Marine moving target detection method and device, electronic equipment and medium

Abstract

The invention provides a maritime moving target detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining maritime multi-source data, and carrying out the fusion of the multi-source data, and the multi-source data comprises at least one of radar data, satellite data and meteorological data; inputting the fused multi-source data into an adaptive feature extraction network, and outputting a feature vector diagram; detecting a maritime moving target from the feature vector diagram based on a target detection algorithm of deep learning; performing real-time tracking on the marine moving target based on a target tracking algorithm, and identifying the category of the marine moving target based on a deep learning classification algorithm; accurate, efficient and real-time detection of the marine dynamic target is achieved, and powerful technical support is provided for the fields of marine safety monitoring, marine rescue and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean monitoring, and in particular relates to a method, device, electronic equipment and storage medium for detecting moving targets at sea. Background Art

[0002] Detection of moving targets at sea is one of the important tasks in the field of ocean monitoring and maritime security.

[0003] Traditional methods for detecting moving targets at sea mainly rely on sensors such as radar, sonar, and infrared, and achieve detection, tracking, and identification of targets at sea by receiving and processing data collected by sensors. However, these methods face many challenges in practical applications, such as a wide variety of targets, complex and changeable backgrounds, severe noise interference, large amounts of data, and high real-time requirements. Summary of the invention

[0004] In a first aspect, an embodiment of the present invention provides a method for detecting moving targets at sea, comprising: acquiring multi-source data at sea and fusing the multi-source data, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data; inputting the fused multi-source data into an adaptive feature extraction network and outputting a feature vector graph; detecting moving targets at sea from the feature vector graph using a target detection algorithm based on deep learning; tracking the moving targets at sea in real time based on a target tracking algorithm, and identifying the category of moving targets at sea based on a classification algorithm based on deep learning.

[0005] In some embodiments, the adaptive feature extraction network includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer and an output layer; the fused multi-source data is input into the adaptive feature extraction network and the feature vector graph is output, including: receiving the fused multi-source data based on the input layer; extracting local features of different scales of the fused multi-source data based on the multiple convolutional layers; downsampling the local features of different scales based on the pooling layer; flattening the downsampled features into a one-dimensional vector based on the fully connected layer, and extracting global features through a fully connected operation; outputting the global features based on the output layer to form a corresponding feature vector graph.

[0006] In some embodiments, the adaptive feature extraction network is obtained by transfer learning or ensemble learning.

[0007] In some embodiments, the deep learning-based target detection algorithm is a YOLO algorithm, and the deep learning-based target detection algorithm detects marine moving targets from the feature vector map, including: generating a series of candidate regions on the feature vector map according to a sliding window or a predefined anchor frame set; detecting the marine moving targets contained in each candidate region, and adjusting the position and size of each candidate region; and removing redundant candidate regions on the feature vector map through a non-maximum suppression algorithm.

[0008] In some embodiments, the target tracking algorithm is used to track the sea moving target in real time, and the deep learning-based classification algorithm is used to identify the category of the sea moving target, including: determining the initial motion state of the detected sea moving target; tracking the sea moving target based on the Kalman filter algorithm, and updating the motion state of the sea moving target in real time to obtain the motion trajectory of the sea moving target; and identifying the category of the sea moving target based on a convolutional neural network model.

[0009] In some embodiments, after acquiring the offshore multi-source data, the method further includes: preprocessing the multi-source data, wherein the preprocessing includes at least one of the following: data cleaning, data calibration, and data format conversion.

[0010] In some embodiments, the fusing of multi-source data includes: fusing the multi-source data using at least one of a weighted average method, a Kalman filter method, and a Bayesian network method.

[0011] In the second aspect, an embodiment of the present invention provides a device for detecting moving targets at sea, including: a data fusion module, used to acquire multi-source data at sea and fuse the multi-source data, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data; a feature extraction module, used to input the fused multi-source data into an adaptive feature extraction network and output a feature vector map; a target detection module, used to detect moving targets at sea from the feature vector map based on a target detection algorithm based on deep learning; a tracking and identification module, used to track the moving targets at sea in real time based on a target tracking algorithm, and identify the category of moving targets at sea based on a classification algorithm based on deep learning.

[0012] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; the memory is used to store computer programs; and the processor is used to implement the steps of the method for detecting moving targets at sea as described in any one of the first aspects when executing the program stored in the memory.

[0013] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method for detecting moving targets at sea as described in any one of the first aspects are implemented.

