Method for fast identification of optical transient source based on multi-band light curve

Abstract

The present application relates to the technical field of transient source identification, in particular to a method for quickly identifying optical transient sources based on multi-band light curve, comprising the following steps: acquiring multi-band light variation data phase, labeling missing interval and constructing atlas, identifying trend interference structure and labeling, compensating missing section to generate continuous sequence, extracting derivative difference to construct candidate mutation index, generating response feature vector and matching type label, and outputting transient source candidate type list; in the present application, the missing section is compensated and interpolated by a Gaussian kernel function, a continuous fitting curve sequence is constructed under the coordinated labeling of brightness change direction and derivative slope, the cross-band derivative range aggregation position is extracted by a sliding time window, the light variation mutation characteristics can be captured, after the multi-dimensional response description vector is extracted, the training network is input, a stable response channel can be formed in a multi-type event concurrent scene, the identification accuracy and robustness are effectively improved, and the limitations caused by the dependence of artificial experience on the identification process are avoided.

Description

Technical Field

[0001] This invention relates to the field of transient source identification technology, and in particular to a rapid identification method for optical transient sources based on multi-band optical curves. Background Technology

[0002] The field of transient source identification technology involves the monitoring and identification of short-lived, sudden astronomical phenomena observed in astronomical observations. Specifically, it includes the detection and classification of transient events of various signal types, such as optical, electromagnetic, and gravitational waves. Core aspects of this technology include: acquiring multi-band observational data of transient sources; constructing and analyzing light curves; determining source types based on time-series data; and rapidly screening and labeling transient sources using machine learning and statistical methods. With the improvement of spatial and temporal resolution of astronomical observation equipment and the continuous growth of observational data, there is an urgent need for effective analysis methods to improve the efficiency and accuracy of transient source identification. Research in this field covers various types of astronomical events, including supernovae, blazars, gamma-ray bursts, and tidal disruption events, typically relying on the analysis of physical characteristics such as light variation, color index changes, and time scales for classification.

[0003] Traditional methods for rapid identification of optical transient sources involve analyzing the brightness variations of the transient source over time (i.e., the light curve) in the optical band to determine its type or physical mechanism. These methods primarily rely on multi-timeframe, single-band, or limited-band observation data acquired by observation equipment. They depend on manually extracting a set of empirical characteristic parameters such as peak brightness, rise and fall times, symmetry, and abrupt changes, and then comparing these parameters with preset rule templates or using manual experience for identification. Alternatively, traditional statistical methods based on single-band light curves are used for fitting, such as exponential decay models or local polynomial regression for trend estimation, to determine the event type. However, these methods are susceptible to noise interference and have limited accuracy when the data band coverage is limited or the data sampling time is uneven. Furthermore, they struggle to handle complex scenarios where multiple types of transient sources occur simultaneously. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for rapid identification of optical transient sources based on multi-band optical curves, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a rapid identification method for optical transient sources based on multi-band optical curves, comprising the following steps: S1: Obtain the observation time series from the multi-band optical variation data, convert the observation time points of each band into corresponding periodic phase values ​​within the periodic range, filter the sparse phase interval index and coverage length, and mark them in combination with the intersection relationship of phase positions between each band to generate an observation phase missing map. S2: Based on the missing sampling interval index marked in the observed phase missing sampling map, extract the differential band brightness sequence in the adjacent phase segment, calculate the first difference value of the brightness of each band, identify and mark the interference area, and generate a trend interference structure annotation set; S3: Based on the phase index in the trend interference structure annotation set, extract the observed brightness points in the adjacent periods before and after the missing segment in the target band as interpolation references, perform continuous value compensation on the missing interval, and generate a set of fitted light variation continuous segments. S4: Based on the brightness sequence and trend labeling information of the fitted continuous optical variation segment set, construct the brightness first derivative difference sequence between each band, extract the range of the first derivative difference within each time window, and aggregate the range change position as candidate recognition region to generate a cross-band change response index set. S5: Call the candidate mutation phase positions recorded in the cross-band mutation response index set, construct a response description feature vector composed of three dimensions: phase position, band difference, and change trend, input the vector into a convolutional neural network structure trained on multiple transient sources, match the output response results with the known type mapping set, and generate a list of candidate optical transient source types.

