Glacier surge detection method and device based on time series analysis

Abstract

This invention provides a method and apparatus for detecting glacier pulsations based on time series analysis, belonging to the field of glacier pulsation detection technology. It aims to solve the technical problem that existing methods limit the accuracy of anomaly detection because they neglect the reasonable fluctuation range of the prediction results. The method includes: acquiring time-series remote sensing images of glacier flow velocity and extracting the glacier surface flow velocity from them to generate historical sequence data of glacier surface flow velocity; predicting the estimated mean of the future glacier flow velocity sequence using a pre-trained deep time series prediction model based on the historical sequence data of glacier surface flow velocity; denoising the estimated mean features using the inverse process of a diffusion model based on the estimated mean to obtain the distribution characteristics of the future flow velocity prediction sequence; and detecting the glacier flow velocity pulsation result based on the distribution characteristics.

Description

Technical Field

[0001] This invention relates to the field of glacier undulation detection technology, and more specifically, to a method and apparatus for glacier undulation detection based on time series analysis. Background Technology

[0002] With the continuous increase in remote sensing satellite data, the field of Earth observation has ushered in unprecedented opportunities, providing abundant data resources for monitoring surface changes. Long-term remote sensing data not only provides routine information on the temporal evolution of monitored areas but also offers valuable opportunities for timely detection of anomalies. This type of data is particularly important for data-driven analysis methods and also provides a more reliable technical means for detecting glacier leaps. Glacier leaps refer to the phenomenon where, during glacier flow, the friction between the glacier base and bedrock decreases sharply, leading to a sudden increase in glacier flow velocity. Glacier flow velocity can be viewed as a fluctuation signal; a sudden increase in velocity indicates the occurrence of a glacier leap. Therefore, glacier flow velocity data obtained through remote sensing, combined with time-series anomaly detection algorithms, can enable automated monitoring of glacier leaps.

[0003] Conventional time series anomaly detection algorithms can be divided into two main categories: prediction-based methods and decomposition-based methods. However, both prediction-based and decomposition-based methods generally neglect the uncertainty modeling in the time series prediction process, that is, they ignore the reasonable fluctuation range of the prediction results, thus limiting the accuracy of anomaly detection. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for detecting glacier undulations based on time series analysis, aiming to solve the technical problem that existing methods limit the accuracy of anomaly detection by ignoring the reasonable fluctuation range of the prediction results.

[0005] One aspect of the present invention provides a glacier pulsation detection method based on time series analysis, comprising: acquiring time series remote sensing images of glacier flow velocity, and extracting the surface flow velocity of the glacier from the time series remote sensing images to generate historical sequence data of glacier surface flow velocity; predicting the estimated mean features of the future glacier flow velocity sequence based on the historical sequence data of glacier surface flow velocity using a pre-trained deep time series prediction model; inputting the estimated mean features into a pre-trained diffusion model and performing the following operations: denoising the estimated mean features using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence, wherein the distribution features characterize the distribution range of glacier flow velocity from the lowest to the highest speed; and detecting the pulsation result of glacier flow velocity based on the distribution features.

[0006] According to an embodiment of the present invention, extracting the surface velocity of a glacier from a time-series remote sensing image and generating historical sequence data of glacier surface velocity includes: identifying and tracking the grayscale features of the glacier surface in two remote sensing images spaced a specified time interval using a cross-correlation model based on the time-series remote sensing image to obtain the surface displacement of the glacier within the specified time interval; calculating the surface flow velocity of the glacier based on the surface displacement; and generating historical sequence data of glacier surface velocity based on the time series corresponding to the surface flow velocity.

[0007] According to an embodiment of the present invention, the distribution characteristics of the future velocity prediction sequence are obtained by denoising the estimated mean features using the inverse process of the diffusion model. This includes: using the estimated mean features as the maximum diffusion step number of the noisy glacier surface velocity future sequence data, using the noise prediction model to predict the noisy glacier surface velocity future sequence data corresponding to the previous diffusion step, until the diffusion step number is 1, to obtain a prediction result of clean glacier surface velocity future sequence data; repeating the above inverse process multiple times to obtain a set of prediction results of glacier surface velocity future sequence data; and estimating the distribution characteristics of the future velocity prediction sequence based on the prediction results of the clean glacier surface velocity future sequence data.

[0008] According to an embodiment of the present invention, the detection of glacier flow velocity fluctuations based on distribution characteristics includes: obtaining an upper limit of glacier flow velocity based on distribution characteristics and a preset quantile, wherein the upper limit represents the maximum range of normal glacier flow velocity; comparing the actual obtained glacier surface flow velocity sequence with the upper limit; and detecting glacier flow velocity fluctuations in response to the actual glacier flow velocity exceeding the upper limit of the corresponding time point at multiple consecutive time points.