[0014] The beneficial effects brought by the present invention are as follows:

[0015] It can be seen from the above scheme that the embodiments of the present invention provide a method, device, electronic device and medium for detecting moving targets at sea. The method achieves accurate, efficient and real-time detection of dynamic targets at sea by fusing multi-source data at sea, constructing an adaptive feature extraction network, a real-time target detection algorithm based on deep learning, and a target tracking and recognition algorithm, thereby providing strong technical support for the fields of maritime safety monitoring, maritime rescue, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for detecting a moving target at sea provided by an embodiment of the present invention;

[0017] Figure 2 for Figure 1 A detailed flow chart of step S102 in the illustrated embodiment;

[0018] Figure 3 for Figure 1 A detailed flow chart of step S103 in the illustrated embodiment;

[0019] Figure 4 A schematic diagram of a flow chart of another method for detecting moving targets at sea provided by an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of the structure of a device for detecting moving targets at sea provided by an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0023] Detection of moving targets at sea is one of the important tasks in the field of ocean monitoring and maritime safety. Traditional methods of detecting moving targets at sea mainly rely on sensors such as radar, sonar, and infrared, and detect, track, and identify targets at sea by receiving and processing data collected by sensors. However, these methods face many challenges in practical applications, such as a wide variety of targets, complex and changeable backgrounds, severe noise interference, large amounts of data, and high real-time requirements.

[0024] With the rapid development of radar technology, especially the emergence of new radar technologies such as synthetic aperture radar (SAR) and phased array radar, the ability to detect moving targets at sea has been significantly improved. These new radars have the advantages of high resolution, wide imaging range, and are not restricted by weather and lighting conditions. They can achieve continuous observation and imaging of targets at sea. However, how to quickly and accurately extract target information from these massive radar data is still a key problem that needs to be solved in the current field of moving target detection at sea.

[0025] In addition, with the rapid development of artificial intelligence technology, especially the widespread application of deep learning, machine learning and other algorithms, new ideas and methods have been provided for the detection of moving targets at sea. By building a deep neural network model, automatic feature extraction and target recognition of radar data can be achieved, thereby improving detection accuracy and efficiency. However, the application of artificial intelligence technology to the detection of moving targets at sea also faces many challenges, such as difficult data labeling, complex model training, and high consumption of computing resources.

[0026] In response to the above technical problems, the technical concept of the present invention is to integrate data including various types of radars, satellites, etc., and combine the target detection algorithm and target tracking and recognition algorithm based on deep learning to achieve accurate, efficient and real-time detection of dynamic targets at sea.

[0027] Figure 1 A schematic diagram of a flow chart of a method for detecting a moving target at sea provided by an embodiment of the present invention. Figure 1 As shown, the method for detecting moving targets at sea includes:

[0028] Step S101, acquiring multi-source data at sea and fusing the multi-source data, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data.

[0029] Specifically, first, multi-source data at sea are collected, including radar data, satellite data, meteorological data, etc., among which radar data can come from different types of radars, such as SAR radar, phased array radar, etc., to obtain radar images and radar signals of moving targets at sea; satellite data can come from remote sensing satellites to obtain remote sensing images and geographic location information of moving targets at sea; meteorological data can come from meteorological satellites or meteorological stations to obtain meteorological information of the environment in which moving targets at sea are located, such as wind speed, wind direction, wave height, etc.

[0030] Then, the multi-source data is fused. In some embodiments, the fusion of the multi-source data in step S101 includes: using at least one of a weighted average method, a Kalman filter method, and a Bayesian network method to fuse the multi-source data. Specifically, a data fusion algorithm such as a weighted average method, a Kalman filter method, and a Bayesian network method can be used to organically combine data from different sources to improve the reliability and accuracy of the data.

[0031] In some embodiments, after acquiring the offshore multi-source data, the method further includes: preprocessing the multi-source data, wherein the preprocessing includes at least one of the following: data cleaning, data calibration, and data format conversion.

[0032] Specifically, after collecting multi-source data and before fusion, the collected multi-source data is preprocessed, including data cleaning, data calibration, data format conversion, etc., where data cleaning is used to remove noise and outliers in the data and improve data quality, such as removing clutter and interference signals in radar images, removing clouds and shadows in remote sensing images, etc.; data calibration is used to align data from different sources in space and time to ensure data accuracy and consistency; data format conversion is used to convert data in different formats into a unified format for subsequent processing and analysis, such as converting radar images and remote sensing images to the same resolution and color space. Then, the preprocessed multi-source data is fused.