[0005] As a further aspect of the present invention, the observed phase missing sampling map includes phase segment missing sampling identifiers, band observation coverage labels, and periodic window sparsity levels; the trend interference structure label set includes band interference combination numbers, phase segment disturbance type identifiers, and slope difference transition levels; the fitted optical variation continuous segment set includes reconstructed brightness time series, phase segment interpolation confidence levels, and trend structure continuity indexes; the cross-band abrupt change response index set includes abrupt change location index, abrupt change response direction label, and abrupt change intensity grading coefficient; and the optical transient source candidate type list includes event response category numbers, type matching confidence labels, and temporal distribution pattern codes.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain the observation time series in the multi-band optical variation data, combine it with the standard reference period parameter, divide each observation time value by the period parameter and extract the remainder as the period normalized phase value, divide the obtained phase value according to the fixed length interval, count the number of observation points in each interval, calculate the difference in the number of observation points between adjacent intervals for each band, and establish a periodic phase density sequence. S102: Based on the difference in the number of observation points recorded in the periodic phase density sequence, the phase intervals are judged one by one. If the number of observation points in a continuous interval is less than the set lower limit threshold of the number of observation points, the interval index and coverage length are extracted as candidate areas for missing data. The index intersection of the phase intervals between the differentiated bands is retrieved, and the candidate areas for missing data are compared with the intersection index to generate a cross-band missing data segment index set. S103: Call the phase index in the cross-band missing segment index set, add density label to the phase interval corresponding to each missing segment, and independently identify and number the missing segments of each band, establish the mapping relationship between the band index, phase segment index and observation density label, aggregate the labeling results into unified structured spectral data, and establish the observation phase missing map.

[0007] As a further aspect of the present invention, the process of setting the lower limit threshold for the number of observation points is specifically as follows: the difference between the mean and standard deviation of the number of observation points in all intervals of the periodic phase density sequence is calculated as the lower limit threshold for the number of observation points. The process of retrieving the intersection of the indexes of phase intervals between differentiated bands is specifically as follows: performing an intersection operation on the set of phase interval indices corresponding to each band to obtain the set of phase intervals covered by the common row of the bands. The process of comparing the candidate region of missing sampling with the intersection index specifically involves matching the phase segment index in the candidate region of missing sampling with the set of intersection phase intervals one by one. If a complete inclusion relationship exists, the candidate region of missing sampling is marked as a cross-band missing sampling segment.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the missing sampling interval index marked in the observation phase missing sampling map, extract the brightness time series data of each band in the adjacent segments before and after each missing sampling phase segment, calculate the difference between adjacent observation points for each band brightness sequence in time order and divide by the time interval to obtain a first-order difference result sequence of equal length, and establish the band brightness change rate information. S202: Call the rate sequence data of each band in the brightness change rate information of the band, calculate the difference between corresponding positions of each pair of bands point by point and form a difference trajectory sequence, perform amplitude statistics on each difference trajectory sequence, determine whether the maximum difference in the continuous time window exceeds the preset brightness rate difference judgment threshold, record the time index range of the segment that meets the change condition, and obtain the rate difference change segment index set. S203: Based on the time position marked by the rate difference abrupt change segment index set, locate the corresponding phase segment index in the original phase sequence, and aggregate the band combination number matched by each abrupt change segment with the difference trajectory trend direction to generate a structured set including the abrupt change segment index, the corresponding band combination, and the jump direction code, and establish a trend interference structure annotation set.

[0009] As a further aspect of the present invention, the setting process of the brightness rate difference judgment threshold is specifically as follows: constructing a difference amplitude distribution set based on all difference data in the brightness change rate difference trajectory sequence in the target band, and using the sum of the average value and standard deviation of the difference amplitude distribution set as the brightness rate difference judgment threshold; The process of determining whether the maximum difference within a continuous time window exceeds the preset brightness rate difference judgment threshold is as follows: within the set time sliding window, the difference between adjacent sampling points is extracted, and the maximum difference within each window is retrieved. If the maximum difference is greater than the brightness rate difference judgment threshold, the index of the time segment included in the corresponding window is marked.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the phase index in the trend interference structure annotation set, locate the front and rear periodic phase boundaries of the corresponding missing segment in the target band, extract the brightness points and corresponding phase values ​​observed on both sides of each boundary, calculate the phase position difference and brightness change value between adjacent observation points, use the difference pairs as the kernel function variable input group and perform standardization processing to establish the Gaussian kernel function input parameter set; S302: Call the phase position difference in the input parameter set of the Gaussian kernel function, construct the correlation function matrix with the difference as the input variable of the kernel function, generate the covariance matrix in the form of squared exponential kernel, normalize the similarity values ​​between input points, adjust the relevant weights of the corresponding interpolation points in the covariance matrix in combination with the brightness change value, and obtain the brightness interpolation estimation result matrix. S303: Based on the phase index corresponding to the brightness interpolation estimation result matrix, reconstruct the brightness sequence trajectory of the missing segment, calculate the difference between adjacent brightness values ​​point by point on the reconstructed sequence, extract the positive and negative signs of the difference values ​​as trend direction indicators, merge the brightness values, phase index and trend signs to generate structured sequence data, and establish a set of fitted light variation continuous segments.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the brightness sequence and trend labeling information of the fitted continuous optical variation segment set, extract the brightness value of each band at the same phase index, divide the brightness difference between adjacent observation points by the corresponding time interval to form the first derivative sequence of each band, and perform first derivative value subtraction between band pairs in a time-series point correspondence manner to generate a cross-band brightness derivative difference sequence set. S402: Call each difference sequence in the cross-band brightness derivative difference sequence set, set a fixed-length sliding time window, extract the maximum and minimum values ​​of the derivative difference in each window, calculate the difference as the derivative range of the window, compare the derivative range with the set derivative abrupt change range judgment threshold, if it is greater than the threshold, record the start time index and band pair number corresponding to the window, and obtain the abrupt change candidate region index set; S403: Based on the time index and band pair number in the mutation candidate region index set, aggregate the repeated mutation positions in adjacent windows, extract stable repeating segments, and combine the corresponding position brightness change direction marker jump trend to merge the mutation phase index, corresponding band combination and direction code to form a structured record, and establish a cross-band mutation response index set.

