Sea ice semi-supervised multi-task inversion method driven by anomaly detection

Abstract

The invention discloses a sea ice semi-supervised multi-task inversion method driven by anomaly detection, and belongs to the technical field of satellite remote sensing and deep learning. The problems that an existing method excessively depends on labels, is redundant in structure and cannot effectively invert sea ice under the condition of sample imbalance are solved. The method comprises the following steps: sending DDM data of TDS-1 into a data preprocessing module to obtain standard data required by a multi-task network; a sea ice detection module in a multi-task network structure is used for carrying out deep feature extraction on input standard DDM data, a potential deep learning guide mechanism is used for mining and sharing detailed information and spatial context, and sea ice data and seawater data are accurately distinguished through the module, so that a sea ice detection task is completed; and inputting the extracted DDM high-dimensional features and module output results into a sea ice concentration inversion module through an inter-module data feature sharing mechanism, and inferring the concentration of sea ice by using the learned features to complete sea ice concentration inversion. According to the invention, the precision and efficiency of sea ice inversion are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of satellite remote sensing and deep learning technology, and specifically to an anomaly detection-driven sea ice semi-supervised multi-task inversion method. Background Art

[0002] The formation, melting and movement of sea ice have a significant impact on the stability of the Earth's climate system, the balance of marine ecology and human socio-economic activities. Therefore, accurate detection and analysis of sea ice has extremely high professional value and practical significance.

[0003] At present, the use of satellite remote sensing for sea ice detection is mainly based on visible light remote sensing and microwave remote sensing. The images obtained by the visible light remote sensing method are affected by weather conditions and cannot be observed in real time around the clock. Microwave remote sensing is not affected by weather, but its spatial resolution is poor. To improve the spatial resolution, the antenna size needs to be increased, which is a challenge under the current space technology conditions.

[0004] With the development of global navigation satellite systems, global navigation satellite system reflectometry (GNSS-R) technology has become a popular means of global sea ice detection as an emerging method. Existing GNSS-R-based sea ice remote sensing methods mainly include methods based on Delay Doppler Maps (DDM) observables, methods based on scattering coefficient retrieval, and methods based on machine learning. Although both methods based on DDM observables and methods based on scattering coefficient retrieval can achieve sea ice detection tasks, these two methods have certain limitations. They cannot directly learn information from DDM data and need to rely on designed rules or feature extraction for sea ice detection. Machine learning can automatically learn the relationship between images and required parameters from a large amount of remote sensing image data without specific rules. On the other hand, both of the above methods only complete the task of sea ice detection and do not perform the task of sea ice concentration inversion. Based on this, the method of using machine learning for sea ice remote sensing has gradually been studied and explored by many scholars. It not only improves the performance of sea ice detection, but most machine learning algorithms also design similarity networks to complete the task of sea ice concentration inversion.

[0005] Although these machine learning-based methods have made significant progress in advancing sea ice inversion technology, they are mainly supervised learning algorithms, which means that they are extremely dependent on accurate and high-quality labeled data. In practice, obtaining large-scale, high-quality labeled datasets is not only challenging but also costly. More importantly, these labeled data are likely to be affected by various errors, thus affecting the accuracy of the final results. Therefore, although these methods are extremely valuable in theory, their flexibility and applicability are still limited in practical applications. In addition, for specific situations in actual application scenarios, although we can obtain a large amount of sea ice DDM data, the number of seawater DDM data far exceeds that of sea ice DDM data on a global scale. This phenomenon leads to the problem of sample imbalance when using large-scale datasets for model training. How to overcome the two major challenges of over-reliance on labeled data and sample imbalance is of great significance to this field. Summary of the invention

[0006] The purpose of the present invention is to propose an anomaly detection driven sea ice semi-supervised multi-task inversion method to solve the problems of over-reliance on labels, structural redundancy and inability to effectively invert sea ice in the case of sample imbalance in the prior art.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] An anomaly detection driven sea ice semi-supervised multi-task inversion method includes the following steps:

[0009] S1. Obtain the DDM data of TDS-1, input the obtained DDM data into the data preprocessing module, and obtain standard DDM data that can be used for sea ice multi-task inversion through data screening, background noise subtraction and normalization;

[0010] S2. Design a multi-task grid structure, wherein the multi-task grid structure includes a sea ice detection module and a sea ice concentration inversion module; use the seawater DDM data in the standard DDM data obtained in S1 as input and input it into the sea ice detection module in the multi-task grid structure; perform deep feature extraction on the input seawater DDM data, accurately distinguish sea ice data from seawater data, and complete the sea ice detection task;

[0011] S3. The high-dimensional features of the seawater DDM data extracted by the sea ice detection module and the module output results are input into the sea ice concentration inversion module through a potential deep learning guidance mechanism. The learned features are used to infer the sea ice concentration and complete the sea ice concentration inversion task.

