Automatic real-time sea ice condition identification method based on shipborne acceleration measurement

Abstract

The invention discloses a sea ice condition automatic real-time identification method based on shipborne acceleration measurement, the identification method is realized through an identification system, and the identification system comprises three modules: a signal preprocessing module, a time-frequency feature extraction module and a sea ice condition reasoning module; the method comprises the following steps: 1, converting acceleration data into a time-frequency image through time-frequency analysis, and adopting short-time Fourier transform and Hough transform; 2, time-frequency image feature recognition is carried out to extract key time-frequency features, and a DETR model can be adopted; 3, establishing a mapping relation between acceleration signal time-frequency characteristics and sea ice conditions, wherein machine learning methods such as a support vector machine and the like can be adopted; compared with an existing direct measurement technology, the sea ice condition is inversed in real time by adopting a reverse engineering method, the identification accuracy is improved, the cost is reduced, and assistance can be provided for polar region safety operation and navigation decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of sea ice recognition, and more specifically, to an automatic real-time recognition method for sea ice conditions based on shipborne acceleration measurement. Background Art

[0002] Due to the diversity of polar maritime risks and the complexity of their interaction mechanisms, real-time risk status monitoring of polar maritime navigation and operations requires more attention. Obviously, effective and real-time detection measures are essential to minimize operational risks and reduce the probability of maritime accidents, which requires a flexible method for automatic real-time identification of sea ice conditions.

[0003] Current technology makes it difficult to measure polar environmental conditions and the stress responses generated by ship contact with sea ice, but ice conditions and ice loads are important parameters that determine the safety and efficiency of polar ships; therefore, it is crucial to accurately extract ship-ice contact information from polar ship accelerations and identify sea ice conditions through acceleration data.

[0004] Existing research mainly focuses on the acceleration frequency domain analysis of specific ship-ice contact events, but has not conducted in-depth research and quantitative analysis of ship-ice contact; given the acceleration measurement signal, there is still a lack of effective methods to automatically and quickly identify sea ice conditions, which makes it difficult for icebreakers to identify current ice conditions in real time, quickly and accurately during polar operations. Summary of the invention

[0005] The present invention provides a method for automatic real-time identification of sea ice conditions based on ship-borne acceleration measurement, so as to solve the problem that in the prior art, there is still a lack of an effective method for automatically and quickly identifying sea ice conditions under a given acceleration measurement signal, which makes it difficult for icebreakers to identify current ice conditions in real time, quickly and accurately during polar operations.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] An automatic real-time sea ice condition recognition method based on shipborne acceleration measurement comprises the following steps: the recognition method is implemented by a recognition system, the recognition system comprises three modules: a signal preprocessing module, a time-frequency feature extraction module and a sea ice condition reasoning module; the specific recognition steps are as follows:

[0008] Step 1: Input the original data into the signal preprocessing module and convert it into a time-frequency diagram through the time-frequency analysis method;

[0009] Step 2: Automatically identify ship-ice contact events and extract acceleration time-frequency features in the time-frequency feature extraction module;

[0010] Step 3: Establish the mapping relationship between acceleration time-frequency characteristics and sea ice conditions.

[0011] Preferably, the step 1 comprises the following operations:

[0012] 1. Collect raw data and perform noise reduction to extract the ship-ice contact signal from the acceleration data;

[0013] 2. Perform time-frequency analysis on the acceleration signal after noise reduction to generate a time-frequency image;

[0014] 3. The time-frequency image is processed for subsequent recognition.

[0015] Preferably, the step 2 comprises the following operations:

[0016] 1. Manually mark the ship-ice contact time on the acceleration time-frequency image and generate a marking file;

[0017] 2. Use image recognition algorithms (such as DETR) to train the image dataset to obtain an automatic ship-ice contact recognition model;

[0018] 3. Apply the ship-ice contact automatic identification model to process the image data set, identify the ship-ice contact, and extract the typical acceleration time-frequency characteristics such as ship-ice contact frequency, pulse width, and acceleration peak.

[0019] Preferably, the step three comprises the following operations:

[0020] 1. Match the acceleration time-frequency characteristics with sea ice characteristics (including thickness, density, and iceberg size) to establish a data set;

[0021] 2. Apply machine learning methods such as support vector machines to establish a mapping model between acceleration time-frequency characteristics and sea ice characteristics;

[0022] 3. The established model is connected with the signal preprocessing module in step one and the DETR model trained in step two to form an automatic sea ice condition recognition system.

[0023] Preferably, the noise reduction process comprises performing low-pass filtering on the acceleration to obtain acceleration data containing only ship-ice contact; and visualizing the extracted acceleration data using short-time Fourier transform to obtain a time-frequency image of the ship-ice contact event.