[0012] As a further embodiment of the present invention, the threshold for determining the range of derivative mutations is set by statistically analyzing the numerical distribution of all difference points in the cross-band brightness derivative difference sequence set, and calculating the sum of the mean and standard deviation based on the distribution as the threshold for determining the range of derivative mutations. The length of the sliding time window is set by expanding it to an integer multiple based on the average phase interval of the concentrated observation points in the fitted continuous optical variation section, and limiting it to the minimum integer window covering no less than five observation points within the coverage period.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the mutation phase index recorded in the cross-band mutation response index set, extract the brightness value of each phase point in all bands and the brightness difference direction of adjacent phase points, calculate the sign of the brightness difference and combine it with the phase position as a structural element, and aggregate it according to the brightness difference between bands, phase sequence and derivative sign encoding to construct a three-dimensional structured numerical set and generate a response description feature vector set. S502: Based on each feature vector in the response description feature vector set, it is input as an input sample into the trained convolutional neural network structure. In the neural network, the convolutional channels and activation units are mapped layer by layer, and the output of each layer is normalized. The position index of the maximum response value of the corresponding category channel of the output layer is extracted and matched with the corresponding type number to obtain the category response result vector set. S503: Based on the type number index in the category response result vector set, correspond to the type label in the preset multi-type transient source label set, organize the response label corresponding to each mutation phase point and aggregate the phase segment range of the same type of label, establish a data structure of three elements: type label, label confidence score and phase index interval, and establish a candidate type list of optical transient sources.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by mapping observation time to periodic phase and establishing a missing sampling spectrum, sparse coverage sections between bands can be accurately located. By constructing the trajectory of the difference in brightness change rate and controlling the threshold, observation noise interference can be separated and trend jump features can be marked. By combining the Gaussian kernel function to perform weight compensation interpolation on the missing segments, a continuous fitting curve sequence can be constructed under the collaborative annotation of brightness change direction and derivative slope. By extracting the cross-band derivative range aggregation position through a sliding time window, the true light change abrupt change features can be captured. After the multi-dimensional response description vector is extracted and input into the training network, a stable response channel can be formed in the concurrent scenarios of multiple types of events, effectively improving the recognition accuracy and robustness, and avoiding the limitations caused by human experience dependence on the recognition process. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for rapid identification of optical transient sources based on multi-band optical curves, comprising the following steps: S1: Obtain the observation time series from the multi-band optical variation data, and convert the observation time points of each band into corresponding periodic phase values ​​within the period range by combining the standard reference period parameters. Divide the periodic phase values ​​of each band according to the interval length, count the number of observation points in each interval, and record the change amplitude of observation density between adjacent phase intervals. After establishing the phase density sequence, filter the index and coverage length of the sparsely observed phase intervals, and mark them according to the intersection relationship of the phase positions between each band to generate the observation phase missing map. S2: Based on the missing interval index marked in the observed phase missing sampling map, extract the differential band brightness sequence in the adjacent phase segment, calculate the first difference value of the brightness of each band, construct the corresponding brightness change rate sequence, calculate the difference trajectory of the rate sequence between multiple bands, and by setting the brightness change rate difference threshold, identify the time period when the rate variation amplitude continuously exceeds the brightness change rate difference threshold, mark it as the interference area, and record the phase index, band combination and jump direction sequence to generate a trend interference structure annotation set; S3: Based on the phase index in the trend interference structure annotation set, extract the observed brightness points in the adjacent periods before and after the missing segment in the target band as interpolation references, use the period phase position difference of adjacent observation points as the input variable of the Gaussian process kernel function, calculate the covariance matrix of the Gaussian kernel function in the current period, adjust the weight distribution of the interpolation points of the brightness sequence, perform continuous value compensation for the missing interval, and simultaneously annotate the brightness change direction and first derivative slope sign sequence corresponding to the current phase point during the compensation process to generate a set of fitted light variation continuous segments; S4: Based on the brightness sequence and trend labeling information of the fitted continuous optical variation segment set, construct the brightness first derivative difference sequence between each band, perform sliding time window operation, extract the range of the first derivative difference within each time window, and record the phase value, abrupt band pair and abrupt direction corresponding to the abrupt change position of the range. Aggregate the abrupt change positions that appear in multiple consecutive windows as candidate recognition regions to generate a cross-band abrupt change response index set. S5: Call the candidate mutation phase positions recorded in the cross-band mutation response index set, combine the brightness change direction and first derivative slope sign sequence in the fitted continuous optical change segment set, construct a response description feature vector composed of three dimensions: phase position, band difference and change trend, input the vector into a convolutional neural network structure trained on multiple transient sources, extract the corresponding response mode channel in the network, and match the output response results with the known type mapping set to generate a list of candidate optical transient source types; The observation phase missing sampling map includes phase segment missing sampling identifiers, band observation coverage labels, and periodic window sparsity levels. The trend interference structure label set includes band interference combination numbers, phase segment disturbance type identifiers, and slope difference transition levels. The fitted optical variation continuous segment set includes reconstructed brightness time series, phase segment interpolation confidence, and trend structure continuity index. The cross-band abrupt change response index set includes abrupt change location index, abrupt change response direction label, and abrupt change intensity grading coefficient. The optical transient source candidate type list includes event response category numbers, type matching confidence labels, and temporal distribution pattern codes.