[0033] Step S102: input the fused multi-source data into an adaptive feature extraction network and output a feature vector graph.

[0034] Specifically, an adaptive feature extraction network is constructed to extract features of marine moving targets from fused multi-source data. The adaptive feature extraction network can use a deep neural network model, such as a convolutional neural network (CNN), a recurrent neural network (RNN), etc., to automatically extract high-dimensional features from the data through training and learning.

[0035] Step S103: Detect marine moving targets from the feature vector graph using a deep learning-based target detection algorithm.

[0036] Specifically, a real-time target detection algorithm based on deep learning is constructed to detect moving targets at sea from the feature vector output by the adaptive feature extraction network. Target detection algorithms based on deep learning, such as You Only Look Once (YOLO) and Single Shot MultiBox Detector (SSD), have the advantages of fast detection speed and high accuracy.

[0037] Step S104: Track the marine moving target in real time based on the target tracking algorithm, and identify the category of the marine moving target based on the deep learning classification algorithm.

[0038] Specifically, on the basis of real-time target detection, target tracking and recognition are further realized. Target tracking can adopt tracking methods based on algorithms such as Kalman filtering and particle filtering, and continuously track the target by continuously observing the target's motion trajectory; target recognition can adopt classification algorithms based on deep learning, such as convolutional neural networks (CNN), etc., and accurately recognize the target by classifying the feature vectors of the tracked marine moving targets.

[0039] In some embodiments, step S104 includes: determining the initial motion state of the detected sea moving target; tracking the sea moving target based on the Kalman filter algorithm, and updating the motion state of the sea moving target in real time to obtain the motion trajectory of the sea moving target; identifying the category of the sea moving target based on the convolutional neural network model.

[0040] Specifically, when performing target tracking and recognition, first, the marine moving targets of real-time target detection are initialized, including the category, position, speed and other information of the marine moving targets; then, the Kalman filter is used to track the target and update the target's motion state; at the same time, the convolutional neural network (CNN) is used to identify the target, confirm the category of the target, and then output the tracking and recognition results.

[0041] The method for detecting moving targets at sea provided in this embodiment achieves accurate, efficient and real-time detection of dynamic targets at sea by fusing multi-source data at sea and constructing an adaptive feature extraction network, a real-time target detection algorithm based on deep learning, and a target tracking and recognition algorithm. Compared with traditional methods based on manually set features and rules, the method has higher detection accuracy and can accurately identify different types of moving targets at sea, providing strong technical support for the fields of maritime safety monitoring, maritime rescue, etc.

[0042] Based on the above embodiments, Figure 2 for Figure 1 A detailed flow chart of step S102 in the embodiment shown, the adaptive feature extraction network includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer and an output layer, such as Figure 2 As shown, step S102 includes the following steps:

[0043] Step S1021: receiving the fused multi-source data based on the input layer.

[0044] Step S1022: extracting local features of different scales of the fused multi-source data based on the multiple convolutional layers.

[0045] Step S1023: down-sample the local features of different scales based on the pooling layer.

[0046] Step S1024: Flatten the downsampled features into a one-dimensional vector based on the fully connected layer, and extract global features through a fully connected operation.

[0047] Step S1025: Output the global features based on the output layer to form a corresponding feature vector graph.

[0048] Specifically, an adaptive feature extraction network is constructed, which is based on a convolutional neural network model (CNN) and includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer is used to receive fused multi-source data as input; each of the multiple convolutional layers contains multiple convolutional kernels, each of which is responsible for extracting a specific feature, thereby extracting local features of different scales of data through convolution operations; the pooling layer is used to downsample the output of the convolutional layer to reduce the dimension and amount of calculation of the data, and commonly used pooling operations include maximum pooling and average pooling; the fully connected layer is used to flatten the output of the pooling layer into a one-dimensional vector, and extract global features through a fully connected operation; the output layer is used to output the extracted feature vector for subsequent target detection tasks.

[0049] In the process of adaptive feature extraction, this embodiment first performs preliminary feature extraction on the input data through multiple convolutional layers to obtain a preliminary feature map; then, the preliminary feature map is reduced in dimension and feature selected through the pooling layer to obtain a higher-level feature map; then, the feature map is further processed through the fully connected layer to obtain a feature vector of the target, that is, a feature vector map. It should be noted that at this time, the feature vector can also be classified according to the classifier to obtain the category information of the target, and the step of identifying the category of the marine moving target based on the deep learning classification algorithm in step S104 can be executed at this time.