[0024] Preferably, the ship-ice contact automatic identification model takes the acceleration time-frequency diagram as input and the ship-ice contact event as the identification result.

[0025] Preferably, the acceleration time-frequency characteristics include three items: ship-ice contact frequency, acceleration response pulse width, and acceleration amplitude.

[0026] Preferably, the sea ice conditions include three items: sea ice density, sea ice growth stage and sea ice morphology.

[0027] The principle and beneficial effects of this technical solution:

[0028] (1) The present invention uses enhanced signal decomposition and data visualization modules to extract ship-ice acceleration features from the original acceleration data, and extracts acceleration data containing only ship-ice contact; uses STFT and HF to visualize the ship-ice contact acceleration data from (ship-ice contact events); constructs a ship-ice contact event data set, and proposes a targeted ICE-DETR model based on the characteristics of the ship-ice contact event image, including the following three improvements: adding a CA attention mechanism, using the FDPN feature diffusion focus pyramid structure to replace the neck network, and implanting a dimension-aware selective integration module (DASI) in the neck network; the present invention solves the interference of noise on acceleration data during the detection process through empirical mode decomposition, absolute median difference, short-time Fourier transform and Hough transform, and realizes accurate extraction of ship-ice acceleration features. The present invention solves the problem that the density of ship-ice contact events is too high to be accurately identified through the CA attention mechanism, the FDPN feature diffusion focus pyramid structure and the DASI dimension-aware selective integration module, and realizes accurate extraction of ship-ice contact event features.

[0029] (2) A deep learning method is used to train the neural network to obtain a ship-ice contact event recognition model. The trained model is verified for confidence using evaluation indicators such as accuracy, recall, and average precision mean, and the optimal ship-ice contact event detection model is finally obtained; the test set is input into the target detection model to obtain the recognition result, and each ship-ice contact event is extracted as the acceleration feature n, the pulse width of the ship-ice contact event is extracted as the acceleration feature w, and the maximum acceleration amplitude is extracted from the time-amplitude curve corresponding to the ship-ice contact event as A. The present invention solves the problem that model parameters cannot be quantified through evaluation indicators such as accuracy, recall, and average precision mean, and realizes the visualization of the accuracy of the ICE-DETR model. The present invention solves the feature extraction of ship-ice contact events through ICE-DETR and realizes the quantification of ship-ice contact event features.

[0030] (3) The acceleration features (n, A, w) obtained by constructing the first two modules are artificially matched to the sea ice characteristics (Ct, Sa, Fa) to construct a sea ice data set; the GAPSO-SVM automatic ice inference model is proposed, and the intelligent optimization algorithm (GAPSO) is used to optimize the support vector machine model to obtain the best parameters (C, g); two single-mode functions and two multi-modal functions are selected to solve the minimum value, and the optimization performance of GAPSO is verified; the performance of GA, PSO and GAPSO is analyzed using the optimal value; the SVM is trained by machine methods to obtain an automatic ice inference model. Therefore, researchers can judge the three major features to obtain a rough ice condition and obtain more accurate research results when studying unknown icebreakers; the acceleration data can automatically infer the three parameters of ice conditions (Ct, Sa, Fa) through the automatic ice inference model. Through these parameters, researchers can roughly understand the sea ice environment experienced by icebreakers, thereby assisting polar engineering research with more detailed environmental information. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is the algorithm flow chart of the present invention;

[0032] Figure 2 A schematic diagram of the noise reduction operation of the enhanced signal decomposition and data visualization module in the embodiment;

[0033] Figure 3 A schematic diagram of the visualization operation of the enhanced signal decomposition and data visualization module in the embodiment;

[0034] Figure 4 It is a structural diagram of the CA attention mechanism of the time-frequency feature extraction module based on ICE-DETR in the embodiment;

[0035] Figure 5 Schematic diagram of the FDPN feature focused diffusion pyramid structure of the time-frequency feature extraction module based on ICE-DETR in the embodiment;

[0036] Figure 6 It is a schematic diagram of the activation function structure of the DASI dimension perception selective integration module based on the ICE-DETR time-frequency feature extraction module in the embodiment;

[0037] Figure 7 Schematic diagram of the recognition result of the time-frequency feature extraction module based on ICE-DETR in the embodiment;

[0038] Figure 8 Schematic diagram of the GAPSO-SVM structure of the ice condition parameter reasoning module in the embodiment;

[0039] Fig. 9 Graph showing the objective function of the ice condition parameter inference module in the embodiment. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments:

[0041] Example:

[0042] S1: input the raw data into the enhanced signal decomposition and data visualization module;

[0043] When analyzing sea ice conditions, the image should contain as little noise and interference as possible. This process needs to be achieved through a variety of means to ensure that the extracted sea ice features are accurate and representative.