[0012] Preferably, the screening conditions for the data screening work in S1 are:

[0013] ① Data with latitudes greater than 55 degrees north and south;

[0014] ②Data with satellite incidence angle less than 35 degrees;

[0015] ③Peak signal-to-noise ratio (SNR) is greater than 0dB;

[0016] ④The antenna gain is greater than 3dB;

[0017] ⑤The sea ice concentration is 0.15 as the dividing line.

[0018] Preferably, the background noise subtraction in S1 specifically includes the following contents:

[0019] The noise floor is calculated as:

[0020]

[0021] Based on formula (1), the formula for subtracting the background noise of each DDM data is:

[0022] DDM remove_noise =DDM(τ,f)-Noise (2)

[0023] Wherein, Noise represents the noise value; τ1 and τ2 represent the pixel limits in the delay frequency band; f1 and f2 represent the pixel limits in the Doppler frequency band; X represents the number of all pixels in the noise box; DDM(τ, f) represents the value of a specific position of the original DDM data in the delay and Doppler frequency directions.

[0024] Preferably, the normalization operation in S1 specifically refers to normalizing each DDM according to its peak power, and the normalization calculation formula is:

[0025]

[0026] Where DDM(τ,f) represents the value of a specific position of the original DDM data in the direction of time delay and Doppler frequency. maxnosie The DDM value represents the peak power after subtracting the noise floor.

[0027] Preferably, the sea ice detection module is used for anomaly detection, and comprises a generator G(x) and a discriminator D(x);

[0028] The DDM image reflected by seawater is used as a normal sample to be input into the sea ice detection module for training, and the sea ice DDM image is used as an abnormal sample. During the training process, the generator G(x) and the discriminator D(x) of the sea ice detection module compete with each other, wherein the generator G(x) is used to generate a reconstructed DDM image that is closer to the real DDM image x. So that the discriminator D(x) cannot distinguish true from false. The discriminator D(x) is used to distinguish between x and

[0029] Preferably, the generator G(x) is composed of an encoder and a decoder; the encoder maps the input image to a latent vector in the latent space by downsampling, and extracts an abstract feature representation; each layer of the encoder includes a convolutional layer, a BatchNorm layer, and a LeakyReLU activation function, reads its input data into the network, and outputs a latent vector in the latent space;

[0030] The decoder maps the latent vector of the latent space back to the original image space by upsampling, that is, reconstructing the abstract feature representation into an image again, and finally generating a realistic image;

[0031] The decoder (Decoder) adopts the structure of transposed convolution and BatchNorm layer, which corresponds to the structure of the encoder (Encoder) one by one; the transposed convolution is used to upsample the input latent space vector to increase the size of the feature map; the decoder (Decoder) uses the Tanh activation function in the output layer, and the other layers all use the ReLU activation function.

[0032] Preferably, the D(x) is composed of a convolutional layer, a BatchNorm layer, and a LeakyReLU activation function; the discriminator D(x) is used to judge the authenticity of the reconstructed image generated by the generator G(x) and the input real image, and classify them. The goal of the discriminator D(x) is to get a high score on the real image and a low score on the reconstructed image;

[0033] The discriminator D(x) also acts as a feature extractor, taking the intermediate layer output of the discriminator as features, and then calculating the feature matching loss between the generated image and the real image.

[0034] Preferably, the potential deep learning guidance mechanism described in S3 is implemented by a jump connection, which enables information to be transmitted in the network by jumping, connecting the feature map of the last layer in the encoder (Encoder) with the feature map of the corresponding layer in the decoder (Decoder) to transmit more detailed information and spatial context, and restore richer detail information in the decoder stage.

[0035] Preferably, the sea ice concentration inversion module in S3 includes a discriminant matching unit and an LSTM (Long Short-Term Memory) network;

[0036] The discrimination and matching unit judges and matches the output result of the sea ice detection module with the original input image, and the data that is sea ice data and is judged as abnormal by the sea ice detection module is used as the input of the LSTM network; the LSTM network receives the output of the matching discrimination unit, and obtains the high-dimensional feature representation of the corresponding DDM data through the inter-module data feature sharing mechanism, thereby completing the sea ice concentration inversion task;

[0037] The LSTM network includes two stacked LSTM layers, each LSTM layer is regarded as a layer that performs sequence modeling on the input data, and the stacking of multiple layers allows the model to capture more complex sequence dependencies; each LSTM layer includes 128 hidden units, each hidden unit includes a memory unit; LSTM includes three key gating mechanisms, namely forget gate, input gate and output gate, to control the flow and update of cell states.