[0023] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain the observation time series in the multi-band optical variation data, combine it with the standard reference period parameter, divide each observation time value by the period parameter and extract the remainder as the period normalized phase value, divide the obtained phase value according to the fixed length interval, count the number of observation points in each interval, calculate the difference in the number of observation points between adjacent intervals for each band, and establish a periodic phase density sequence. The system reads the raw astronomical observation logs recorded in the storage medium and parses out structured data columns containing observation time points, corresponding band numbers, and brightness magnitude values. The processing unit calls a pre-determined standard reference period parameter, derived from the dominant peak value of the Fourier transform spectrum analysis of historical long-period observation records; for example, the standard reference period parameter is set to five days. The processing unit performs a modulo operation on each discrete observation time value, specifically dividing the observation time value by the standard reference period parameter and retaining only the remainder. This remainder is then divided by the standard reference period parameter, thus converting it into a period-normalized phase value within the interval between zero and one. The processing unit sets a fixed phase interval length; for example, dividing the zero-to-one phase range into one hundred equal-length micro-intervals, each interval being 0.01 units long. The processing unit iterates through all converted period-normalized phase values, mapping them to the corresponding phase intervals and counting the total number of observation data points falling within each phase interval. After completing the statistics for each band, the processing unit calculates the difference in the number of observation points between adjacent phase intervals for each band. This is achieved by subtracting the number of observation points in the previous interval from the number in the later interval, thus generating a periodic phase density sequence that reflects the variation in observation density within that band. This sequence visually demonstrates the completeness of observation coverage and sampling non-uniformity under a specific periodic phase.

[0024] S102: Based on the difference in the number of observation points recorded in the periodic phase density sequence, the phase interval is judged one by one. If the number of observation points in a continuous interval is less than the set lower limit threshold of the number of observation points, the interval index and coverage length are extracted as candidate areas for missing data. The index intersection of the phase intervals between the differentiated bands is retrieved, and the candidate areas for missing data are compared with the intersection index to generate a cross-band missing data segment index set. The specific process for setting the lower limit threshold for the number of observation points is as follows: the difference between the mean and standard deviation of the number of observation points in all intervals of the periodic phase density sequence is calculated as the lower limit threshold for the number of observation points. The process of retrieving the intersection of the indexes of phase intervals between differentiated bands is as follows: perform an intersection operation on the set of phase interval indices corresponding to each band to obtain the set of phase intervals covered by the common row of the bands; The process of comparing the candidate region of missing sampling with the intersection index is as follows: the phase segment index in the candidate region of missing sampling is matched one by one with the intersection phase interval set. If there is a complete inclusion relationship, the candidate region of missing sampling is marked as a cross-band missing sampling segment. The processing unit receives the periodic phase density sequence and first calculates the arithmetic mean of the number of observation points corresponding to all phase intervals in the sequence, while simultaneously calculating the standard deviation of these values. The processing unit subtracts the standard deviation from the arithmetic mean, and the result is defined as the lower limit threshold for the number of observation points. For example, if the average number of observation points across all intervals in a certain band is fifty and the standard deviation is ten, then the lower limit threshold is set to forty. The processing unit then performs a traversal scan of the periodic phase density sequence. When it detects that the number of observation points in multiple consecutive phase intervals is lower than the lower limit threshold, the processing unit records the start and end index numbers of these consecutive intervals, calculates the phase length they cover, and marks them as candidate regions for low density. The processing unit performs the above operation on different bands involved, obtaining a set of candidate region indices for low density in each band. Subsequently, the processing unit performs an inter-band index intersection operation, that is, finds the phase interval indices that are simultaneously marked as low density in both band A and band B. For example, if the low-density range of band A is indexed as 5 to 10, and the low-density range of band B is indexed as 8 to 12, then the intersection index is indexed as 8 to 10. The processing unit compares the previously identified candidate areas for missing data with this intersection index set. Only when the index range of a candidate area for missing data falls entirely within the intersection index range, or highly overlaps with the intersection index, is that area finally confirmed as a cross-band missing data segment index set. This process eliminates accidental missing data that only occurs in a single band, focusing on critical phase segments that are commonly missing across multiple bands.