[0044] The method for extracting ship-ice acceleration data includes: collecting original acceleration data, and performing noise reduction by using empirical mode decomposition (EMD) to obtain IMF1 containing the maximum feature; and removing discrete data points in IMF1 by using median absolute deviation (MAD) to obtain acceleration data containing only ship-ice contact.

[0045] Since ice condition datasets need to extract features from different dimensions, there are few high-quality features of single time-amplitude that can be used for training. In this case, using visualization technology can provide more features for analysis; Figure 3 As shown in the figure, STFT is used to convert the acceleration data from a time-amplitude image to a time-frequency image. Compared with the time-frequency function drawn without EMDMAD, it is found that this method can effectively suppress noise; the ship-ice contact characteristics are extracted and the main components of the original signal are well preserved without losing information. Therefore, this method can effectively extract the ship-ice contact acceleration data characteristics and has good time resolution; in order to better learn the pictures generated by EMDMAD-STFT for ICE-DETR, the line segment detection algorithm based on Hough transform is used to process the pictures, in which the impact pulses will be extracted as line features, and these features can be detected by HF due to the color change.

[0046] S2: Input the ship-ice contact time into the time-frequency feature extraction module based on ICE-DETR

[0047] The labelimg software was used to annotate images of ship-ice contact events and generate the corresponding xml annotation file; the annotation file included the annotation box, the location information of the annotation box, and the true category information of the content annotated by the annotation box; at the same time, the MixUp technology was used to enhance the original dataset to obtain a dataset consisting of 8000 ship-ice contact events; the dataset was split into 80% training set and 20% test set.

[0048] The ICE-DETR model is proposed, including the following three improvements: adding CA attention mechanism, such as Figure 4 shown.

[0049] Use FDPN feature diffusion focused pyramid structure to replace the neck network, such as Figure 5 shown.

[0050] The Dimensionality Aware Selective Integration (DASI) module is implanted in the neck network, such as Figure 6 shown.

[0051] The CA module enables the model to better understand the importance of different positions in the image by weighting the feature map in spatial coordinates. The DASI module adaptively selects and fuses features and dynamically adjusts the feature fusion strategy to improve the detection accuracy and robustness of dense small targets; FDPN enhances the performance of target detection and classification through feature focusing and diffusion mechanisms, enabling the model to achieve efficient and accurate target detection in complex scenes.

[0052] The dataset was input into ICE-DETR for training, thus obtaining a ship-ice contact event recognition model. The modified modules were combined and compared to determine the practical effectiveness of the proposed ICE-DETR.

[0053]

[0054] Using the trained ICE-DETR model, new images of ship-ice contact events are identified, such as Figure 7 As shown;

[0055] Each ship-ice contact event is extracted as the acceleration feature n, the pulse width of the ship-ice contact event is extracted as the acceleration feature w, and the maximum acceleration amplitude is extracted as A from the time-amplitude curve corresponding to the ship-ice contact event.

[0056] Step S3, establishing a connection between the obtained acceleration characteristics (n, w, A) and the sea ice conditions (Ct, Sa and Fa) and inputting them into the ice condition parameter inference module;

[0057] The (C,g) in SVM is optimized using an intelligent optimization algorithm that combines genetic algorithm and particle swarm algorithm.

[0058] like Fig. 9 As shown in Figure 2, two single-mode functions and two multi-mode functions are selected to solve the minimum value and verify the optimization performance of GAPSO. The obtained sea ice condition data set is input into GAPSO-SVM and the SVM is trained by machine method to obtain the automatic ice condition reasoning model, as shown in Figure 2. Figure 8 As shown, researchers can study the unknown icebreaker by judging the three major characteristics, obtain the approximate ice conditions and obtain more accurate research results.

[0059] The specific usage and function of this embodiment are as follows:

[0060] like Figures 1 to 9 As shown, the present invention provides a method for automatic real-time recognition of sea ice conditions based on shipborne acceleration measurement. The recognition method is implemented by a recognition system, which includes three modules: a signal preprocessing module, a time-frequency feature extraction module, and a sea ice condition reasoning module. The specific recognition steps are as follows:

[0061] Step 1: Input the original data into the signal preprocessing module and convert it into a time-frequency diagram through the time-frequency analysis method. ;

[0062] Step 2: Automatically identify ship-ice contact events and extract acceleration time-frequency features in the time-frequency feature extraction module;

[0063] Step 3: Establish the mapping relationship between acceleration time-frequency characteristics and sea ice conditions.