[0050] In some embodiments, the adaptive feature extraction network is obtained by transfer learning or ensemble learning. Specifically, in order to improve the performance and generalization ability of the adaptive feature extraction network, it is trained by transfer learning, ensemble learning and other strategies. Among them, transfer learning refers to using a model pre-trained on a large data set as the initial model, migrating it to a new task, namely, the marine moving target detection task, for fine-tuning, and using existing knowledge and experience to improve the performance of the model. Ensemble learning can combine the prediction results of multiple models to improve the stability and accuracy of the model.

[0051] On the basis of the aforementioned embodiments, by constructing an adaptive feature extraction network, high-dimensional features in the data can be automatically extracted to achieve accurate detection of moving targets at sea. In addition, the adaptive feature extraction network can make full use of existing knowledge and experience through strategies such as transfer learning and ensemble learning to improve the generalization ability of the model, so that even in the face of new moving targets at sea or complex environmental conditions, it can maintain high detection accuracy and stability.

[0052] Based on the above embodiments, Figure 3 for Figure 1 A detailed flow chart of step S103 in the embodiment shown is as follows: Figure 3 As shown, step S103 includes:

[0053] Step S1031: Generate a series of candidate regions on the feature vector map according to a sliding window or a predefined anchor frame set.

[0054] Step S1032: Detect the marine moving targets contained in each candidate area, and adjust the position and size of each candidate area.

[0055] Step S1033: removing redundant candidate regions on the feature vector map by using a non-maximum suppression algorithm.

[0056] Specifically, after extracting the feature vector map corresponding to the fused multi-source data, real-time target detection can be achieved based on the YOLO algorithm, including the following steps: first, generating candidate regions, that is, generating a series of candidate regions (also called anchor frames, detection frames) on the gridded feature vector map through a sliding window or a predefined anchor frame set, which may contain targets; then performing classification and regression operations on each candidate region, the classification operation determines whether the candidate region contains the target and the category of the target; the regression operation adjusts the position and size of the candidate region to make it match the target more accurately; finally, performing redundancy elimination, that is, removing redundant detection frames through the non-maximum suppression (NMS) algorithm to obtain the final detection result.

[0057] On the basis of the foregoing embodiments, by adopting the YOLO algorithm to optimize the network structure and utilizing technologies such as parallel computing, the computational complexity of the model can be significantly reduced and the detection speed can be improved. Compared with traditional methods based on sliding windows, this optimization improves the detection efficiency and makes real-time detection of dynamic targets at sea possible.

[0058] Figure 4 A schematic diagram of another method for detecting a moving target at sea provided by an embodiment of the present invention is shown below. Figure 4 The embodiment of the present invention is described in detail, which is mainly divided into four stages: data processing, adaptive feature extraction, real-time target detection and target recognition and tracking.

[0059] 1. Data Processing

[0060] (1) Obtaining sensor data: Obtaining multi-source data from different sensors. For example, radar data can come from different types of radars, such as SAR radar, phased array radar, etc., to obtain the radar image and radar signal of the ship; satellite data can come from remote sensing satellites to obtain the remote sensing image and geographic location information of the ship; meteorological data can come from meteorological satellites or meteorological stations to obtain meteorological information of the ship's environment, such as wind speed, wind direction, wave height, etc.

[0061] (2) Data preprocessing: Data cleaning of multi-source data, such as removing clutter and interference signals from radar images, removing clouds and shadows from remote sensing images, etc.; data format conversion, such as converting radar images and remote sensing images to the same resolution and color space.

[0062] (3) Data fusion: The pre-processed multi-source data is fused using the weighted average method. A weight coefficient can be assigned to each data source based on the reliability and importance of the data. Since SAR radar data is not affected by conditions such as lighting and weather, its data weight is higher than that of remote sensing data and meteorological data. Then the data from different data sources are weighted averaged to obtain the fused data.

[0063] 2. Adaptive Feature Extraction

[0064] (1) Input layer: receives fused multi-source data as input.

[0065] (2) Convolutional layer: Each convolutional layer contains multiple convolution kernels. Each convolution kernel is responsible for extracting a specific feature. The convolution operation is used to extract local features of the data at different scales.

[0066] (3) Pooling layer: Downsample the output of the convolutional layer to reduce the dimension of the data and the amount of computation. Commonly used pooling operations include maximum pooling and average pooling.

[0067] (4) Fully connected layer: Flatten the output of the pooling layer into a one-dimensional vector and extract global features through a fully connected operation.