[0009] According to an embodiment of the present invention, pre-training a deep time-series prediction model includes: acquiring glacier surface velocity sequences from historical periods where no glacier surge events have occurred, and constructing a training dataset; inputting the training dataset into the pre-constructed deep time-series prediction model, and obtaining the output of the deep time-series prediction model, wherein the output of the deep time-series prediction model includes the estimated mean of the future glacier velocity sequence; constructing a loss function based on the difference between the output of the deep time-series prediction model and the actual future glacier velocity sequence, and training the parameters of the deep time-series prediction model by minimizing the loss function.

[0010] According to an embodiment of the present invention, pre-training the diffusion model includes: using the forward process of the diffusion model, randomly adding Gaussian noise to the estimated mean to obtain noisy future sequence data of glacier surface flow velocity; using the noisy future sequence data of glacier surface flow velocity as input to the reverse process of the diffusion model, and using the estimated mean as conditional information of the diffusion model, predicting the Gaussian noise component added in the forward process through a noise prediction model; constructing a loss function based on the difference between the Gaussian noise component output by the noise prediction model and the Gaussian noise component added in the forward process, and training the parameters of the noise prediction model by minimizing the loss function.

[0011] According to an embodiment of the present invention, the method of obtaining noisy glacier surface velocity future sequence data by randomly adding Gaussian noise to the estimated mean during the forward process of the diffusion model includes: configuring a variance scheduling table, wherein the variance scheduling table is used to control the rate at which Gaussian noise is added to the estimated mean and the actual glacier future velocity sequence during the forward process; and calculating the noisy glacier surface velocity future sequence data corresponding to a specified number of diffusion steps based on the variance scheduling table, using the estimated mean, the actual glacier future velocity sequence, and the randomly sampled Gaussian noise.

[0012] Another aspect of the present invention provides a glacier pulsation detection device based on time series analysis, comprising: an acquisition module for acquiring time series remote sensing images of glacier flow velocity and extracting the surface flow velocity of the glacier from the time series remote sensing images to generate historical sequence data of glacier surface flow velocity; a prediction module for predicting the estimated mean features of the future glacier flow velocity sequence based on the historical sequence data of glacier surface flow velocity using a pre-trained deep time series prediction model; and inputting the estimated mean features into a pre-trained diffusion model and executing the following modules: a denoising module for denoising the estimated mean features using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence, wherein the distribution features characterize the distribution range of glacier flow velocity from the lowest to the highest speed; and a detection module for detecting the pulsation result of glacier flow velocity based on the distribution features.

[0013] Another aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0014] Another aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method described above.

[0015] Another aspect of the present invention provides a computer program product including computer-executable instructions that, when executed, are used to implement the method described above.

[0016] Compared with the prior art, the glacier surge detection method and apparatus based on time series analysis provided in this invention have at least the following beneficial effects:

[0017] (1) The glacier pulsation detection method and device based on time series analysis provided in this embodiment of the invention estimates the average future glacier flow velocity through a depth model and introduces a diffusion model to generate a reasonable range of flow velocity fluctuations, thereby replacing the fixed threshold with a dynamic confidence interval to achieve more accurate pulsation anomaly identification.

[0018] (2) The glacier pulsation detection method and device based on time series analysis provided in this embodiment of the invention uses a carefully designed depth time series prediction model to process historical data, capture the estimated mean as the conditional information of the diffusion model, and enhance the uncertainty capture capability of the model.

[0019] (3) The glacier pulsation detection method and device based on time series analysis provided in the embodiments of the present invention uses a diffusion model as a model uncertainty modeling tool. Unlike the ordinary diffusion model which adds noise during the diffusion process until the standard normal distribution noise is processed, the endpoint of the diffusion process is modeled as a Gaussian distribution with the initial estimate of the future flow velocity sequence obtained by the depth time series prediction model as the mean, which improves the accuracy and robustness of the model uncertainty estimation.

[0020] (4) The glacier pulsation detection method and device based on time series analysis provided in the embodiments of the present invention model the uncertainty interval of the glacier flow velocity sequence through diffusion model, realizes high-precision distribution estimation of future flow velocity sequence, and does not require manual setting of anomaly detection threshold, providing a scientific reference for judging glacier pulsation events.

[0021] (5) The glacier pulsation detection method and device based on time series analysis provided in this embodiment of the invention integrates a diffusion model and a depth time series prediction model. The two models are trained simultaneously through a hybrid optimization strategy. It can be used as a plug-and-play module, compatible with various depth time series prediction models, and has high universality and scalability. Attached Figure Description

[0022] The above and other objects, features and advantages of the present invention will become more apparent from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0023] Figure 1 A flowchart illustrating a glacier pulsation detection method based on time series analysis according to an embodiment of the present invention is shown schematically.

[0024] Figure 2A schematic diagram illustrates the structure of a glacier pulsation detection device based on time series analysis according to an embodiment of the present invention;

[0025] Figure 3 The diagram schematically illustrates a structural block diagram of an electronic device suitable for implementing a time-series analysis-based glacier undulation detection method according to an embodiment of the present invention. Detailed Implementation

[0026] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] In the embodiments of this invention, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to maintain the security of user personal information and network security.

