Method and device for feature extraction and classification of spatial target luminosity signals

Abstract

The present application relates to target recognition technical field, especially a kind of space target luminosity signal feature extraction and classification method, device.Method includes: the luminosity signal of space target to be handled is preprocessed, and the luminosity signal after preprocessing is obtained;The luminosity signal after preprocessing is down-sampled based on multiple preset sampling intervals, and the luminosity signal of different time resolution is obtained;The cumulative quantity feature of preset number of the luminosity signal of different time resolution is calculated;From the cumulative quantity feature of preset number, the optimal feature subset calculated in advance is extracted, and the optimal feature subset is calculated based on the luminosity signal set of multiple known categories target, and space target to be handled is one kind in known category;Space target to be handled is classified based on optimal feature subset, and the category of space target to be handled is determined.The present application method can accurately extract the effective feature in space target luminosity signal, and space target is accurately classified based on the feature extracted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target recognition, in particular to a feature extraction and classification method and device for space target photometric signals. BACKGROUND

[0002] Space targets include satellites, spaceships, rocket debris and fragments, etc. Such space targets usually orbit around the earth, and the detection, tracking, characteristic analysis and identification of such targets are one of the main tasks of space monitoring systems.

[0003] At present, space target photometric feature extraction is a hot research direction in the space field. The photometric signal curve is affected by many factors such as observation position, target orbit, attitude, surface material, illumination, etc., and contains a large amount of feature information of space targets. The original photometric curve is not suitable for direct analysis of space targets. On the one hand, the dimension is very high, which is not suitable for the design of the classifier, and on the other hand, such direct description cannot reflect the essential characteristics of the target. Therefore, some descriptions reflecting the essence of the target must be found from the original photometric curve, which is called feature extraction. The result of feature extraction should be "less and precise", and the dimension of feature data should be as low as possible, while effectively reflecting the discriminability between different categories.

[0004] In related technologies, the method of inversion is mainly used to analyze the characteristics of space targets. Such method analyzes the influence of different characteristic parameters on the photometric signal observation results by constructing a ground-based observation model and a motion model of space targets, and compares the measured signal with the simulated signal, so as to infer the characteristics and corresponding categories of space targets. However, the inversion method is affected by factors such as model accuracy and observation noise, and the applicable conditions are relatively harsh, and the analysis result is not high enough in accuracy.

[0005] Therefore, there is an urgent need for a feature extraction and classification method and device for space target photometric signals to solve the above technical problems. SUMMARY

[0006] The embodiments of the present application provide a feature extraction and classification method and device for space target photometric signals, which can accurately extract effective features in the space target photometric signals, and accurately classify space targets based on the extracted features.

[0007] In a first aspect, the embodiments of the present application provide a feature extraction and classification method for space target photometric signals, comprising:

[0008] Preprocessing the photometric signal of the space target to be processed to obtain a preprocessed photometric signal;

[0009] Down-sampling the preprocessed photometric signal based on a plurality of preset sampling intervals to obtain photometric signals with different time resolutions;

[0010] calculate a preset number of accumulation quantity features of the photometric signals with different time resolutions;

[0011] extract a pre-calculated optimal feature subset from the preset number of accumulation quantity features, the optimal feature subset being calculated based on a set of photometric signals of a plurality of known category targets, and the spatial target to be processed being one of the known categories;

[0012] classify the spatial target to be processed based on the optimal feature subset, and determine the category of the spatial target to be processed.

[0013] In a possible design, the pre-processing of the photometric signal of the spatial target to be processed to obtain the pre-processed photometric signal comprises:

[0014] processing of outliers in the photometric signal of the spatial target to be processed;

[0015] suppression of long-time noise in the photometric signal that has been processed for outliers;

[0016] mapping of the photometric signal after suppression of long-time noise to a preset observation distance to obtain the pre-processed photometric signal.

[0017] In a possible design, the calculation of the preset number of accumulation quantity features of the photometric signals with different time resolutions comprises:

[0018] for each time resolution of the photometric signal, calculating a preset order of moment features of the photometric signal of the current time resolution; and based on the preset order of moment features, calculating the preset order of accumulation quantity features of the photometric signal of the current time resolution;

[0019] collecting the preset order of accumulation quantity features of each time resolution of the photometric signal to obtain the preset number of accumulation quantity features of the photometric signals with different time resolutions.