[0068] (5) Output layer: Output the extracted feature vector to form a feature vector graph for subsequent target detection tasks.

[0069] 3. Real-time object detection

[0070] (1) Candidate region generation: Generate a series of candidate regions on the gridded feature vector map, which may contain objects. The generation of candidate regions can be achieved through a sliding window or a set of predefined anchor boxes.

[0071] (2) Classification and regression: Classification and regression operations are performed on each candidate region. The classification operation determines whether the candidate region contains the target and the category of the target; the regression operation adjusts the position and size of the candidate region to make it match the target more accurately.

[0072] (3) Redundancy elimination: Redundant detection boxes are removed through the non-maximum suppression (NMS) algorithm to obtain the final detection result.

[0073] 4. Target recognition and tracking

[0074] (1) Detection result initialization: Initialize the results of real-time target detection, including target category, location, speed and other information.

[0075] (2) Kalman filter: Use Kalman filter to track the target and update the target's motion state.

[0076] (3) Feature vector classification: Use convolutional neural network (CNN) to identify the target, confirm the target category, and then output the tracking and recognition results.

[0077] In summary, this embodiment has the following beneficial effects. The first is to improve the detection accuracy: by constructing an adaptive feature extraction network and a real-time target detection algorithm, it is possible to automatically extract high-dimensional features in the data and realize accurate detection of moving targets at sea. The second is to improve the detection efficiency: by optimizing the network structure through the YOLO algorithm and adopting technologies such as parallel computing, the computational complexity of the model can be significantly reduced and the detection speed can be improved. The third is to enhance the generalization ability: through strategies such as transfer learning and ensemble learning, it is possible to make full use of existing knowledge and experience, improve the generalization ability of the adaptive feature extraction network model, so that when facing new moving targets at sea or complex environmental conditions, it is possible to maintain high detection accuracy and stability.

[0078] Figure 5 A schematic diagram of the structure of a marine moving target detection device provided by an embodiment of the present invention is shown in FIG. Figure 5As shown, the marine moving target detection device comprises:

[0079] The data fusion module 501 is used to obtain and fuse multi-source data at sea, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data;

[0080] A feature extraction module 502 is used to input the fused multi-source data into an adaptive feature extraction network and output a feature vector graph;

[0081] A target detection module 503, configured to detect a moving target at sea from the feature vector map based on a deep learning target detection algorithm;

[0082] The tracking and identification module 504 is used to track the marine moving target in real time based on a target tracking algorithm, and to identify the category of the marine moving target based on a deep learning classification algorithm.

[0083] In some embodiments, the adaptive feature extraction network includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer and an output layer; the feature extraction module 502 is specifically used to:

[0084] Receiving the fused multi-source data based on the input layer;

[0085] Extracting local features of different scales of the fused multi-source data based on the multiple convolutional layers;

[0086] Downsampling the local features of different scales based on the pooling layer;

[0087] Flattening the downsampled features into a one-dimensional vector based on the fully connected layer, and extracting global features through a fully connected operation;

[0088] The global features are output based on the output layer to form a corresponding feature vector graph.

[0089] In some embodiments, the adaptive feature extraction network is obtained by transfer learning or ensemble learning.

[0090] In some embodiments, the target detection module 503 is specifically used to:

[0091] Generate a series of candidate regions on the feature vector map according to a sliding window or a predefined anchor box set;

[0092] Detect the marine moving targets contained in each candidate area, and adjust the position and size of each candidate area;

[0093] The redundant candidate regions on the feature vector map are removed by a non-maximum suppression algorithm.

[0094] In some embodiments, the tracking and identification module 504 is specifically used to:

[0095] Determine the initial motion state of the detected marine moving target;

[0096] Track the moving targets at sea based on the Kalman filter algorithm and update the moving state of the moving targets at sea in real time to obtain the moving trajectory of the moving targets at sea;

[0097] The category of the marine moving target is identified based on a convolutional neural network model.

[0098] In some embodiments, the data fusion module 501 is further used to:

[0099] The multi-source data is preprocessed, and the preprocessing includes at least one of the following: data cleaning, data calibration, and data format conversion.

[0100] In some embodiments, the data fusion module 501 is specifically used to:

[0101] At least one of weighted average method, Kalman filter method and Bayesian network method is used to fuse multi-source data.

[0102] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and corresponding beneficial effects of the marine moving target detection device described above can refer to the corresponding process in the aforementioned method example and will not be repeated here.