[0031] With the continuous increase in remote sensing satellite data, the field of Earth observation has ushered in unprecedented opportunities, providing abundant data resources for monitoring surface changes. Long-term remote sensing data not only provides routine information on the temporal evolution of monitored areas but also offers valuable opportunities for timely detection of anomalies. This type of data is particularly important for data-driven analysis methods and also provides a more reliable technical means for detecting glacier leaps. Glacier leaps refer to the phenomenon where, during glacier flow, the friction between the glacier base and bedrock decreases sharply, leading to a sudden increase in glacier flow velocity. Glacier flow velocity can be viewed as a fluctuation signal; a sudden increase in velocity indicates the occurrence of a glacier leap. Therefore, glacier flow velocity data obtained through remote sensing, combined with time-series anomaly detection algorithms, can enable automated monitoring of glacier leaps.

[0032] Conventional time series anomaly detection algorithms can be divided into two main categories: prediction-based methods and decomposition-based methods.

[0033] The core idea of ​​prediction-based anomaly detection methods is to construct a time series prediction model, use historical data to predict future values, and then compare the actual observed values ​​with the predicted values. When the deviation exceeds a preset threshold, it is judged as an anomaly. One of the most classic models is ARIMA (Autoregressive Integrated Moving Average), which mainly consists of three parts: autoregression (AR), integration (I), and moving average (MA). The autoregressive part models the linear dependence of the series' historical values, reflecting the dependence of the current value on data from several past periods. The integration operation processes non-stationary series through differencing, eliminating trend terms and seasonal fluctuations through first-order or multi-order differencing, so that the data satisfies the stationarity assumption. The moving average part optimizes the current prediction based on a linear combination of historical prediction errors, reflecting the cumulative error correction mechanism. This model achieves dynamic prediction through a recursive mechanism. When the deviation between the actual value and the predicted value exceeds a set threshold (such as a multiple of the standard deviation), it can be judged as an anomaly. It is suitable for time series data with linear trends or periodicity, but its adaptability to nonlinear abrupt changes is limited.

[0034] Decomposition-based time series anomaly detection algorithms identify anomalies by breaking down a series into components such as trend, seasonality, periodicity, and residuals, and analyzing residuals exceeding expected ranges or abrupt changes in relationships between components. A classic example is BFAST (Breaks For Additive Season and Trend), a statistical method combining time series decomposition and structural abrupt change detection. Its core lies in identifying trend and seasonal abrupt changes in time series through an iterative process. This method first uses locally weighted regression to decompose the original time series into trend, seasonal, and residual components. The trend component captures long-term changes, the seasonal component reflects periodic fluctuations, and the residuals contain random noise. In the abrupt change detection phase, BFAST applies the Bai-Perron multiple structural abrupt change test to the trend and seasonal components respectively, using a piecewise linear regression framework to find the optimal combination of breakpoints and identify time points where significant changes in trend or seasonal patterns occur. The BFAST method has advantages such as handling missing data and adapting to different time scales, providing a solution for analyzing structural changes in non-stationary time series. It is suitable for data with obvious trends and periodicities, but it has lower sensitivity to non-periodic anomalies and parameter tuning relies on experience.

[0035] It is evident that existing prediction-based anomaly detection methods, whether traditional ARIMA models or advanced deep time series prediction networks, primarily focus on single-point predictions, i.e., the expected mean of future time series, while giving less consideration to the uncertainties in the prediction process. However, uncertainty modeling is crucial in time series anomaly detection. This is because the task not only focuses on the expected value of the predicted value but also needs to determine the reasonable fluctuation range of the future series (such as upper and lower tolerances). Therefore, the accuracy of uncertainty modeling directly affects the accuracy of anomaly detection. In summary, existing methods generally neglect uncertainty modeling in the time series prediction process, that is, they ignore the reasonable fluctuation range of the prediction results, thus limiting the accuracy of anomaly detection.

[0036] In recent years, deep learning-based time series prediction models have made significant progress. Deep time series prediction models based on Recurrent Neural Networks (RNNs), Temporal Convolutional Networks (TCNs), and Transformers have demonstrated their respective advantages in different application scenarios. RNNs are deep learning models capable of processing sequential data. Through recurrent connections, the network can capture long-term dependencies in time series. Their advantage lies in predicting future values ​​based on historical information, making them particularly suitable for time series with sequential correlations. TCNs are time series prediction models based on convolutional neural network structures. They use convolutional operations to capture local dependencies in sequences and employ extended convolutional layers to handle long-term dependencies. Their advantages include suitability for parallel computation, reduced training time, and a better ability to address the poor performance of RNNs in long-term dependency tasks. Transformer models capture long-range dependencies through self-attention mechanisms, avoiding the gradient problem of RNNs when processing long sequences. They also utilize multi-head attention mechanisms to process multiple information sources in parallel, demonstrating outstanding performance in multivariate time series prediction and have become a research hotspot in the field of time series analysis in recent years.