[0020] In a possible design, the optimal feature subset is calculated based on the following method:

[0021] obtaining a set of photometric signals of a plurality of known category targets;

[0022] calculating a preset number of accumulation quantity features of each photometric signal in the set of photometric signals;

[0023] based on the preset number of accumulation quantity features of each photometric signal, determining a plurality of feature subsets, each of the feature subsets containing at least one of the accumulation quantity features, and each of the feature subsets containing different accumulation quantity features;

[0024] For each of the feature subsets, a probability of each photometric signal belonging to any of the known categories is calculated, and a score of each feature subset is determined based on a ranking of the probability of each photometric signal belonging to the true category in the ranking of all probabilities.

[0025] The feature subset with the highest score is determined as the optimal feature subset.

[0026] In a possible design, the score of each feature subset is determined based on the probability of each photometric signal belonging to the true category in the ranking of all probabilities, including:

[0027] For each of the photometric signal sets corresponding to the known category targets, the following is performed:

[0028] For each feature in the feature subset, a mean value, a variance, and a distribution function of the current feature of each photometric signal in the current category are calculated; a prior distribution probability of the current photometric signal is determined based on the mean value, the variance, and the distribution function of each feature; and a probability of the current photometric signal belonging to any of the known categories is calculated based on the prior distribution probability of the current photometric signal, where the true category of the current photometric signal is known.

[0029] The score of the current feature subset is determined based on a position of the probability of each photometric signal belonging to the true category in the ranking of all probabilities.

[0030] In a possible design, the score of the current feature subset is determined based on the position of the probability of each photometric signal belonging to the true category in the ranking of all probabilities, including:

[0031] A total ranking is determined according to the size of the probability value, and the position of the probability of the current photometric signal belonging to the true category in the total ranking is determined;

[0032] Based on a preset position scoring criterion, the score of the current feature subset in distinguishing the current photometric signal is determined according to the position.

[0033] The scores of the current feature subset in distinguishing all photometric signals are added to obtain the score of the current feature subset.

[0034] In a possible design, the class of the to-be-processed space target is determined based on the optimal feature subset, including:

[0035] The prior distribution probability of the photometric signal of the to-be-processed space target is calculated based on the optimal feature subset.

[0036] Based on the prior distribution probability, a probability that the photometric signal belongs to any of the known category targets is calculated according to a Bayesian posterior probability formula.

[0037] The category with the highest probability is determined as the category of the space target to be processed.

[0038] In a second aspect, an embodiment of the present application further provides a feature extraction and classification device for a photometric signal of a space target, comprising:

[0039] A preprocessing module is configured to preprocess a photometric signal of a space target to be processed to obtain a preprocessed photometric signal.

[0040] A sampling module is configured to down-sample the preprocessed photometric signal based on a plurality of preset sampling intervals to obtain photometric signals with different time resolutions.

[0041] A first calculation module is configured to calculate cumulative quantity features of a preset number of the photometric signals with different time resolutions.

[0042] An extraction module is configured to extract a pre-calculated optimal feature subset from the preset number of the cumulative quantity features, the optimal feature subset being calculated based on a photometric signal set of a plurality of known category targets, and the space target to be processed being one of the known categories.

[0043] A determination module is configured to classify the space target to be processed based on the optimal feature subset to determine a category of the space target to be processed.

[0044] In a third aspect, an embodiment of the present application further provides an electronic device, comprising a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in any of the embodiments of the present application.

[0045] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium storing a computer program, and the computer program, when executed in a computer, causes the computer to execute the method described in any of the embodiments of the present application.

[0046] The embodiment of the present application provides a feature extraction and classification method of a space target luminosity signal. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0048] Figure 1 FIG. 1 is a flow chart of a feature extraction and classification method of a space target luminosity signal according to an embodiment of the present application;

[0049] Figure 2 FIG. 2 is a hardware architecture diagram of an electronic device according to an embodiment of the present application;

[0050] Figure 3 FIG. 3 is a structure diagram of a feature extraction and classification device of a space target luminosity signal according to an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present application.

[0052] Please refer to Figure 1 The embodiment of the present application provides a feature extraction and classification method of a space target luminosity signal, which comprises the following steps.