[0064] like Figure 1 As shown, step one includes the following operations:

[0065] 1. Collect raw data and perform noise reduction to extract the ship-ice contact signal from the acceleration data;

[0066] 2. Perform time-frequency analysis on the acceleration signal after noise reduction to generate a time-frequency image;

[0067] 3. The time-frequency image is processed for subsequent recognition.

[0068] like Figure 1 As shown, step 2 includes the following operations:

[0069] 1. Manually mark the ship-ice contact time on the acceleration time-frequency image, generate an XML annotation file, and use MixUp technology to enhance the data set to obtain the image data set;

[0070] 2. The CA attention mechanism is added to the backbone network of the ICE-DETR model, the original network is replaced by the FDPN neck network with feature diffusion focusing, and the dimension-aware selective integration module (DASI) is implanted in the FDPN. The image dataset is trained to obtain the ship-ice contact automatic recognition model.

[0071] 3. Apply the ship-ice contact automatic identification model to the image dataset.

[0072] like Figure 1 As shown, step three includes the following operations:

[0073] 1. Match the acceleration time-frequency characteristics with sea ice characteristics (including thickness, density, and iceberg size) to establish a data set;

[0074] 2. Apply machine learning methods such as support vector machines to establish a mapping model between acceleration time-frequency characteristics and sea ice characteristics;

[0075] 3. The established model is connected with the signal preprocessing module in step one and the DETR model trained in step two to form an automatic sea ice condition recognition system.

[0076] like Figure 3 As shown in FIG, the noise reduction process includes low-pass filtering the acceleration to obtain acceleration data containing only ship-ice contact; the extracted acceleration data is visualized using short-time Fourier transform to obtain a time-frequency image of the ship-ice contact event.

[0077] like Figure 4 As shown in Figure 1, the ship-ice contact automatic identification model takes the acceleration time-frequency diagram as input and the ship-ice contact event as the identification result.

[0078] like Figure 5 As shown in the figure, the time-frequency characteristics of acceleration include ship-ice contact frequency, acceleration response pulse width and acceleration amplitude.

[0079] like Figure 5 As shown in the figure, sea ice conditions include sea ice density, sea ice growth stage and sea ice morphology.

[0080] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement, characterized in that: The following steps are involved: The identification method is implemented by an identification system, which includes three modules: a signal preprocessing module, a time-frequency feature extraction module, and a sea ice condition reasoning module; the specific identification steps are as follows: Step 1: Input the original data into the signal preprocessing module and convert it into a time-frequency diagram through the time-frequency analysis method; Step 2: Automatically identify ship-ice contact events and extract acceleration time-frequency features in the time-frequency feature extraction module; Step 3: Establish the mapping relationship between acceleration time-frequency characteristics and sea ice conditions.

2. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 1, characterized in that: The step 1 comprises the following operations:

1. Collect raw data and perform noise reduction to extract the ship-ice contact signal from the acceleration data; 2. Perform time-frequency analysis on the acceleration signal after noise reduction to generate a time-frequency image; 3. The time-frequency image is processed for subsequent recognition.

3. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 2, characterized in that: The step 2 includes the following operations:

1. Manually mark the ship-ice contact time on the acceleration time-frequency image and generate a marking file; 2. Use image recognition algorithms (such as DETR) to train the image dataset to obtain an automatic ship-ice contact recognition model; 3. Apply the ship-ice contact automatic identification model to process the image data set, identify the ship-ice contact, and extract the typical acceleration time-frequency characteristics such as ship-ice contact frequency, pulse width, and acceleration peak.

4. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 3 is characterized in that: The step three includes the following operations:

1. Match the acceleration time-frequency characteristics with sea ice characteristics (including thickness, density, and iceberg size) to establish a data set; 2. Apply machine learning methods such as support vector machines to establish a mapping model between acceleration time-frequency characteristics and sea ice characteristics; 3. The established model is connected with the signal preprocessing module in step one and the DETR model trained in step two to form an automatic sea ice condition recognition system.

5. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 4 is characterized in that: The noise reduction process includes low-pass filtering the acceleration to obtain only ship-ice contact acceleration data; and visualizing the extracted acceleration data using short-time Fourier transform to obtain a time-frequency image of the ship-ice contact event.

6. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 5, characterized in that: The ship-ice contact automatic identification model takes the acceleration time-frequency diagram as input and the ship-ice contact event as the identification result.

7. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 6 is characterized in that: The acceleration time-frequency characteristics include three items: ship-ice contact frequency, acceleration response pulse width, and acceleration amplitude.

8. The method for automatic real-time identification of sea ice conditions based on shipborne acceleration measurement according to claim 7, characterized in that: The sea ice conditions include sea ice density, sea ice growth stage and sea ice morphology.

Images