[0037] Thanks to significant advancements in deep time-series prediction models, prediction-based anomaly detection algorithms are gradually becoming the mainstream trend in this field. Developing specialized time-series anomaly detection methods for glacier undulation detection holds promise for overcoming the limitations of traditional methods in terms of automation and detection efficiency, providing a more advanced and reliable technological approach for this field.

[0038] Based on this, this invention provides a glacier pulsation detection method based on time series analysis, aiming to fully leverage the powerful capabilities of depth time series prediction models in mean estimation, while effectively modeling the uncertainties in time series predictions using diffusion models. In this way, based on accurately predicting the mean of future flow velocity sequences, a reasonable fluctuation range of the flow velocity sequence can be captured, thereby dynamically determining the threshold for outlier detection and achieving more accurate and efficient glacier pulsation detection.

[0039] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0040] Figure 1 A flowchart illustrating a glacier pulsation detection method based on time series analysis according to an embodiment of the present invention is shown.

[0041] like Figure 1As shown, the glacier surge detection method based on time series analysis in this embodiment may include, for example, operations S1 to S4.

[0042] In operation S1, time-series remote sensing images of glacier flow velocity are acquired, and the surface flow velocity of the glacier is extracted from the time-series remote sensing images to generate historical sequence data of glacier surface flow velocity.

[0043] In operation S2, based on the historical sequence data of glacier surface flow velocity, a pre-trained deep time series prediction model is used to predict the estimated mean characteristics of the future glacier flow velocity sequence.

[0044] Input the estimated mean features into the pre-trained diffusion model and perform the following operations S3~S4:

[0045] In operation S3, the estimated mean features are denoised using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence. The distribution features characterize the distribution range of glacier flow velocity from the lowest to the highest speed.

[0046] In operation S4, based on the distribution characteristics, the fluctuation results of glacier flow velocity were detected.

[0047] The glacier pulsation detection method based on time series analysis provided in this invention estimates the average future glacier flow velocity through a depth model and introduces a diffusion model to generate a reasonable range of flow velocity fluctuations. Thus, a dynamic confidence interval replaces a fixed threshold, achieving more accurate identification of pulsation anomalies.

[0048] According to an embodiment of the present invention, operation S1 extracts the surface velocity of a glacier from a time-series remote sensing image to generate historical sequence data of glacier surface velocity, which may specifically include:

[0049] Based on time-series remote sensing images, a cross-correlation model is used to identify and track the grayscale features of the glacier surface in two remote sensing images spaced at a specified time interval, thereby obtaining the surface displacement of the glacier within the specified time.

[0050] The surface flow velocity of the glacier is calculated based on the surface displacement.

[0051] Based on the time series corresponding to the surface flow velocity, historical sequence data of glacier surface flow velocity is generated.

[0052] In this embodiment, for example, a set of remote sensing images arranged in chronological order can be used to extract the surface velocity of the glacier using a feature tracking method.

[0053] Feature tracking, also known as pixel tracking, incorporates a core algorithm called normalized cross-correlation. The basic principle of this algorithm is to extract cells of a certain size from the main image and then search for the cell that best matches the image from the search area—that is, the cell with the highest correlation. This algorithm can automatically track the grayscale features of the glacier surface, such as cracks or small ice mounds, on two optical remote sensing images taken at a certain time interval, thereby obtaining the surface displacement during that period and ultimately the surface flow velocity of the entire ice surface.

[0054] The calculation formula for the cross-correlation model is as follows:

[0055]

[0056] in, This refers to the reference image in the two registered images; Indicates a search for images; assuming , The two images show relative displacement, so the glacier displacement occurring during the time interval between the two image acquisitions is represented by (…). , )express.

[0057] To obtain more accurate information about glacier movement, a Fourier transform can be performed on the calculation formula of the above cross-correlation model in the frequency domain. After the transform, the following relationship exists:

[0058]

[0059] in, Fourier spectrum image; Fourier spectrum image. , These represent the changes in the spectrum along the row and column directions, respectively.

[0060] Considering the reliability and simplicity of the normalization form, the normalized cross-correlation value of the image in the frequency domain is obtained, where the normalization coefficients are as follows:

[0061]

[0062] The superscript * indicates the conjugate complex number. Representing the normalized cross-power spectrum, we can obtain the following by performing an inverse Fourier transform:

[0063]

[0064] After applying the above formula, the displacement of the two images on the glacier surface can be obtained by taking the peak point of the two-dimensional impulse function. .

[0065] According to an embodiment of the present invention, operation S2 uses a pre-trained deep time series prediction model based on the historical sequence data of glacier surface flow velocity to predict the estimated mean characteristics of the future glacier flow velocity sequence. For example, any deep time series prediction model can be selected, such as LSTM Encoder-Decoder, Informer (efficient long-sequence time series prediction model), CrossFormer (multivariate time series prediction model), etc., with the historical sequence data of glacier surface flow velocity as the model input to obtain the estimated mean characteristics of the future flow velocity sequence.