[0053] Step 100, pre-processing the photometric signal of the space target to be processed to obtain a pre-processed photometric signal;

[0054] Step 102, downsampling the pre-processed photometric signal based on a plurality of preset sampling intervals to obtain photometric signals with different time resolutions;

[0055] Step 104, calculating a preset number of cumulative quantity features of the photometric signals with different time resolutions;

[0056] Step 106, extracting a pre-calculated optimal feature subset from the preset number of cumulative quantity features, the optimal feature subset being calculated based on a plurality of known category target photometric signal sets, and the space target to be processed being one of the known categories;

[0057] Step 108, classifying the space target to be processed based on the optimal feature subset to determine the category of the space target to be processed.

[0058] This embodiment first obtains photometric signals with different time resolutions through downsampling to represent the overall trend of photometric changes using low-resolution signals and represent high-frequency and short-time information using high-resolution signals, and since the statistics between different resolutions are approximately independent, feature extraction based on the signals obtained through downsampling can obtain more target information. Then, the calculation of the cumulative quantity features of the photometric signals can describe the nonlinear characteristics of the signals. For photometric signals, more features are not necessarily better, and more features sometimes hinder the description of the target, so the present application calculates an optimal feature subset that can accurately represent a plurality of category targets based on a plurality of known category target photometric signal sets, uses the optimal feature subset to effectively and accurately represent target information, and then uses the feature subset to classify the space target to be processed, which can achieve accurate classification of the space target.

[0059] The execution of each step will be described in detail below. Figure 1

[0060] First, for step 100, the photometric signal of the space target to be processed is pre-processed to obtain a pre-processed photometric signal.

[0061] In some embodiments, the pre-processing process includes:

[0062] A1, processing outliers and noise in the photometric signal of the space target to be processed;

[0063] A2, suppressing long-time noise in the photometric signal that has been processed for outliers and noise;

[0064] A3, mapping the photometric signal after suppressing long-time noise to a preset observation distance to obtain a pre-processed photometric signal.​

[0065] In step A1, the wild value noise has a large amplitude and only appears at individual points or small areas in the time domain, which can be effectively identified and repaired. The wild value noise types include zero value and strong spike noise, wherein the zero value refers to a phenomenon that the observation signal suddenly returns to zero and then quickly recovers, and the strong spike noise refers to a phenomenon that the observation signal suddenly rises or falls and then quickly recovers.

[0066] In step A2, the long-time noise in the original signal is suppressed, and the noise includes step noise and full-time domain superimposed noise. The step noise refers to a step phenomenon existing at both ends of a part of the signal, that is, the signal amplitude is greatly increased or decreased as a whole. In this embodiment, the method of setting a gradient threshold is used for discrimination. The full-time domain noise is generally distributed in the high-frequency region, and the observation signal of this type of noise needs to be quantitatively evaluated. In this embodiment, a low-pass filter is reasonably designed for processing, so as to filter out the high-frequency noise as much as possible while retaining the characteristics of the original signal.

[0067] In step A3, the photometric signal is normalized to map the original photometric signal to a preset observation distance. This step needs to calculate the shortest observation distance (i.e., the actual observation distance) according to the observation position, the azimuth angle track and the pitch angle track, and then convert the original signal to a signal at the same observation distance according to the quantitative relationship between the distance and the photometric intensity. Let the photometric signal processed by step A2 be B=[b1,b2,...,b n ], and the normalized photometric signal be A=[a1,a2,...,a n ], and the specific calculation formula is as follows:

[0068] a i =b i -5log 10 (s / S)

[0069] In the formula, S is the preset distance, s is the shortest observation distance, a i and b i represent the i-th signal in the photometric signals A and B, respectively.

[0070] Then, for step 102, the preprocessed photometric signal is down-sampled based on a plurality of preset sampling intervals, to obtain photometric signals with different time resolutions.

[0071] The photometric sequence with different time resolutions reflects different characteristics. Generally, the low-resolution signal is more suitable for representing the overall trend of photometric change, and the high-resolution signal shows more high-frequency and short-time information. The statistics between different resolutions are approximately independent, and the statistics within each resolution also maintain independence as much as possible, thereby facilitating the subsequent design of the classifier.