[0025] S103: Call the phase index in the cross-band missing segment index set, add density label to the phase interval corresponding to each missing segment, and independently identify and number the missing segments of each band, establish the mapping relationship between the band index, phase segment index and observation density label, aggregate the labeling results into unified structured spectral data, and establish the observation phase missing map; Based on the specific information in the cross-band missing data segment index set, the processing unit constructs a multi-dimensional mapping structure in memory. First, the processing unit locates the original observation phase sequence and finds the phase value range corresponding to the missing data segment index. Within these defined phase intervals, the processing unit inserts a specific density label, distinct from normal observation data points, to alert subsequent processing steps that a decrease in data confidence exists in this area. The processing unit assigns a unique identifier to each band within these specific missing data segments, combining the band ID and the missing data segment sequence number. The processing unit establishes a lookup table with the band index and phase segment index as a joint primary key, and observation density labels (such as "high density," "low density," and "missing data") as values. The processing unit aggregates these scattered labeling information to form a unified structured spectral data object. In this spectral, the horizontal axis represents phase, the vertical axis represents band, and specific areas in the spectral are highlighted as missing data states, thus establishing an intuitive observation phase missing data spectral, providing precise spatial positioning guidance for subsequent interpolation and reconstruction.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the index of the missing sampling interval marked in the observation phase missing sampling map, extract the brightness time series data of each band in the adjacent segments before and after each missing sampling phase segment, calculate the difference between adjacent observation points in the brightness sequence of each band in time order and divide it by the time interval to obtain a first-order difference result sequence of equal length, and establish the band brightness change rate information. For each missing phase segment, the processing unit traces backward and extends forward, extracting brightness time-series data from the adjacent regions before and after the missing segment. For example, if the missing segment is phase 0.4 to 0.5, the processing unit extracts data between phases 0.3 and 0.4, and 0.5 and 0.6. For each extracted brightness sequence, the processing unit sorts them according to the order of observation time. Then, the processing unit calculates the brightness difference between two adjacent observation points in the sequence and divides this difference by the time interval difference between the two observation points, thus obtaining a first-order difference result sequence reflecting the rate of brightness change. Since the original observation time may be uneven, the processing unit resamples these first-order difference results into sequences of equal length using linear interpolation to ensure data alignment between different bands. This process generates information on the rate of brightness change of bands, quantifying the instantaneous trend of celestial brightness change around the missing segment.

[0027] S202: Call the rate sequence data of each band in the brightness change rate information, calculate the difference between corresponding positions of each pair of bands point by point and form a difference trajectory sequence, perform amplitude statistics on each difference trajectory sequence, determine whether the maximum difference in the continuous time window exceeds the preset brightness rate difference judgment threshold, record the time index range of the segment that meets the abrupt change condition, and obtain the rate difference abrupt change segment index set. The specific process of setting the brightness rate difference judgment threshold is as follows: construct a set of difference amplitude distributions based on all difference data in the brightness change rate difference trajectory sequence in the target band, and use the sum of the average value and standard deviation of the difference amplitude distribution set as the brightness rate difference judgment threshold. The process of determining whether the maximum difference within a continuous time window exceeds the preset brightness rate difference judgment threshold is as follows: within the set time sliding window, the difference between adjacent sampling points is extracted, and the maximum difference within each window is retrieved. If the maximum difference is greater than the brightness rate difference judgment threshold, the index of the time segment included in the corresponding window is marked. The processing unit retrieves the brightness change rate information for each band. For any two different bands involved in the target observation (e.g., band R and band G), it calculates the numerical difference of their rate sequences at each time point. Specifically, it subtracts the brightness change rate of band G at the same time from the brightness change rate of band R at a given moment, forming a continuous difference trajectory sequence. The processing unit then analyzes the absolute value distribution of all data points in this difference trajectory sequence, calculating the average and standard deviation of these difference amplitudes. The processing unit adds the average and standard deviation, and sets the sum as the brightness rate difference judgment threshold. For example, if the average difference amplitude is 0.5 star magnitude per day and the standard deviation is 0.2 star magnitude per day, then the threshold is set to 0.7 star magnitude per day. The processing unit sets a fixed-length time sliding window, for example, covering five sampling points. The processing unit slides this window across the difference trajectory sequence, retrieving the maximum value of all difference points within the window at each window position. If the maximum difference within a window exceeds the aforementioned brightness rate difference judgment threshold, the processing unit records the time segment index corresponding to that window. The processing unit merges and organizes all windows that meet the conditions to obtain an index set of rate difference abrupt change segments. This set indicates the time regions where the brightness change behavior between different bands is significantly separated.

