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0143] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable ROM (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0144] The volatile memory may be a random access memory (RAM) which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DRRAM).

[0145] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0146] Those skilled in the art will appreciate that in one or more of the above examples, the functions described in the present invention can be implemented using a combination of hardware and software. When software is used, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any media that facilitates the transmission of computer programs from one place to another. Storage media can be any available media that can be accessed by general or special-purpose computers.

[0147] The above specific implementation methods further illustrate the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present invention should be included in the scope of protection of the present invention.

Claims

1. A river outlet identification system based on deep learning, characterized in that: The system comprises: An image acquisition module is used to respectively acquire original image data of multiple monitoring areas of a target river channel through multiple monitors within a target monitoring period; A data transmission module, used for storing in real time and uploading the original image data to a server; A data processing module, used for performing image processing on the original image data to obtain model input data; An automatic identification module, used for storing a pre-trained deep learning model, wherein the model input data is input into the deep learning model to obtain a river outlet identification result output by the deep learning model; The deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

2. The deep learning-based river outlet identification system according to claim 1, characterized in that: A plurality of the monitors are arranged at intervals in a zigzag structure along the target river channel.

3. The river outlet identification system based on deep learning according to claim 1 is characterized in that: The data transmission module is also used for: intercepting the original image data at intervals of a preset time length; Constructing a file name for the captured image data, wherein the file name at least includes the name of the target river, the number of the monitor, and the image capture time; The image data with the file name is uploaded to the server.

4. The river outlet identification system based on deep learning according to claim 1 is characterized in that: Performing image processing on the original image data to obtain model input data specifically includes: Using image edge detection algorithm to pre-process the acquired raw image data; The preprocessed image is subjected to secondary denoising using a morphological algorithm, wherein the morphological algorithm includes dilation, erosion, opening operation and closing operation.

5. The river outlet identification system based on deep learning according to any one of claims 1 to 4, characterized in that: The pre-built deep learning network is trained using the image sample data of the target river to obtain the deep learning model, specifically including: Acquire image sample data of the target river channel, and perform preprocessing and data calibration on the image sample data to construct a river outlet image database; Based on the image database, the pre-built deep learning network is trained, verified and tested in multiple training dimensions to obtain the deep learning model.

6. The deep learning-based river outlet identification system according to claim 5, characterized in that: The training dimensions specifically include: Image feature extraction, including color, texture and edge feature extraction of the input image, as well as the feature extraction of the outlet shape, outlet structure and outlet pattern of the input image; Classification task learning, including category recognition and multi-category classification; Object detection learning, which involves determining the location and bounding box of objects in an image and identifying the category of the object; Image generation learning, including generating images with specific styles or contents by learning the distribution and features of images; learning to repair and reconstruct partially damaged or missing images; Transfer learning, which includes fine-tuning for a specific task, transferring knowledge learned in one domain to another related domain, and adjusting the number of times the model is trained based on the learning rate presented by the model.

7. The river outlet identification system based on deep learning according to any one of claims 1 to 4, characterized in that: The system is integrated on a web page and / or an APP.

8. A method for identifying river outlets based on deep learning, characterized in that: The method comprises: During the target monitoring period, collect original image data of multiple monitoring areas of the target river; Performing image processing on the original image data to obtain model input data; Inputting the model input data into the deep learning model to obtain the river outlet identification result output by the deep learning model; The deep learning model is obtained by training a pre-constructed deep learning network using the image sample data of the target river.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images