[0072] In this step, the number of sampling intervals and the interval of each sampling interval are set as required, and the application does not make specific limitations. For example, the sampling interval can be 4, and the interval is [0.1, 1, 6, 30] seconds, that is, the first sampling interval is sampled once every 0.1 seconds, the second sampling interval is sampled once every 1 second, the third sampling interval is sampled once every 6 seconds, and the fourth sampling interval is sampled once every 30 seconds. The sampling interval gradually increases, thereby obtaining photometric signals with different time resolutions, which can represent more comprehensive information of the target.

[0073] Then, for step 104, the cumulative quantity features of the preset number of photometric signals with different time resolutions are calculated, including:

[0074] For each time resolution photometric signal, the moment features of the preset order of the current time resolution photometric signal are calculated; based on the moment features of the preset order, the cumulative quantity features of the preset order of the current time resolution photometric signal are calculated:

[0075] The cumulative quantity features of the preset order of each time resolution photometric signal are collected to obtain the cumulative quantity features of the preset number of photometric signals with different time resolutions.

[0076] It can be understood that for multiple signals of the same category, the same feature value should theoretically tend to be consistent, and have a unimodal symmetric structure in form. Since the high-order cumulative quantity of the signal is always zero for Gaussian colored noise, it can be used to extract non-Gaussian signals. In addition, the high-order cumulative quantity contains the phase information of the system and can accurately describe the nonlinear characteristics of the signal. Therefore, the embodiment uses Gaussian distribution to describe the feature distribution of the signal, and uses high-order cumulative quantity as the feature of the photometric signal.

[0077] In this embodiment, the preset order is 4 times, and the specific calculation process is as follows:

[0078] First, for each time resolution photometric signal, the fourth-order moments m1-m4 of the current time resolution photometric signal are calculated, where:

[0079]

[0080]

[0081]

[0082]

[0083] In the formula, a i is the i-th signal in the photometric signal, and n is the number of signals of the current time resolution photometric signal.

[0084] Then, according to the calculated fourth-order moments, fourth-order cumulant features C1-C4 of the current time resolution photometric signal are calculated, wherein,

[0085] C1 = m1

[0086]

[0087]

[0088]

[0089] As can be seen, four cumulant features can be obtained for each time resolution photometric signal, and the above operation is performed on each time resolution photometric signal to obtain a preset number of cumulant features. For example, when the down-sampling time interval is 3, there are three different time resolution photometric signals, and the preset number is 3 x 4 = 12.

[0090] It should be noted that the preset order and the down-sampling time interval can be set as needed, and the present application does not make specific limitations.

[0091] For step 106, a pre-calculated optimal feature subset is extracted from the preset number of cumulant features, and the optimal feature subset is calculated based on a photometric signal set of a plurality of known category targets, and the space target to be processed is one of the known categories.

[0092] When performing target recognition, the more features used, the better the classification effect, therefore, an optimal feature subset needs to be extracted from all features, and target classification is performed using the optimal feature subset, which can obtain the most accurate result.

[0093] It can be understood that the optimal feature subset is different for different categories of targets. For example, when distinguishing between a football and an American football, only one feature, shape, is needed (that is, shape is the optimal feature subset), and when color, material and other features are added, the classification difficulty is increased and the classification accuracy is reduced. When distinguishing between men and women, the combination of hair length, height and bust may be the optimal feature subset. As can be seen, when determining the optimal feature subset, statistics need to be performed based on a photometric signal set of a plurality of known category targets. Of course, the target to be processed needs to be one of the known category targets. In the present application, in the same observation area and the same observation time period, the types of space targets that usually appear are relatively fixed, such as satellites, rocket debris, spaceships, etc., therefore, it is feasible and effective to determine the optimal feature subset based on the photometric signal set of the known category targets, and then use the optimal feature subset to extract features and classify unknown targets.

[0094] In some embodiments, the optimal feature subset is calculated based on the following method:

[0095] B1, obtaining a plurality of sets of photometric signals of known category targets;

[0096] B2, calculating a preset number of cumulative quantity features of each photometric signal in the set of photometric signals;

[0097] B3, determining a plurality of feature subsets based on the preset number of cumulative quantity features of each photometric signal, each feature subset containing at least one cumulative quantity feature, and each feature subset containing different cumulative quantity features;

[0098] B4, for each feature subset, calculating the probability of each photometric signal belonging to any category in the known category, and determining the score of each feature subset based on the ranking of the probability of each photometric signal belonging to the true category in all probabilities;

[0099] B5, determining the feature subset with the highest score as the optimal feature subset.