[0038] Preferably, the forget gate is used to determine how much of the unit state at the previous moment is retained to the unit state at the current moment, and the specific function is expressed as:

[0039] f t =σ(W f ·[h t-1 ,x t ]+b f ) (4)

[0040] Among them, f t represents the output value of the forget gate; σ represents the sigmoid layer; w f represents the weight matrix of the forget gate; x t Represents the input of the current time step; h t-1 represents the hidden state of the previous time step; [h t-1 ,x t ] represents the vector formed by connecting the hidden state of the previous time step and the input of the current time step; b f represents the bias vector;

[0041] The input gate determines which information should enter the cell state through the learned weights. It consists of two parts: the sigmoid layer and the tanh layer. The sigmoid layer determines what value to update, and the tanh layer updates the information to be updated into the cell state. The specific function is expressed as follows:

[0042] i t =σ(W i ·[h t -1,x t ])+bi (5)

[0043]

[0044] Among them, i t Represents the output value of the input gate; represents the candidate value of the input gate; W i , W f represents the weight matrix; b i 、b c represents the bias vector;

[0045] New cell state C t The memory in is composed of the old memory left after the memory cells of the previous moment forget the useless memory and the useful information filtered out by the input gate as new memory. The function of the new cell state is expressed as follows:

[0046]

[0047] The output gate is used to determine which information should be output to the next layer or as the final output. The specific function is as follows:

[0048] o t =σ(W o [h t-1 ,x t ]+b o ) (8)

[0049] h t =ot*tanh(C t ) (9)

[0050] Among them, t Represents the output value of the output gate; h t represents the hidden state of the current time step;

[0051] The hidden state is part of the LSTM network output and is used to carry the information of the network at the current time step. The hidden state is useful information extracted from the cell state and is used for network prediction or input of the next time step.

[0052] Compared with the prior art, the present invention provides an anomaly detection driven sea ice semi-supervised multi-task inversion method, which has the following beneficial effects:

[0053] (1) In order to solve the problem of huge data labeling workload and overcome the imbalance of sea ice and seawater DDM data, this paper applies the idea of ​​semi-supervised anomaly detection to the field of remote sensing sea ice inversion for the first time, and realizes the sea ice inversion task under limited data labels.

[0054] (2) In order to optimize the algorithm structure of sea ice inversion and reduce its redundancy, the present invention proposes an innovative integrated sea ice inversion framework. This multi-task network structure uniquely integrates the two modules of sea ice detection and sea ice concentration inversion, and implements two tasks through a single network. In the sea ice detection module, the present invention uses a potential deep learning guidance mechanism to mine and share detailed information and spatial context. At the same time, in the sea ice concentration inversion module, the DDM data features of the sea ice detection module are shared to the time series feature extraction network through the inter-module data feature sharing mechanism, which reduces the processing process of repeated data feature extraction and significantly reduces the model structure and parameter redundancy.

[0055] (3) The present invention not only focuses on spatial texture features, but also considers the temporal relationship between different features. While extracting DDM texture features, the data regression strategy is further optimized according to the temporal relationship between features. Through the construction of high-dimensional texture features and special attention to temporal features, the algorithm can more effectively capture the local details of the data and its relationship with global features, thereby effectively enhancing the model's data understanding and generalization capabilities in complex data environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is an overall structural diagram of the anomaly detection driven sea ice semi-supervised multi-task inversion method mentioned in Example 1 of the present invention;

[0057] Figure 2 Schematic diagram of the specific structure of the multi-task network mentioned in Example 1 of the present invention;

[0058] Figure 3 is a schematic diagram of the DDM image mentioned in Example 1 of the present invention; wherein, Figure 3 (a) is the seawater image. Figure 3 (b) is the image of sea ice concentration of 50%. Figure 3 (c) is the image when the sea ice concentration is 94%. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.

[0060] Embodiment 1:

[0061] See also Figure 1 ,This paper proposes a hybrid model sea ice extent prediction algorithm ,based on the feature focus capture mechanism, such as Figure 1 As shown, the following steps are included:

[0062] Step 1: Obtain the DDM data of TDS-1, input the DDM image into the data preprocessing module, and obtain standard DDM data that can be used for sea ice multi-task inversion through data screening, background noise subtraction and standardization.