[0066] According to embodiments of the present invention, pre-training a deep temporal prediction model may include, for example:

[0067] Obtain glacier surface velocity sequences from historical periods when no glacier surge events occurred, and construct a training dataset;

[0068] The training dataset is input into a pre-built deep time series prediction model, and the output of the deep time series prediction model is obtained. The output of the deep time series prediction model includes the estimated mean of the future flow velocity sequence of the glacier.

[0069] A loss function is constructed based on the difference between the output of the deep time series prediction model and the actual future glacier flow velocity sequence. The parameters of the deep time series prediction model are trained by minimizing the loss function.

[0070] In this embodiment, using the same generation method as the glacier surface velocity history sequence data, glacier surface velocity sequences from historical periods without glacier surge events are obtained pixel-by-pixel to construct a training dataset. Then, each glacier velocity sequence in the training dataset is segmented and sliced ​​to obtain a pair of historical sequences (input values) and future sequences (values ​​to be predicted) samples. .

[0071] Choose an arbitrary depth time series prediction model to train the historical glacier surface flow sequences in the dataset. As input to the model, the estimated mean of the future flow velocity sequence The deep time series prediction model is trained using the model output.

[0072] Specifically, the selected deep time series prediction model is denoted as... The parameters are It can be done through The estimated mean of the future glacier flow velocity sequence was obtained. .

[0073] Assume the true value of the future glacier flow velocity sequence (i.e., the actual future glacier flow velocity sequence) is: The loss is estimated by calculating the mean value using the following formula. :

[0074]

[0075] in, This represents the Kullback-Leibler divergence from distribution p to distribution q.

[0076] The deep time series prediction model is trained by continuously minimizing the loss function.

[0077] The glacier surge detection method based on time series analysis provided in this invention uses a carefully designed depth time series prediction model to process historical data, capture the estimated mean as conditional information for the diffusion model, and enhance the model's ability to capture uncertainty.

[0078] According to an embodiment of the present invention, operation S3 uses the reverse process of the diffusion model to denoise the estimated mean features to obtain the distribution features of the future flow velocity prediction sequence, which may specifically include:

[0079] Using the estimated mean characteristics as the maximum diffusion step number, the future sequence data of noisy glacier surface velocity is predicted by the noise prediction model to obtain the future sequence data of noisy glacier surface velocity corresponding to the previous diffusion step number, until the diffusion step number is 1, and a prediction result of clean glacier surface velocity future sequence data is obtained.

[0080] By repeating the above reverse process multiple times, a set of prediction results for the future sequence data of glacier surface flow velocity is obtained;

[0081] Based on the prediction results of future flow velocity sequence data from clean glacier surfaces, the distribution characteristics of the future flow velocity prediction sequence are estimated.

[0082] In this embodiment, the reverse process of the diffusion model iterates from the maximum diffusion step number T→T-1→T-2… all the way to t=1, which is regarded as a complete sampling process. Each reverse process produces an estimate of the future flow velocity sequence.

[0083] By repeating the above reverse process multiple times, a set of estimated results for the future flow velocity sequence can be obtained. This set of results represents the distribution characteristics of the future flow velocity sequence.

[0084] The future flow velocity sequence fluctuates within a possible data range. Each reverse process is equivalent to obtaining a sample result within this data range. Using a set of sample results (such as 100 estimation results), the distribution characteristics of this data range can be estimated.

[0085] According to embodiments of the present invention, pre-training a diffusion model may include, for example:

[0086] Input the estimated mean of the deep time series prediction model into the pre-built diffusion model and perform the following operations:

[0087] By using the forward process of the diffusion model, Gaussian noise is randomly added to the estimated mean to obtain noisy future sequence data of glacier surface flow velocity;

[0088] Noisy glacier surface flow velocity future sequence data are used as input to the reverse process of the diffusion model, and the estimated mean is used as the conditional information of the diffusion model. The Gaussian noise component added in the forward process is predicted by the noise prediction model.

[0089] A loss function is constructed based on the difference between the Gaussian noise component output by the noise prediction model and the Gaussian noise component added during the forward process. The parameters of the noise prediction model are trained by minimizing the loss function.

[0090] According to an embodiment of the present invention, by utilizing the forward process of a diffusion model, Gaussian noise is randomly added to the estimated mean to obtain noisy future sequence data of glacier surface flow velocity, which may specifically include:

[0091] Configure a variance scheduling table, which is used to control the rate at which Gaussian noise is added to the estimated mean and the actual future glacier flow velocity sequence during the forward process;

[0092] Based on the variance scheduling table, the noisy future glacier surface velocity sequence data corresponding to a specified number of diffusion steps is calculated by estimating the mean, the actual future glacier velocity sequence, and randomly sampled Gaussian noise.

[0093] In this embodiment, the diffusion model is mainly used for uncertainty estimation.

[0094] The forward process of the diffusion model, also known as the diffusion process, involves processing clean, noise-free data. By continuously adding noise disturbances, a pure noise data set eventually emerges. ,in This represents the total number of diffusion steps.