[0100] For step B1, when the set of photometric signals is the original obtained signal, preprocessing is also needed, including processing the wild value noise in each photometric signal, suppressing the long time noise in the photometric signal after wild value noise processing, and mapping the photometric signal after long time noise suppression to a preset observation distance to obtain the preprocessed photometric signal. In addition, when the set of photometric signals is the original obtained signal, each photometric signal can also be down-sampled to obtain photometric signals with different time resolutions.

[0101] For step B2, the calculation process is the same as that of step 106, and this application will not be repeated.

[0102] For step B3, let the preset number be L, that is, each photometric signal has L cumulative quantity features, then the following two methods can be used to determine the feature subset.

[0103] The first one is that the number of non-empty subsets is 2 K -1, this method can contain all feature subsets, but when L is large, the one-by-one traversal evaluation of the discrimination effect of each feature subset will reduce the calculation efficiency of the training algorithm, the advantage is that it will not miss any subset, and the calculation accuracy is high.

[0104] Secondly, start from a subset containing all features, and reduce one feature each time, compare the evaluation results of the L+1 subsets obtained, determine the first subset with the best distinguishing effect according to the evaluation results, then reduce one feature from the first subset each time to obtain L-2 subsets, compare the evaluation results of the L-2 subsets obtained, determine the second subset with the best distinguishing effect according to the evaluation results, then reduce one feature from the second subset each time, and so on, until the features are no longer reduced. This method can improve the calculation efficiency, and some better subsets may be missed. The determination method of the feature subset is not limited in the present application.

[0105] For step B4, in some embodiments, the specific execution process includes:

[0106] For each set of luminosity signals corresponding to a known category target, the following are performed:

[0107] For each feature in the feature subset, the mean value, variance and distribution function of the current feature of each luminosity signal in the current category are calculated; based on the mean value, variance and distribution function of each feature, the prior distribution probability of the current luminosity signal is determined; based on the prior distribution probability of the current luminosity signal, the probability that the current luminosity signal belongs to any category in the known categories is calculated respectively, wherein the true category of the current luminosity signal is known;

[0108] Based on the position of the probability that each luminosity signal belongs to the true category in the total probability ranking, the score of the current feature subset is determined.

[0109] In this step, for each feature in the feature subset, the calculation formula of the mean value, variance and distribution function of the current feature of each luminosity signal in the current category is:

[0110]

[0111] In the formula, q is the number of luminosity samples of the current category signal, x is any feature in the preset number of accumulated features, x i is the x feature value corresponding to the i th luminosity sample, μ is the mean value of the feature x, σ is the variance of the feature x, and f(x) is the distribution function of the feature x.

[0112] In some embodiments, the calculation formula of the prior distribution probability of the current luminosity signal based on the mean value, variance and distribution function of each feature is:

[0113]

[0114] In the formula, f i (A) represents the prior distribution probability of the luminosity signal A in the i th known target, L is the number of features in the current feature subset, σi，j is the variance of the jth feature in the ith known target, u i，j is the mean of the jth feature in the ith known target, x j is the jth feature value of the photometric signal A.

[0115] In some embodiments, the calculation formula of the probability that the current photometric signal belongs to any one of the known categories based on the prior distribution probability of the current photometric signal is as follows:

[0116]

[0117] In the formula, P(i|A) is the probability that the photometric signal A belongs to the ith known target, p i and p j are the weights of the ith known target and the jth known target in the plurality of known category targets, respectively, the value is determined according to experience, M is the number of known category targets, f i (A) represents the prior distribution probability of the photometric signal A in the ith known target, f j (A) represents the prior distribution probability of the photometric signal A in the jth known target.

[0118] In order to reduce the calculation complexity, the above formula can be approximated to obtain the following formula:

[0119]

[0120] In the formula, x j is the jth feature value of the photometric signal A, s i，j is the variance of the jth feature in the ith known target, u i，j is the mean of the jth feature in the ith known target.