[0063] TDS-1 is a satellite launched in 2014 for spaceborne GNSS-R research. Its orbit has an inclination of 98.4° and reaches an altitude of 635 km. The satellite carries 8 independent payloads. One of them is the Space GNSS Receiver Remote Sensing Instrument (SGR-ReSI), which generates DDM data by cross-correlating the scattered signal with a locally generated code replica with different time delays and Doppler shifts.

[0064] The DDM is a scattered power map consisting of 128 delay bins, each with a resolution of 244 ns, and 20 Doppler bins, each with a resolution of 500 Hz. The DDM is affected by key parameters such as the location of the specular reflection point (SP), the incidence angle of the satellite, the gain of the antenna, and the time of data collection. In this study, we use the SP location provided in the DDM metadata as the acquisition location of the DDM. The SP can be used to extract the SIC corresponding to the sea ice DDM from the reference data.

[0065] like Figure 3 Shown are DDM images of seawater, sea ice concentration of 50%, and sea ice concentration of 94%. The SIC value is obtained from the reference data of the SP position corresponding to each DDM. Figure 3 It can be seen that for different ocean surface conditions (seawater, sea ice, and different sea ice concentrations), DDM exhibits different image characteristics. Therefore, by analyzing the characteristics of the DDM image, sea ice and seawater, as well as different sea ice distribution states, can be effectively distinguished. This is because the smoothness of the reflecting surface plays a key role in the intensity of coherent reflection. Specifically, the smoother the surface, the stronger the coherent reflection, which is manifested as a smaller degree of diffusion of the DDM glittering area. Conversely, the greater the degree of diffusion. Since the surface of sea ice is relatively smooth, the scattered power of the sea ice surface comes almost entirely from SP, so the DDM image presents a more significant peak and a smaller glittering area, such as Figure 3 (c) In contrast, the DDM diagram of seawater usually shows a larger diffusion area and presents a horseshoe shape, such as Figure 3 (a).

[0066] According to the ratio of seawater to sea ice DDM obtained by TDS-1 in 2018, sea ice DDM data is 8.8% of seawater DDM data, and sea ice DDM has different forms according to different concentrations. Therefore, in this invention, seawater DDM is regarded as normal data for training, and sea ice DDM is regarded as abnormal data.

[0067] The data set preparation process in this embodiment is as follows:

[0068] (1) Data screening. Since not all acquired DDMs can be used for sea ice inversion, quality control is required according to certain conditions. The data set used in the present invention is mainly screened according to the following conditions:

[0069] The latitudes of north and south are greater than 55 degrees. Since sea ice data generally exist in high latitudes, DDM data are selected from areas above 55 degrees latitude in the north and south as sea ice data in the SND2018 dataset.

[0070] The satellite incident angle is less than 35 degrees. The smaller the satellite incident angle, the more vertically the signal is received by the satellite, and the smaller the path loss. Therefore, the smaller the incident angle, the smaller the coherent scattering component, and the better the signal transmission quality.

[0071] The peak signal-to-noise ratio (SNR) is greater than 0dB. The signal-to-noise ratio is an indicator used to measure the relationship between signal and noise. It is used to describe the relationship between the strength of the useful signal and the noise level. A lower SNR indicates that the DDM is affected by noise and cannot correctly display the characteristics of the reflective surface. Therefore, data with an SNR greater than 0dB are selected to remove all noise points below the threshold.

[0072] The antenna gain is greater than 3dB. Antenna gain is the efficiency of the antenna in receiving or sending signals in a certain direction. Therefore, the higher the antenna gain, the higher the efficiency of the antenna in receiving or sending signals in a certain direction, which means the better the communication quality.

[0073] The sea ice concentration is 0.15, and data with a SIC value higher than 15% are considered to contain sea ice reflection signals, otherwise they are considered to contain sea water reflection signal data.

[0074] (2) Data background noise subtraction. In order to further improve data quality and reduce the impact of noise generated by antennas and instruments on the data, the background noise subtraction method is used for processing. Background noise is a pixel frame without signal, specifically referring to the first four delay frequency bands of all Doppler frequency bands contained in each DDM. The noise calculation formula is shown in formula (1).

[0075]

[0076] Among them, τ1 and τ2 represent the pixel limits in the delay frequency band; f1 and f2 represent the pixel limits in the Doppler frequency band; X represents the number of all pixels in the noise box; DDM(τ,f) represents the value of a specific position of the original DDM data in the delay and Doppler frequency directions; Noise represents the noise value. After calculating the noise, subtract the noise from each bin of the DDM, as shown in Equation 2.