[0095] Specifically, given the variance scheduling table of the diffusion process. The following formula is used to utilize the true value of the future flow velocity sequence. The estimated mean of the deep time series prediction model And randomly sampled Gaussian noise ε, to calculate the noisy data at the t-th diffusion step. :

[0096]

[0097] in, , , Noisy data at the t-th diffusion step Can be considered clean data and estimated mean The interpolation process between them.

[0098] In this embodiment, the reverse process of the diffusion model, i.e., the denoising process, is to process the noisy data using a neural network model (e.g., a noise prediction model). By performing gradual denoising, a clean, noise-free dataset is eventually recovered from the noisy data. This enables data generation.

[0099] The estimated mean of the deep time series prediction model Noisy data at step t and randomly sampled Gaussian noise Estimate the noisy data at the (t-1)th diffusion step. :

[0100]

[0101] in,

[0102]

[0103]

[0104]

[0105] By sampling multiple times using a diffusion model, an uncertainty estimate of the future flow velocity sequence is obtained. Specifically, the mean of the future sequence estimate obtained based on a deep time series prediction model is... and a set of randomly sampled Gaussian noise By following the denoising steps of the reverse process, the noisy data is gradually restored to clean data, and the distribution of the predicted sequence is obtained.

[0106] The glacier pulsation detection method based on time series analysis provided in this invention uses a diffusion model as a model uncertainty modeling tool. Unlike ordinary diffusion models that add noise during the diffusion process until the standard normal distribution noise is processed, this method models the endpoint of the diffusion process as a Gaussian distribution with the initial estimate of the future flow velocity sequence obtained by the depth time series prediction model as the mean, thereby improving the accuracy and robustness of the model uncertainty estimation.

[0107] For example, the endpoint of the diffusion process can be modeled as follows:

[0108]

[0109] Then, the generation process is carried out by introducing a denoising process with conditional information.

[0110] In this embodiment, to improve the overall performance of the model, the deep time series prediction model and the noise prediction model can be jointly optimized through the reverse process of the diffusion model during training.

[0111] For example, first, the ground truth of future flow rate sequences in the training dataset. Add noise interference to generate noisy data. .

[0112] Specifically, for a certain sample in the training dataset ,Will As the starting point of the forward process (diffusion step count is 0), it is denoted as A random sample is taken from the standard normal distribution. Noise sequences with the same dimension Randomly sample diffusion steps t from the integer set [0, T], and use... and the generated estimated mean According to the preset noise scheduling strategy The noisy data at step t of the diffusion process is calculated using the following formula. :

[0113]

[0114] Then, the generated noisy data As input to the diffusion model, the deep time-series prediction model and the noise prediction model are simultaneously optimized through the reverse process of the diffusion model.

[0115] Specifically, given The diffusion step number t is first obtained from the following formula. :

[0116]

[0117] in, This represents a noise prediction model for a neural network, with parameters as follows: To estimate the mean and the noisy data at step t As input, predict the noise component added during the forward pass.

[0118] Secondly, the coefficients of the three denoising processes are calculated using the following formula. , :

[0119]

[0120]

[0121]

[0122] Secondly, based on the coefficients of the denoising process , and Calculate the posterior mean :

[0123]

[0124] when At that time, noisy data can be obtained. The noisy data at step t-1 after one step of denoising. :

[0125]

[0126] in, This represents noise sampled from a standard normal distribution.

[0127] The posterior variance is calculated using the following formula:

[0128]

[0129] when In this case, no noise needs to be added, meaning the final denoised prediction data is obtained directly. :

[0130]

[0131] According to forecast data Calculate diffusion loss :

[0132]

[0133] Combined with mean loss The final objective function for joint training of the two networks is obtained as follows:

[0134]

[0135] The glacier undulation detection method based on time series analysis provided in this invention integrates a diffusion model and a depth time series prediction model. It trains both models simultaneously through a hybrid optimization strategy, making it a plug-and-play module compatible with various depth time series prediction models and exhibiting high universality and scalability.

[0136] According to an embodiment of the present invention, operation S4, based on distribution characteristics, detects the fluctuation results of glacier flow velocity, which may specifically include:

[0137] Based on the distribution characteristics, the upper bound of glacier flow velocity is obtained according to the preset quantiles, where the upper bound represents the maximum range of normal glacier flow velocity.

[0138] The actual obtained glacier surface velocity sequence was compared with the upper bound;

[0139] When the actual glacier flow velocity at multiple consecutive time points exceeds the upper limit of the corresponding time point, a jump in glacier flow velocity is detected.

[0140] In this embodiment, after obtaining the expected distribution of the glacier surface velocity sequence over a future period, an upper bound for a reasonable velocity range is obtained based on preset quantiles. Then, the actual obtained glacier surface velocity sequence is compared with the predicted upper bound. If multiple consecutive true values ​​exceed the reasonable range, an anomaly is determined, indicating a glacier surge event.