[0028] S203: Based on the time position marked by the rate difference abrupt change segment index set, locate the corresponding phase segment index in the original phase sequence, and aggregate the band combination number matched by each abrupt change segment with the difference trajectory trend direction to generate a structured set including the abrupt change segment index, the corresponding band combination, and the jump direction code, and establish a trend interference structure annotation set. The processing unit maps the temporal information from the rate difference abrupt change segment index back to the original period-normalized phase sequence, thereby determining the specific phase segment in which the abrupt change occurred. The processing unit identifies band combinations exhibiting differences within the abrupt change segment (e.g., "band R - band G") and analyzes the trajectory of the difference. If the difference is positive and continuously increasing, it is marked as "positive divergence"; if the difference is negative and continuously decreasing, it is marked as "negative divergence". The processing unit packages and combines the start and end phase indices of the abrupt change segment, the involved band combination numbers, and the jump direction encoding (e.g., using the number 1 to represent positive and -1 to represent negative). This information is stored in a structured dataset, establishing a trend interference structure annotation set. This annotation set not only records when and where the abrupt change occurred but also qualitatively describes the morphology of the abrupt change, providing constraints for subsequent targeted interpolation corrections.

[0029] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the phase index in the trend interference structure annotation set, locate the front and rear periodic phase boundaries of the corresponding missing segment in the target band, extract the brightness points and corresponding phase values ​​observed on both sides of each boundary, calculate the phase position difference and brightness change value between adjacent observation points, use the difference pairs as the kernel function variable input group and perform standardization processing to establish the Gaussian kernel function input parameter set; The processing unit locates the preceding and following valid observation boundary points on the phase axis for the missing segment. It extracts the observed brightness values ​​and corresponding precise phase values ​​at these two boundary points. The processing unit calculates the difference between the phase value of the following boundary point and the phase value of the preceding boundary point as the phase position difference; simultaneously, it calculates the difference between the brightness value of the following boundary point and the brightness value of the preceding boundary point as the brightness change value. The processing unit combines these two differences into a data pair, which serves as the input variable for the Gaussian kernel function. To eliminate the influence of dimensions, the processing unit standardizes the input variable by subtracting the global mean of that variable type and dividing by the global standard deviation, thus centering the data distribution. The processing unit stores the standardized data pair into the Gaussian kernel function input parameter set, ready to construct a kernel matrix describing the correlation of the data.

[0030] S302: Call the phase position difference in the input parameter set of the Gaussian kernel function, construct the correlation function matrix with the difference as the input variable of the kernel function, generate the covariance matrix in the form of squared exponential kernel, normalize the similarity values ​​between input points, and adjust the relevant weights of the corresponding interpolation points in the covariance matrix in combination with the brightness change value to obtain the brightness interpolation estimation result matrix. The processing unit uses a squared exponential kernel function as the core for calculating covariance. Specifically, for any two elements in the matrix, the processing unit calculates the square of the Euclidean distance between their corresponding phase positions, divides this squared value by the square of a preset length scale parameter (e.g., the length scale is set to 0.1), takes the negative sign, and calculates the exponential function value to obtain the basic correlation weight between the two points. Subsequently, the processing unit normalizes these weight values ​​so that their sum is one. The processing unit further introduces the brightness change value as an adjustment factor. If the brightness change value between two points is large, the basic correlation weight is multiplied by a decay coefficient less than one (e.g., 0.8) to reflect the decrease in correlation caused by drastic changes. The adjusted weight matrix constitutes the final covariance matrix. The processing unit uses this covariance matrix to perform matrix multiplication with the brightness values ​​of known boundary points to calculate the expected brightness value of the unknown phase points in the missing segment, thereby obtaining the brightness interpolation estimation result matrix.

[0031] S303: Based on the phase index corresponding to the brightness interpolation estimation result matrix, reconstruct the brightness sequence trajectory of the missing segment, calculate the difference between adjacent brightness values ​​point by point on the reconstructed sequence, extract the positive and negative signs of the difference values ​​as trend direction indicators, merge the brightness values, phase index and trend signs to generate structured sequence data, and establish a set of fitted continuous light variation segments. The processing unit, based on the predicted brightness values ​​recorded in the brightness interpolation estimation matrix, maps them to each discrete phase index within the missing segment, and plots a completed brightness sequence trajectory. Following this reconstructed trajectory, the processing unit calculates the difference between adjacent brightness values ​​point by point, starting from the starting point. The processing unit extracts the sign of each difference result: a plus sign for a positive difference, a minus sign for a negative difference, and zero for zero. This sequence of signs constitutes the trend direction indicator. The processing unit merges the reconstructed brightness values, corresponding phase indices, and trend direction indicators column-wise, generating a structured sequence containing three columns of data. This sequence fills the gaps in the original data and includes metadata about local change trends, establishing a set of fitted continuous light variation segments and achieving temporal completeness of the light variation curve.