[0077] DDM remove_noise =DDM(τ,f)-Noise (2)

[0078] (3) Data standardization

[0079] Since the DDM data obtained by TDS-1 is not calibrated, each DDM needs to be normalized according to its peak power after removing the background noise. The normalization formula is shown in Equation 3.

[0080]

[0081] Where DDM(τ,f) represents the value of a specific position of the original DDM data in the direction of time delay and Doppler frequency. maxnosie The DDM value represents the peak power after subtracting the noise floor.

[0082] Step 2: The seawater DDM data in the obtained standard DDM data is used as input to the sea ice detection module in the multi-task network structure. By performing deep feature extraction on the input DDM data, the sea ice data and seawater data are accurately distinguished to complete the sea ice detection task.

[0083] The specific structural diagram of the multi-task network proposed in the present invention is as follows: Figure 2 The network includes a sea ice detection module and a sea ice concentration inversion module.

[0084] The main idea of ​​the sea ice detection module is anomaly detection. Since sea ice DDM data only accounts for 8% of seawater DDM data, the present invention uses the DDM image reflected by seawater as a normal sample and the sea ice DDM image as an abnormal sample. This module uses Deep Convolutional Genrative Adversarial Networks (DCGAN) as the main architecture to perform feature learning and reconstruction of normal samples. It consists of a generator G(x) and a discriminator D(x). During the training process, G(x) and D(x) compete with each other. The purpose of G(x) is to generate a reconstructed DDM image that is closer to the real DDM image x. D(x) cannot distinguish true from false, and the purpose of D(x) is to perfectly distinguish between x and The final result of this game is x and It becomes increasingly indistinguishable, eventually enabling G(x) to generate realistic This means that the network has learned the characteristics of normal seawater data. In subsequent tests, if sea ice data (that is, abnormal data) is input, since the generator has not learned the characteristics of sea ice data, the G(x) generated There will be a big difference with the input sea ice data, so D(x) can easily judge the difference between the input image and the reconstructed image, thereby achieving the purpose of sea ice detection.

[0085] The generator consists of an encoder and a decoder. The encoder maps the input image to a latent vector in the latent space by downsampling and extracts an abstract feature representation. Each layer of the encoder contains a convolution layer, a BatchNorm layer, and a LeakyReLU activation function. These four layers read the input data into the network and output the latent vector of the latent space. The feature maps output by each layer of the encoder are (10, 64, 64), (5, 32, 128), and (2, 9, 256). The decoder maps the latent vector of the latent space back to the original image space by upsampling, that is, reconstructing the abstract feature representation into an image again. Finally, the decoder generates a realistic image that has similar features to the original input image. The structure of the decoder corresponds to that of the encoder one by one, but the decoder uses the structure of the transposed convolution and BatchNorm layer. The transposed convolution is used to upsample the input latent space vector and increase the size of the feature map. Since the sea ice DDM data size used in the present invention is rectangular, in order to reconstruct the (20,128) image, the convolution kernel, step size, and padding of some layers also need to be adjusted accordingly. For example, the convolution kernel of the first layer of transposed convolution is 4, the step size is (3,4), and the padding is (1,2). In addition, except for the Tanh activation function in the output layer, the other layers of the decoder use the ReLU activation function. The feature maps output by each layer of the decoder are (2,9,256), (5,32,128), (10,64,64), and finally the reconstructed image (20,128,1) is output.

[0086] D(x) is used to judge the authenticity of the reconstructed image generated by G(x) and the input real image, and classify them. The goal of D(x) is to get a high score on the real image and a low score on the reconstructed image. In addition to the above functions, the D(x) of the SIMTRN proposed in the present invention also acts as a feature extractor, taking the intermediate layer output of the discriminator as a feature, and then calculating the feature matching loss between the generated image and the real image. This helps to ensure that the generated image is similar to the real image in the feature space. D(x) consists of a convolutional layer and a BatchNorm layer and a LeakyReLU activation function. The outputs of each layer are (20,128,64), (10,64,128), (5,32,256), (2,9,512), and (2,8,100).

[0087] Step 3: The high-dimensional features of the seawater DDM extracted by the sea ice detection module and the module output results are input into the sea ice concentration inversion module through a potential deep learning guidance mechanism, and the learned features are used to infer the sea ice concentration to complete the sea ice concentration inversion task.

[0088] Potential deep learning guidance mechanism, implemented through jump connections. This mechanism allows information to jump in the network, connecting the feature map of the last layer in the encoder with the feature map of the corresponding layer in the decoder, which helps to transmit more detailed information and spatial context, and restore richer detail information in the decoder stage, especially when processing edges and local features, maximizing the retention and restoration of detail loss in the indirect propagation process. This also helps to alleviate the gradient vanishing problem, making the network easier to train, reducing the difficulty of reconstruction, and generating higher quality images, thereby improving the performance of the entire network.