[0121] According to the above formula, the probability that each photometric sample in the photometric signal set belongs to any category can be calculated. For example, it is known that there are three categories X, Y and Z of known category targets, among which there are 30 photometric samples in the X category target, 30 photometric samples in the Y category target, and 40 photometric samples in the Z category target, each photometric sample has 12 features, and one feature subset contains 4 features, A, B, C and D. Then, through the above calculation formula, the mean, variance and distribution function of the A, B, C and D features in the X category target can be calculated first, and then the prior distribution probability of each photometric sample in the X category target can be determined according to the mean, variance and distribution function of the four features. Similarly, the prior distribution probability of each photometric sample in the three categories of targets can be calculated. Finally, the probability that each photometric sample belongs to any category is obtained.

[0122] In some embodiments, the score of the current feature subset is determined based on the position of the probability ranking of each photometric signal belonging to the true class in the total probability ranking, including:

[0123] The total ranking is determined according to the size of the probability value, and the position of the probability ranking of the current photometric signal belonging to the true class in the total ranking is determined;

[0124] Based on the preset position scoring standard, the score of the current feature subset for distinguishing the current photometric signal is determined according to the position;

[0125] The scores of the current feature subset for distinguishing all photometric signals are added to obtain the score of the current feature subset.

[0126] In this embodiment, for any photometric signal, which class it belongs to is known. After calculating the probability of each photometric signal belonging to each class by using a certain feature subset according to the above scheme, the corresponding score is calculated, and the scores of all photometric signals are added to obtain the final score of the feature subset.

[0127] In this embodiment, let the ith photometric sample A i The performance score under the feature subset Φ is V(A i , Φ), and the score of the feature subset Φ is:

[0128]

[0129] In the formula, N is the number of photometric samples.

[0130] For example, there are 100 classes, each class includes 20 photometric samples, so there are 2000 photometric samples, each photometric sample corresponds to 5 features, and one feature subset is A. The preset scoring standard is: the scores of the photometric sample true label appearing in the known class in the front [1 3 7 10 15 30] are set to [1 0.9 0.7 0.6 0.55 0.5].

[0131] Then, the A subset is used to classify all the photometric samples, and for any photometric sample X, 100 probabilities can be obtained, for example, the probability of X belonging to the first category is 0.85, the probability of X belonging to the second category is 0.7, the probability of X belonging to the third category is 0.8, the probability of X belonging to the fourth category is 0.9, and so on, and the probability of X belonging to the one hundredth category is 0.3. Then, the 100 probabilities are sorted according to the numerical value. Assuming that the true category of X is the third category, when the probability of X belonging to the third category ranks the third in the total probability ranking, the score of the A subset in judging the photometric sample X is 0.9; when the probability of X belonging to the third category ranks the ninth in the total probability ranking, the score of the A subset in judging the photometric sample X is 0.6; when the probability of X belonging to the third category ranks the thirty-fifth in the total probability ranking, the score of the A subset in judging the photometric sample X is 0; and so on. A score can be obtained each time, and the scores of the 2000 samples are added to obtain the score of the A subset. Similarly, the scores of all the feature subsets can be obtained. The feature subset with the highest score is the optimal feature subset.

[0132] Finally, for step 108, the optimal feature subset is used to classify the space target to be processed, and the category of the space target to be processed is determined, including:

[0133] The prior distribution probability of the photometric signal of the space target to be processed is calculated based on the optimal feature subset.

[0134] Based on the prior distribution probability, the probability of the photometric signal belonging to any category of the known category target is calculated according to the Bayesian posterior probability formula.

[0135] The category with the highest probability is determined as the category of the space target to be processed.

[0136] The optimal feature subset is used for classification, and an accurate classification result can be obtained.

[0137] It should be noted that the effectiveness of the optimal feature subset Φ opt can also be evaluated by the following formula:

[0138]

[0139] In the formula, A i is the i th photometric sample, V(A i , Φ opt ) is the performance score of A i under the optimal feature subset Φ opt , and N is the number of photometric samples.

[0140] The greater the value of H is, the more accurate the optimal feature subset of the multi-resolution describes the photometric set, the stronger the effectiveness of the features is, and the better the classification effect can be achieved.

[0141] It should be noted that the photometric signal and the photometric sample have similar meanings. For the target to be processed, they are collectively referred to as photometric signals. For the known target used to calculate the optimal feature subset, they can be referred to as photometric signals or photometric samples.