[0141] The glacier pulsation detection method based on time series analysis provided in this invention achieves high-precision distribution estimation of future flow velocity sequences by modeling the uncertainty interval of glacier flow velocity sequences through a diffusion model. It eliminates the need for manually setting anomaly detection thresholds and provides a scientific reference for determining glacier pulsation events.

[0142] Figure 2 A schematic diagram of a glacier pulsation detection device based on time series analysis according to an embodiment of the present invention is shown.

[0143] like Figure 2 As shown, the glacier pulsation detection device 200 based on time series analysis according to an embodiment of the present invention includes: an acquisition module 210, a prediction module 220, a denoising module 230, and a detection module 240.

[0144] The acquisition module 210 is used to acquire time-series remote sensing images of glacier flow velocity and extract the surface flow velocity of the glacier from the time-series remote sensing images to generate historical sequence data of glacier surface flow velocity.

[0145] The prediction module 220 is used to predict the estimated mean characteristics of the future glacier flow velocity sequence based on the historical sequence data of glacier surface flow velocity and using a pre-trained depth time series prediction model; and inputs the estimated mean characteristics into a pre-trained diffusion model and executes the following modules:

[0146] The denoising module 230 is used to denoise the estimated mean features using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence, wherein the distribution features characterize the distribution range of glacier flow velocity from the lowest to the highest speed.

[0147] The detection module 240 is used to detect the fluctuation results of glacier flow velocity based on distribution characteristics.

[0148] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention, or at least part of the functions of any one or more of them, can be implemented in a single module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be implemented by being divided into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, and firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present invention can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0149] For example, any plurality of the acquisition module 210, prediction module 220, denoising module 230, and detection module 240 can be combined into one module / unit / subunit, or any one of these modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits can be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of the present invention, at least one of the acquisition module 210, prediction module 220, denoising module 230, and detection module 240 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 210, prediction module 220, denoising module 230 and detection module 240 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0150] It should be noted that the glacier undulation detection device based on time series analysis in the embodiments of the present invention corresponds to the glacier undulation detection method based on time series analysis in the embodiments of the present invention. For a detailed description of the glacier undulation detection device based on time series analysis, please refer to the glacier undulation detection method based on time series analysis, which will not be repeated here.

[0151] Figure 3 The diagram schematically illustrates a structural block diagram of an electronic device suitable for implementing a time-series analysis-based glacier undulation detection method according to an embodiment of the present invention. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0152] like Figure 3 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage portion 308 into a random access memory (RAM) 303. The processor 301 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 301 may also include onboard memory for caching purposes. The processor 301 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0153] The storage section 308 stores various programs and data required for the operation of the electronic device 300. The processor 301, ROM 302, and storage section 308 are interconnected via bus 304. The processor 301 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 302 and / or storage section 308. It should be noted that the programs may also be stored in one or more memories other than ROM 302 and storage section 308. The processor 301 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0154] According to an embodiment of the present invention, the electronic device 300 may further include an input / output (I / O) interface 305, which is also connected to a bus 304. The electronic device 300 may also include one or more of the following components connected to the input / output (I / O) interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the input / output (I / O) interface 305 as needed. A removable medium 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 310 as needed so that computer programs read from it can be installed into the storage section 308 as needed.

[0155] According to embodiments of the present invention, the method flow according to embodiments of the present invention can be implemented as a computer software program. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by processor 301, it performs the functions defined in the system of the embodiments of the present invention. According to embodiments of the present invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0156] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0157] According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0158] For example, according to embodiments of the present invention, a computer-readable storage medium may include the ROM 302 and / or storage portion 308 described above and / or one or more memories other than ROM 302 and storage portion 308.

[0159] Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of the present invention. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the methods provided in the embodiments of the present invention.

[0160] When the computer program is executed by the processor 301, it performs the functions defined in the system / apparatus of this embodiment of the invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0161] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via communication section 309, and / or installed from removable medium 311. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0162] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0163] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or pairings fall within the scope of this invention.

[0164] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for detecting glacier undulations based on time series analysis, characterized in that, The method includes: Time-series remote sensing images of glacier flow velocity are acquired, and the surface flow velocity of the glacier is extracted from the time-series remote sensing images to generate historical sequence data of glacier surface flow velocity. Based on the historical sequence data of glacier surface flow velocity, the estimated mean characteristics of the future glacier flow velocity sequence are predicted using a pre-trained deep time series prediction model. The estimated mean features are input into a pre-trained diffusion model, and the following operations are performed: The estimated mean features are denoised using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence, including: Using the estimated mean feature as the future sequence data of noisy glacier surface velocity with the maximum diffusion step, the noise prediction model is used to predict the future sequence data of noisy glacier surface velocity corresponding to the previous step for each diffusion step, until the diffusion step is 1, and a prediction result of clean glacier surface velocity future sequence data is obtained. By repeating the above reverse process multiple times, a set of prediction results for the future sequence data of glacier surface flow velocity is obtained; Based on the prediction results of future sequence data of clean glacier surface flow velocity, the distribution characteristics of the future flow velocity prediction sequence are estimated, wherein the distribution characteristics characterize the distribution range of glacier flow velocity from the lowest to the highest speed. Based on the aforementioned distribution characteristics, the fluctuations in glacier flow velocity were detected, including: Based on the aforementioned distribution characteristics, the upper bound of the glacier flow velocity is obtained according to a preset quantile, wherein the upper bound represents the maximum range of normal glacier flow velocity. The actual obtained glacier surface flow velocity sequence is compared with the upper bound. When the actual glacier flow velocity at multiple consecutive time points exceeds the upper limit of the corresponding time point, a jump in glacier flow velocity is detected.