[0089] The sea ice concentration inversion module includes a discriminant matching unit and an LSTM (Long Short-Term Memory) network. The discriminant matching unit matches the output of the sea ice detection module with the original input image, and the data that is sea ice data and is judged as abnormal by the sea ice detection module is used as the input of the LSTM network. The LSTM network receives the output of the matching discriminant unit and obtains the high-dimensional feature representation of the corresponding DDM data through the inter-module data feature sharing mechanism to complete the sea ice concentration inversion task.

[0090] The LSTM used in the present invention includes two stacked LSTM layers, each of which can be regarded as a layer that performs sequence modeling on the input data, and the stacking of multiple layers allows the model to capture more complex sequence dependencies. Each LSTM layer has 128 hidden units, and each hidden unit contains a memory unit. The memory unit consists of a cell state and a hidden state. Both are information transmitted in the network.

[0091] Specifically, the cell state is the main memory unit of the LSTM network, responsible for storing and transmitting information. t is the state of the cell at time t. At each time step, the cell state is updated, and some information can be selectively retained or deleted. In LSTM, there are three key gating mechanisms to control the flow and update of cell states: forget gate, input gate, and output gate.

[0092] The forget gate determines how much of the cell state at the previous moment is retained in the cell state at the current moment, that is, it determines which information should be forgotten. t is the output value of the forget gate, as shown in Formula 4.

[0093] f t =σ(W f ·[h t-1 ,x t ]+b f ) (4)

[0094] Among them, f t represents the output value of the forget gate; σ represents the sigmoid layer; w f represents the weight matrix of the forget gate; x t Represents the input of the current time step; h t-1 represents the hidden state of the previous time step; [h t-1 ,x t ] represents the vector formed by connecting the hidden state of the previous time step and the input of the current time step; b f represents the bias vector;

[0095] The input gate uses the learned weights to decide which information should enter the cell state. It consists of two parts. The first part is the sigmoid layer, which determines what value to update. The second part is the tanh layer, which updates the information that needs to be updated to the cell state. t is the output value of the input gate. t and As shown in formula 5 and 6.

[0096] i t =σ(W i ·[h t -1,x t ])+b i (5)

[0097]

[0098] Among them, i t Represents the output value of the input gate; represents the candidate value of the input gate; W i , W f represents the weight matrix; b i 、b c represents the bias vector;

[0099] New cell state C t The memory in is composed of two parts. One part is the old memory left after the memory cells of the previous moment forget the useless memory. The other part is the useful information filtered out in the input gate as the new memory. The formula of the new cell state is shown in formula 7.

[0100]

[0101] The output gate determines which information should be output to the next layer or as the final output. t is the output value of the output gate. t With h t As shown in formulas 8 and 9.

[0102] o t =σ(W o [h t-1 ,x t ]+b o ) (8)

[0103] h t =ot*tanh(C t ) (9)

[0104] Among them, t Represents the output value of the output gate; h t represents the hidden state of the current time step;

[0105] The hidden state is the part of the LSTM network output that carries the information of the network at the current time step. The hidden state can be thought of as useful information extracted from the cell state for the network's prediction or input for the next time step.

[0106] In the sea ice detection module, the ultimate goal of the game between the generator and the discriminator is to make the generator generate more and more realistic reconstructed images, that is, the generator learns the characteristics of normal data. In order to achieve this goal, a loss function is needed to optimize the model. The sea ice detection module proposed in this invention uses three loss functions:

[0107] (1) Fighting against losses:

[0108] Adversarial loss is the game loss between the generator and the discriminator. Its goal is to make the reconstructed image of the generator more and more realistic. The goal of the generator is to reduce the probability that the reconstructed image is judged as false by the discriminator as much as possible, and the goal of the discriminator is to increase the probability of correctly distinguishing the real image and the reconstructed image as much as possible. The expression of adversarial loss is shown in formula 10:

[0109]

[0110] in, It is the expectation for the real data, which represents the expected probability that the discriminator will identify the real data as real; is the expectation for the reconstructed data, which represents the expectation of the probability that the discriminator will judge the generated data as false; x~p x represents sampling from the true data distribution.

[0111] (2) Contextual loss:

[0112] Adversarial loss can ensure that the reconstructed images of the generator are more and more realistic, but it cannot ensure that the model can learn the context of the input. Context loss is used to measure the difference in structure and contextual information between the generated image and the real image. The goal is to ensure that the generated image is similar to the real image at the pixel level while taking into account their contextual relationship. This loss can ensure that the model generates images similar to the context of normal samples, rather than generating reconstructed images that deviate from the structure of the real image. The expression of context loss is shown in Equation 11:

[0113]

[0114] in, Represents L1 loss, which is the mean absolute error (MAE).