[0142] As shown in Figure 2 , Figure 3 , an embodiment of the present application provides a feature extraction and classification device for a spatial target photometric signal. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware layer, as shown in Figure 2 , a hardware architecture diagram of an electronic device in which the feature extraction and classification device for a spatial target photometric signal is located, in addition to the processor, the memory, the network interface, and the non-volatile memory shown in Figure 2 , the electronic device in which the device is located in the embodiment can usually include other hardware, such as a forwarding chip responsible for processing packets, and the like. Taking the software implementation as an example, as shown in Figure 3 , as a logically meaningful device, it is formed by the CPU of the electronic device in which it is located reading the corresponding computer program in the non-volatile memory into the memory for running. The feature extraction and classification device for a spatial target photometric signal provided in the embodiment includes:

[0143] The preprocessing module 300 is configured to preprocess the photometric signal of the spatial target to be processed to obtain a preprocessed photometric signal.

[0144] The sampling module 302 is configured to down-sample the preprocessed photometric signal based on a plurality of preset sampling intervals to obtain photometric signals of different time resolutions.

[0145] The first calculation module 304 is configured to calculate a preset number of cumulative quantity features of the photometric signals of different time resolutions.

[0146] The extraction module 306 is configured to extract a previously calculated optimal feature subset from the preset number of cumulative quantity features. The optimal feature subset is calculated based on a photometric signal set of a plurality of known category targets, and the spatial target to be processed is one category of the known categories.

[0147] The determination module 308 is configured to classify the spatial target to be processed based on the optimal feature subset to determine the category of the spatial target to be processed.

[0148] In the embodiments of the present application, the preprocessing module 300 can be configured to perform step 100 in the above-mentioned method embodiments, the sampling module 302 can be configured to perform step 102 in the above-mentioned method embodiments, the first calculation module 304 can be configured to perform step 104 in the above-mentioned method embodiments, the extraction module 306 can be configured to perform step 106 in the above-mentioned method embodiments, and the determination module 308 can be configured to perform step 108 in the above-mentioned method embodiments.

[0149] In some embodiments, the preprocessing module 300 is configured to perform the following operations:

[0150] processing outliers in the photometric signals of the space target to be processed;

[0151] suppressing long-time noise in the photometric signals that have been processed for outliers;

[0152] mapping the photometric signals after the long-time noise is suppressed to a preset observation distance to obtain the preprocessed photometric signals.

[0153] In some embodiments, the first calculation module 304 is configured to perform the following operations:

[0154] for each time resolution photometric signal, calculating a preset order moment feature of the current time resolution photometric signal, and based on the preset order moment feature, calculating a preset order cumulative quantity feature of the current time resolution photometric signal;

[0155] collecting the preset order cumulative quantity features of each time resolution photometric signal to obtain a preset number of cumulative quantity features of different time resolution photometric signals.

[0156] In some embodiments, the optimal feature subset is calculated based on the following method:

[0157] obtaining a set of photometric signals of a plurality of known category targets;

[0158] calculating a preset number of cumulative quantity features of each photometric signal in the set of photometric signals;

[0159] based on the preset number of cumulative quantity features of each photometric signal, determining a plurality of feature subsets, each feature subset containing at least one cumulative quantity feature, and each feature subset containing different cumulative quantity features;

[0160] for each feature subset, calculating a probability that each photometric signal belongs to any category of the known categories, and based on the ranking of the probability that each photometric signal belongs to the true category in all probabilities, determining a score of each feature subset;

[0161] determining the feature subset with the highest score as the optimal feature subset.

[0162] In some embodiments, for each feature subset, the probability of each photometric signal belonging to any of the known classes is calculated, and the score of each feature subset is determined based on the ranking of the probability of each photometric signal belonging to the true class in the ranking of all probabilities, including:

[0163] For each set of photometric signals corresponding to a known class target, the following is performed:

[0164] For each feature in the feature subset, the mean, variance and distribution function of the current feature of each photometric signal in the current class are calculated; based on the mean, variance and distribution function of each feature, the prior distribution probability of the current photometric signal is determined; based on the prior distribution probability of the current photometric signal, the probability of the current photometric signal belonging to any of the known classes is calculated, respectively, wherein the true class of the current photometric signal is known;

[0165] Based on the position of the probability of each photometric signal belonging to the true class in the total probability ranking, the score of the current feature subset is determined.