[0032] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the brightness sequence and trend labeling information of the fitted continuous optical variation segment set, extract the brightness value of each band at the same phase index, divide the brightness difference between adjacent observation points by the corresponding time interval to form the first derivative sequence of each band, and perform first derivative value subtraction between band pairs in the manner of corresponding time points to generate a cross-band brightness derivative difference sequence set. The processing unit reads the fitted continuous optical variation segment set and extracts the brightness values ​​at the same phase index position for each band. The processing unit performs first-order derivative calculations on the brightness sequence of each band, i.e., subtracting the brightness value of the previous point from the current point's brightness value and then dividing by the time interval between the two, generating a first-order derivative sequence. Subsequently, the processing unit performs synchronous subtraction operations between different bands (e.g., band A and band B). The processing unit ensures that the first-order derivative values ​​of band A and band B are included in the calculation at the same time point, subtracting the derivative value of band B from the derivative value of band A to obtain a new sequence data. This process is repeated for all possible band combinations, generating multiple sets of cross-band brightness derivative difference sequences. These sequences reflect the evolution of differences in the rate of brightness change across different bands over time, and can sensitively capture asynchronous bursts or decay behaviors across bands.

[0033] S402: Call each difference sequence in the cross-band brightness derivative difference sequence set, set a fixed-length sliding time window, extract the maximum and minimum values ​​of the derivative difference in each window, calculate the difference as the derivative range of the window, compare the derivative range with the set derivative abrupt change range judgment threshold, if it is greater than the threshold, record the start time index and band pair number corresponding to the window, and obtain the abrupt change candidate region index set; The threshold for determining the range of derivative mutations is set by statistically analyzing the numerical distribution of all difference points in the cross-band brightness derivative difference sequence set, and using the sum of the mean and standard deviation calculated based on the distribution as the threshold for determining the range of derivative mutations. The length of the sliding time window is set by expanding it to an integer multiple based on the average phase interval of the concentrated observation points in the fitted continuous optical variation section, and limiting it to the minimum integer window of no less than five observation points within the coverage period. For each cross-band brightness derivative difference sequence, a fixed-length sliding time window is defined. The window length is determined by multiplying the average phase interval (e.g., 0.02 phase units) of all observation points in the fitted continuous light variation segment set by five and rounding up to ensure the window covers at least five observation points. Within each sliding window, the processor identifies the maximum and minimum derivative difference values ​​and calculates the difference, defined as the derivative range. The processor then statistically analyzes the derivative ranges calculated for all windows, calculating their mean and standard deviation. The sum of the mean and standard deviation is set as the threshold for determining the derivative abrupt change range. For example, if the mean is 2.0 and the standard deviation is 0.5, the threshold is 2.5. The processor compares the derivative range of each window with this threshold. If the range is greater than the threshold, it indicates that the derivative difference within that window fluctuates drastically. The processor records the start time index and the corresponding band pair number of that window and adds it to the abrupt change candidate region index set.

[0034] S403: Based on the time index and band pair number in the candidate mutation region index set, aggregate the repeated mutation positions in adjacent windows, extract stable repeating segments, and combine the brightness change direction markers of the corresponding positions to mark the jump trend. Merge the mutation phase index, corresponding band combination and direction code to form a structured record and establish a cross-band mutation response index set. The processor examines all recorded window time indices. If adjacent window indices are found to be consecutive or overlapping, the processor merges these windows into a continuous, stable, repeating segment. The processor then returns to the original brightness data corresponding to this segment and observes its direction of change. If the brightness of the involved bands shows an upward trend within this segment, the processor marks the abrupt change as a "brightening abrupt change"; otherwise, it marks it as a "darkening abrupt change." The processor integrates the merged abrupt change phase index range, the band combination triggering the abrupt change, and the trend direction encoding (e.g., using the letters U to represent an increase and D to represent a decrease) into a single record. These records constitute a cross-band abrupt change response index set, concisely summarizing all significant cross-band anomalous behaviors in the light curve, providing core features for the final physical classification.

[0035] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the mutation phase index recorded in the cross-band mutation response index set, extract the brightness value of each phase point in all bands and the brightness difference direction of adjacent phase points, calculate the sign of the brightness difference and combine it with the phase position as a structural element, and aggregate it according to the brightness difference between bands, phase sequence and derivative sign encoding to construct a three-dimensional structured numerical set and generate a response description feature vector set. For each index point, the processor retrieves its brightness value from the raw or fitted data across all bands. Simultaneously, it calculates the brightness difference between that point and the previous phase point, and determines the derivative sign encoding based on the sign of the difference (e.g., positive is represented by the number 1, negative by the number 0). The processor combines the brightness differences between bands (numerical features), the current phase's position within the period (positional features), and the derivative sign encoding (morphological features) to form a basic structural element. The processor then arranges these elements in three dimensions according to band and time order, constructing a multidimensional tensor. This tensor is the response description feature vector, comprehensively encoding the spectral, temporal, and morphological attributes of the abrupt change event, serving as the input to the machine learning model.