[0103] like Figure 6 As shown, an embodiment of the present invention provides an electronic device, including a processor 601, a communication interface 602, a memory 603 and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604.

[0104] Memory 603, used for storing computer programs;

[0105] In one embodiment of the present invention, the processor 601 is used to implement the steps of the marine moving target detection method provided by any one of the aforementioned method embodiments when executing the program stored in the memory 603.

[0106] The implementation principle and technical effect of the electronic device provided by the embodiment of the present invention are similar to those of the above embodiment and will not be described in detail here.

[0107] The above-mentioned memory 603 can be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk or a ROM. The memory 603 has a storage space for program codes for executing any method steps in the above-mentioned method. For example, the storage space for program codes may include various program codes for implementing various steps in the above method respectively. These program codes can be read from or written into one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks. Such computer program products are usually portable or fixed storage units. The storage unit may have a storage segment or storage space arranged similarly to the memory 603 in the above-mentioned electronic device. The program code can be compressed, for example, in an appropriate form. Generally, the storage unit includes a program for executing the method steps according to an embodiment of the present invention, that is, a code that can be read by a processor such as 601, which, when run by an electronic device, causes the electronic device to execute various steps in the method described above.

[0108] The embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for detecting moving targets at sea are implemented.

[0109] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0110] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0111] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0112] The above are preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for detecting moving targets at sea, characterized in that: include: Acquire multi-source data at sea and fuse the multi-source data, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data; Input the fused multi-source data into the adaptive feature extraction network and output the feature vector graph; A target detection algorithm based on deep learning detects moving targets at sea from the feature vector graph; The marine moving target is tracked in real time based on a target tracking algorithm, and the category of the marine moving target is identified based on a deep learning classification algorithm.

2. The method according to claim 1, characterized in that The adaptive feature extraction network includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer and an output layer; the fused multi-source data is input into the adaptive feature extraction network, and a feature vector graph is output, including: Receiving the fused multi-source data based on the input layer; Extracting local features of different scales of the fused multi-source data based on the multiple convolutional layers; Downsampling the local features of different scales based on the pooling layer; Flattening the downsampled features into a one-dimensional vector based on the fully connected layer, and extracting global features through a fully connected operation; The global features are output based on the output layer to form a corresponding feature vector graph.

3. The method according to claim 2, characterized in that The adaptive feature extraction network is obtained by using transfer learning or ensemble learning.

4. The method according to any one of claims 1 to 3, characterized in that: The target detection algorithm based on deep learning is a YOLO algorithm, and the target detection algorithm based on deep learning detects a moving target at sea from the feature vector map, including: Generate a series of candidate regions on the feature vector map according to a sliding window or a predefined anchor box set; Detect the marine moving targets contained in each candidate area, and adjust the position and size of each candidate area; The redundant candidate regions on the feature vector map are removed by a non-maximum suppression algorithm.

5. The method according to any one of claims 1 to 3, characterized in that: The real-time tracking of the marine moving target based on the target tracking algorithm and the identification of the category of the marine moving target based on the deep learning classification algorithm include: Determine the initial motion state of the detected marine moving target; Track the moving targets at sea based on the Kalman filter algorithm and update the moving state of the moving targets at sea in real time to obtain the moving trajectory of the moving targets at sea; The category of the marine moving target is identified based on a convolutional neural network model.

6. The method according to any one of claims 1 to 3, characterized in that: After acquiring the offshore multi-source data, the method further includes: The multi-source data is preprocessed, and the preprocessing includes at least one of the following: data cleaning, data calibration, and data format conversion.

7. The method according to any one of claims 1 to 3, characterized in that: The fusing of multi-source data includes: At least one of weighted average method, Kalman filter method and Bayesian network method is used to fuse multi-source data.

8. A device for detecting moving targets at sea, characterized in that: include: A data fusion module, used to obtain and fuse multi-source data at sea, wherein the multi-source data includes at least one of the following: radar data, satellite data, and meteorological data; A feature extraction module is used to input the fused multi-source data into an adaptive feature extraction network and output a feature vector graph; A target detection module, used for detecting marine moving targets from the feature vector image based on a deep learning target detection algorithm; The tracking and identification module is used to track the marine moving target in real time based on a target tracking algorithm, and to identify the category of the marine moving target based on a deep learning classification algorithm.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the steps of the method for detecting moving targets at sea as described in any one of claims 1 to 7 when executing the program stored in the memory.

10. A 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 for detecting moving targets at sea as described in any one of claims 1 to 7 are implemented.

Images