[0115] (3) Potential losses:

[0116] The potential loss is used to measure the difference between the latent vector of the real image and the latent vector of the reconstructed image. Its goal is to ensure that the latent vector of the reconstructed image is similar to the latent vector of the real image, thereby improving the performance of the model. The present invention uses the last convolutional layer of the discriminator to extract x and The potential loss expression is shown in Equation 12.

[0117]

[0118] in, Represents L2 loss, which is the mean squared error (MSE).

[0119] Finally, the total loss of the sea ice detection module is the weighted sum of the above three losses, as shown in Equation 13. adv 、w con 、w lat is a weighting parameter used to adjust the dominance weight of each loss component in the overall objective function:

[0120] L=w adv L adv +w con L con +w lat L lat (13)

[0121] The sea ice concentration inversion module uses the MSE function, as shown in Equation 14:

[0122]

[0123] This loss function is used to measure the difference between the model's predicted sea ice value and the actual sea ice value, y~p y represents sampling from the true data distribution.

[0124] According to the above-mentioned DDM data screening method based on anomaly detection network, a graphics workstation with a CPU of Intel E-2124 with a main frequency of 3.3GHz and a single NVIDIA2080TI graphics card, an operating system of Ubuntu16.04, and a deep learning framework of Pytorch2.0 were used for verification.

[0125] The experimental parameters are as follows: the experimental learning rate is set to 0.0001, 500 epochs of training are performed on the DDM image data, and the input batch size is set to 64.

[0126] In order to verify the effectiveness and advancement of the present invention, a comparative test was conducted between the present invention and four existing deep learning-based sea ice detection algorithms and sea ice concentration inversion algorithms. The experiment proved that the sea ice detection accuracy of the present invention algorithm reached 0.9879, and the correlation coefficient of sea ice concentration inversion reached 0.9525, both of which were better than the existing comparative algorithms. This shows that the present invention achieves better sea ice detection and sea ice concentration inversion effects.

[0127] It should be noted that, in the present invention patent, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0128] 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. An anomaly detection driven sea ice semi-supervised multi-task inversion method, characterized in that: The following steps are involved: S1. Obtain the DDM data of TDS-1, input the obtained DDM data into the data preprocessing module, and obtain standard DDM data that can be used for sea ice multi-task inversion through data screening, background noise subtraction and normalization; S2. Design a multi-task grid structure, wherein the multi-task grid structure includes a sea ice detection module and a sea ice concentration inversion module; use the seawater DDM data in the standard DDM data obtained in S1 as input and input it into the sea ice detection module in the multi-task grid structure; perform deep feature extraction on the input seawater DDM data, accurately distinguish sea ice data from seawater data, and complete the sea ice detection task; S3. The high-dimensional features of the seawater DDM data extracted by the sea ice detection module and the module output results are input into the sea ice concentration inversion module through a potential deep learning guidance mechanism. The learned features are used to infer the sea ice concentration and complete the sea ice concentration inversion task.

2. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 1, characterized in that: The screening conditions for the data screening work described in S1 are: ① Data with latitudes greater than 55 degrees north and south; ②Data with satellite incidence angle less than 35 degrees; ③Peak signal-to-noise ratio is greater than 0dB; ④The antenna gain is greater than 3dB; ⑤The sea ice concentration is 0.15 as the dividing line.

3. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 2 is characterized in that: The background noise subtraction described in S1 specifically includes the following: The noise floor is calculated as: Based on formula (1), the formula for subtracting the background noise of each DDM data is: DDM remove_noise =DDM(τ,f)-Noise (2) Wherein, Noise represents the noise value; τ1 and τ2 represent the pixel limits in the delay frequency band; f1 and f2 represent the pixel limits in the Doppler frequency band; X represents the number of all pixels in the noise box; DDM(τ, f) represents the value of a specific position of the original DDM data in the delay and Doppler frequency directions.

4. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 3 is characterized in that: The normalization operation described in S1 specifically refers to normalizing each DDM according to its peak power. The normalization calculation formula is: Where DDM(τ,f) represents the value of a specific position of the original DDM data in the direction of time delay and Doppler frequency. maxnosie The DDM value represents the peak power after subtracting the noise floor.

5. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 1, characterized in that: The sea ice detection module is used for anomaly detection and includes a generator G(x) and a discriminator D(x); The DDM images of seawater reflection are input into the sea ice detection module as normal samples for training, and the sea ice DDM images are used as abnormal samples; During the training process, the generator G(x) and the discriminator D(x) of the sea ice detection module compete with each other, wherein the generator G(x) is used to generate a reconstructed DDM image that is closer to the real DDM image x. So that the discriminator D(x) cannot distinguish true from false. The discriminator D(x) is used to distinguish between x and 6. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 5, characterized in that: The generator G(x) consists of an encoder and a decoder; the encoder maps the input image to a latent vector in the latent space by downsampling and extracts an abstract feature representation; each layer of the encoder contains a convolutional layer, a BatchNorm layer, and a LeakyReLU activation function, which reads its input data into the network and outputs a latent vector in the latent space; The decoder maps the latent vector of the latent space back to the original image space by upsampling, that is, reconstructs the abstract feature representation into an image again, and finally generates a realistic image; The decoder adopts the structure of transposed convolution and BatchNorm layer, which corresponds to the structure of the encoder one by one; the transposed convolution is used to upsample the input latent space vector to increase the size of the feature map; the decoder uses the Tanh activation function in the output layer, and the other layers all use the ReLU activation function.

7. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 6, characterized in that: The D(x) is composed of a convolutional layer, a BatchNorm layer, and a LeakyReLU activation function; the discriminator D(x) is used to judge the authenticity of the reconstructed image generated by the generator G(x) and the input real image, and classify them. The goal of the discriminator D(x) is to get a high score on the real image and a low score on the reconstructed image; The discriminator D(x) also acts as a feature extractor, taking the intermediate layer output of the discriminator as features, and then calculating the feature matching loss between the generated image and the real image.

8. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 1, characterized in that: The potential deep learning guidance mechanism described in S3 is implemented through skip connections. By using this mechanism, information is transmitted in the network by jumping, connecting the feature map of the last layer in the encoder with the feature map of the corresponding layer in the decoder to transmit more detailed information and spatial context, and recovering richer detail information in the decoder stage.

9. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 8, characterized in that: The sea ice concentration inversion module described in S3 includes a discriminant matching unit and an LSTM network; The discrimination and matching unit judges and matches the output result of the sea ice detection module with the original input image, and the data that is sea ice data and is judged as abnormal by the sea ice detection module is used as the input of the LSTM network; the LSTM network receives the output of the matching discrimination unit, and obtains the high-dimensional feature representation of the corresponding DDM data through the inter-module data feature sharing mechanism, thereby completing the sea ice concentration inversion task; The LSTM network includes two stacked LSTM layers, each of which is regarded as a layer that performs sequence modeling on the input data, and the stacking of multiple layers allows the model to capture more complex sequence dependencies; each LSTM layer includes 128 hidden units, each of which contains a memory unit; the LSTM includes three key gating mechanisms, namely the forget gate, input gate and output gate, to control the flow and update of cell states.

10. The anomaly detection driven sea ice semi-supervised multi-task inversion method according to claim 9, characterized in that: The forget gate is used to determine how much of the unit state at the previous moment is retained to the unit state at the current moment. The specific function is expressed as: f t =σ(W f ·[h t-1 ,x t ]+b f ) (4) Among them, f t represents the output value of the forget gate; σ represents the sigmoid layer; w f represents the weight matrix of the forget gate; x t Represents the input of the current time step; h t-1 represents the hidden state of the previous time step; [h t-1 ,x t ] represents the vector formed by connecting the hidden state of the previous time step and the input of the current time step; b f represents the bias vector; The input gate determines which information should enter the cell state through the learned weights. It consists of two parts: the sigmoid layer and the tanh layer. The sigmoid layer determines what value to update, and the tanh layer updates the information to be updated into the cell state. The specific function is expressed as follows: i t =σ(W i ·[h t -1,x t ])+b i (5) Among them, i t Represents the output value of the input gate; represents the candidate value of the input gate; W i , W f represents the weight matrix; b i , b c represents the bias vector; New cell state C t The memory in is composed of the old memory left after the memory cells of the previous moment forget the useless memory and the useful information screened out by the input gate as new memory. The function of the new cell state is expressed as follows: The output gate is used to determine which information should be output to the next layer or as the final output. The specific function is expressed as follows: the t =σ(W o [h t-1 ,x t ]+b o ) (8) h t =o t *tanh(C t ) (9) Among them, t Represents the output value of the output gate; h t represents the hidden state of the current time step; The hidden state is part of the LSTM network output and is used to carry the information of the network at the current time step. The hidden state is useful information extracted from the cell state and is used for network prediction or input of the next time step.

Images