[0036] S502: Based on each feature vector in the response description feature vector set, it is fed into the trained convolutional neural network structure as an input sample. In the neural network, the convolutional channels and activation units are mapped layer by layer, and the output of each layer is normalized. The position index of the maximum response value of the corresponding category channel of the output layer is extracted and matched with the corresponding type number to obtain the category response result vector set. The processing unit inputs the constructed response description feature vector into a pre-trained convolutional neural network model. This network contains multiple convolutional layers, each composed of several convolutional kernels. Inside the network, the input vector first passes through the first convolutional layer, where the kernels slide across the vector and perform a weighted summation operation, extracting preliminary local features. The output is then non-linearly mapped by activation units (such as rectified linear units) to enhance the model's expressive power. Subsequently, the data flows through subsequent convolutional and pooling layers, where features are abstracted layer by layer. At the output of each layer, the processing unit performs normalization to constrain the numerical range within a reasonable interval, preventing gradient explosion. Finally, the data reaches the fully connected output layer, where each neuron corresponds to a predefined category. The processing unit compares the activation values ​​of all neurons in the output layer, finds the index of the location of the maximum response value, and matches the corresponding internal type number based on this index to obtain the category response result vector set.

[0037] S503: Based on the type number index in the category response result vector set, correspond to the type label in the preset multi-type transient source label set, organize the response label corresponding to each mutation phase point and aggregate the phase segment range of the same type of label, establish a data structure of three elements: type label, label confidence score and phase index interval, and establish a candidate type list of optical transient sources. The system receives the type number index from the category response result vector set and uses it as a key to query a pre-defined multi-category transient source label database. This database stores specific physical type labels such as "Type Ia supernova," "core collapse supernova," and "tidal disintegration event." The processing unit associates the retrieved text labels with the original abrupt change phase points. To improve the robustness of the results, the processing unit performs aggregation analysis on the labels obtained from adjacent phase points. If points within a continuous phase segment are classified into the same type, the processing unit merges this segment into a defined phase index interval and calculates the average classification confidence score of all points within this interval as the label confidence score. The processing unit ultimately constructs a data structure containing three core elements: "inferred type label," "comprehensive confidence score," and "effective phase index interval," generating a detailed list of candidate optical transient source types, thus completing the automated identification process from raw light variation data to specific astrophysical types.

[0038] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for rapid identification of optical transient sources based on multi-band optical curves, characterized in that, Includes the following steps: S1: Obtain the observation time series from the multi-band optical variation data, convert the observation time points of each band into corresponding periodic phase values ​​within the periodic range, filter the sparse phase interval index and coverage length, and mark them in combination with the intersection relationship of phase positions between each band to generate an observation phase missing map. S2: Based on the missing sampling interval index marked in the observed phase missing sampling map, extract the differential band brightness sequence in the adjacent phase segment, calculate the first difference value of the brightness of each band, identify and mark the interference area, and generate a trend interference structure annotation set; S3: Based on the phase index in the trend interference structure annotation set, extract the observed brightness points in the adjacent periods before and after the missing segment in the target band as interpolation references, perform continuous value compensation on the missing interval, and generate a set of fitted light variation continuous segments. S4: Based on the brightness sequence and trend labeling information of the fitted continuous optical variation segment set, construct the brightness first derivative difference sequence between each band, extract the range of the first derivative difference within each time window, and aggregate the range change position as candidate recognition region to generate a cross-band change response index set. S5: Call the candidate mutation phase positions recorded in the cross-band mutation response index set, construct a response description feature vector composed of three dimensions: phase position, band difference, and change trend, input the vector into a convolutional neural network structure trained on multiple transient sources, match the output response results with the known type mapping set, and generate a list of candidate optical transient source types.

2. The method for rapid identification of optical transient sources based on multi-band optical curves according to claim 1, characterized in that, The observed phase missing sampling map includes phase segment missing sampling identifiers, band observation coverage labels, and periodic window sparsity levels. The trend interference structure label set includes band interference combination numbers, phase segment disturbance type identifiers, and slope difference transition levels. The fitted optical variation continuous segment set includes reconstructed brightness time series, phase segment interpolation confidence, and trend structure continuity index. The cross-band abrupt change response index set includes abrupt change location index, abrupt change response direction label, and abrupt change intensity grading coefficient. The optical transient source candidate type list includes event response category numbers, type matching confidence labels, and temporal distribution pattern codes.

3. The method for rapid identification of optical transient sources based on multi-band optical curves according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the observation time series in the multi-band optical variation data, combine it with the standard reference period parameter, divide each observation time value by the period parameter and extract the remainder as the period normalized phase value, divide the obtained phase value according to the fixed length interval, count the number of observation points in each interval, calculate the difference in the number of observation points between adjacent intervals for each band, and establish a periodic phase density sequence. S102: Based on the difference in the number of observation points recorded in the periodic phase density sequence, the phase intervals are judged one by one. If the number of observation points in a continuous interval is less than the set lower limit threshold of the number of observation points, the interval index and coverage length are extracted as candidate areas for missing data. The index intersection of the phase intervals between the differentiated bands is retrieved, and the candidate areas for missing data are compared with the intersection index to generate a cross-band missing data segment index set. S103: Call the phase index in the cross-band missing segment index set, add density label to the phase interval corresponding to each missing segment, and independently identify and number the missing segments of each band, establish the mapping relationship between the band index, phase segment index and observation density label, aggregate the labeling results into unified structured spectral data, and establish the observation phase missing map.

Images