2. The method according to claim 1, characterized in that, The step of extracting the surface velocity of the glacier from the time-series remote sensing images and generating historical sequence data of glacier surface velocity includes: Based on the time-series remote sensing images, a cross-correlation model is used to identify and track the grayscale features of the glacier surface in two remote sensing images spaced at a specified time interval, thereby obtaining the surface displacement of the glacier within the specified time. The surface flow velocity of the glacier is calculated based on the surface displacement. Based on the time series corresponding to the surface flow velocity, historical sequence data of glacier surface flow velocity is generated.

3. The method according to claim 1, characterized in that, Pre-training the deep temporal prediction model includes: Obtain glacier surface velocity sequences from historical periods when no glacier surge events occurred, and construct a training dataset; The training dataset is input into a pre-built deep time series prediction model, and the output of the deep time series prediction model is obtained, wherein the output of the deep time series prediction model includes the estimated mean of the future flow velocity sequence of the glacier; A loss function is constructed based on the difference between the output of the depth time series prediction model and the actual future glacier flow velocity sequence. The parameters of the depth time series prediction model are trained by minimizing the loss function.

4. The method according to claim 3, characterized in that, Pre-training the diffusion model includes: The estimated mean output by the depth time series prediction model is input into the pre-built diffusion model, and the following operations are performed: By using the forward process of the diffusion model, Gaussian noise is randomly added to the estimated mean to obtain noisy future sequence data of glacier surface flow velocity; The noisy glacier surface flow velocity future sequence data is used as input to the reverse process of the diffusion model, and the estimated mean is used as the conditional information of the diffusion model. The Gaussian noise component added in the forward process is predicted by the noise prediction model. A loss function is constructed based on the difference between the Gaussian noise component output by the noise prediction model and the Gaussian noise component added during the forward process. The parameters of the noise prediction model are trained by minimizing the loss function.

5. The method according to claim 4, characterized in that, The step of using the forward process of the diffusion model to randomly add Gaussian noise to the estimated mean to obtain noisy future sequence data of glacier surface flow velocity includes: Configure a variance scheduling table, wherein the variance scheduling table is used to control the rate at which Gaussian noise is added to the estimated mean and the actual future glacier flow velocity sequence during the forward process; Based on the variance scheduling table, the future sequence data of glacier surface velocity containing noise corresponding to a specified number of diffusion steps is calculated using the estimated mean, the actual future glacier velocity sequence, and randomly sampled Gaussian noise.

6. A glacier pulsation detection device based on time series analysis, characterized in that, The device includes: The acquisition module is used to acquire time-series remote sensing images of glacier flow velocity, extract the surface flow velocity of the glacier from the time-series remote sensing images, and generate historical sequence data of glacier surface flow velocity. The prediction module is used to predict the estimated mean characteristics of the future glacier flow velocity sequence based on the historical sequence data of the glacier surface flow velocity and using a pre-trained depth time series prediction model; and inputs the estimated mean characteristics into a pre-trained diffusion model, and executes the following modules: A denoising module is used to denoise the estimated mean features using the reverse process of the diffusion model to obtain the distribution features of the future flow velocity prediction sequence, including: Using the estimated mean feature as the future sequence data of noisy glacier surface velocity with the maximum diffusion step, the noise prediction model is used to predict the future sequence data of noisy glacier surface velocity corresponding to the previous step for each diffusion step, until the diffusion step is 1, and a prediction result of clean glacier surface velocity future sequence data is obtained. By repeating the above reverse process multiple times, a set of prediction results for the future sequence data of glacier surface flow velocity is obtained; Based on the prediction results of future sequence data of clean glacier surface flow velocity, the distribution characteristics of the future flow velocity prediction sequence are estimated, wherein the distribution characteristics characterize the distribution range of glacier flow velocity from the lowest to the highest speed. The detection module is used to detect the fluctuations in glacier flow velocity based on the distribution characteristics, including: Based on the aforementioned distribution characteristics, the upper bound of the glacier flow velocity is obtained according to a preset quantile, wherein the upper bound represents the maximum range of normal glacier flow velocity. The actual obtained glacier surface flow velocity sequence is compared with the upper bound. When the actual glacier flow velocity at multiple consecutive time points exceeds the upper limit of the corresponding time point, a jump in glacier flow velocity is detected.

7. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having executable instructions stored thereon, characterized in that, When this instruction is executed by the processor, it causes the processor to perform the method according to any one of claims 1 to 5.

Images