[0166] In some embodiments, the score of the current feature subset is determined based on the position of the probability of each photometric signal belonging to the true class in the total probability ranking, including:

[0167] The total ranking is determined according to the size of the probability value, and the position of the probability of the current photometric signal belonging to the true class in the total ranking is determined;

[0168] Based on the preset position scoring standard, the score of the current feature subset for distinguishing the current photometric signal is determined according to the position;

[0169] The scores of the current feature subset for distinguishing all photometric signals are added to obtain the score of the current feature subset.

[0170] In some embodiments, the determining module 308 is configured to perform the following operations:

[0171] Based on the optimal feature subset, the prior distribution probability of the photometric signal of the space target to be processed is calculated;

[0172] Based on the prior distribution probability, the probability of the photometric signal belonging to any of the known class targets is calculated according to the Bayes posterior probability formula;

[0173] The class with the highest probability is determined as the class of the space target to be processed.

[0174] It can be understood that the structural schematic of the embodiments of the present application does not constitute a specific limitation on the space target luminosity signal feature extraction and classification device. In other embodiments of the present application, the space target luminosity signal feature extraction and classification device can include more or fewer components than the schematic, or combine certain components, or split certain components, or different component arrangements. The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0175] The information interaction between the modules in the device, the execution process, and the like, are based on the same concept as the method embodiments of the present application, and the specific content can be referred to the description in the method embodiments of the present application, which will not be described here.

[0176] The embodiments of the present application also provide an electronic device including a memory and a processor, the memory stores a computer program, and the processor implements the space target luminosity signal feature extraction and classification method in any of the embodiments of the present application when executing the computer program.

[0177] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, causes the processor to execute the space target luminosity signal feature extraction and classification method in any of the embodiments of the present application.

[0178] Specifically, a system or device equipped with a storage medium can be provided, and the storage medium stores a software program code for implementing the functions of any of the above embodiments, and the computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage medium.

[0179] In this case, the program code read from the storage medium itself can implement the functions of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute a part of the present application.

[0180] The storage medium for providing the program code includes a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer via a communication network.

[0181] In addition, it should be clear that not only the program code read by the computer can be executed, but also part or all of the actual operations can be completed by the operating system and the like operating on the computer based on the instructions of the program code, so as to implement the functions of any of the above embodiments.

[0182] Further, it is understood that the programs while being read by the storage media can be written into the memory provided in the extension board inserted into the computer or the memory provided in the extension module connected to the computer, and then the CPU or the like installed in the extension board or the extension module is caused to perform part or all of the actual operations based on the instructions of the program codes, thereby realizing the functions of any of the above-described embodiments.

[0183] It is noted that, in this document, the terms such as first and second are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Also, the terms "comprising", "containing" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements not only includes those elements, but also includes other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0184] It is understood by those skilled in the art that all or part of the steps of the above-described method embodiments can be completed by program instruction related hardware, and the aforementioned program can be stored in a computer readable storage medium, and the program performs the steps including the above-described method embodiments when executed; and the aforementioned storage medium includes ROM, RAM, magnetic disc or optical disc and various program code storage media.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for feature extraction and classification of a spatial target photometric signal, characterized in that, The method comprises the following steps: preprocessing a luminosity signal of a to-be-processed space target to obtain a preprocessed luminosity signal; down-sampling the preprocessed luminosity signal based on a plurality of preset sampling intervals to obtain luminosity signals with different time resolutions; calculating a preset number of accumulation quantity features of the luminosity signals with different time resolutions; extracting a pre-calculated optimal feature subset from the preset number of accumulation quantity features, the optimal feature subset being calculated based on a luminosity signal set of a plurality of known category targets, and the to-be-processed space target being one of the known categories; classifying the to-be-processed space target based on the optimal feature subset to determine the category of the to-be-processed space target; the optimal feature subset being calculated based on the following method: obtaining a luminosity signal set of a plurality of known category targets; calculating a preset number of accumulation quantity features of each luminosity signal in the luminosity signal set; determining a plurality of feature subsets based on the preset number of accumulation quantity features of each luminosity signal, each of the feature subsets containing at least one accumulation quantity feature, and each of the feature subsets containing different accumulation quantity features; for each luminosity signal set corresponding to each of the known category targets, performing the following steps: for each feature in the feature subset, calculating the mean value, variance and distribution function of the current feature of each luminosity signal in the current category; determining the prior distribution probability of the current luminosity signal based on the mean value, variance and distribution function of each feature; and calculating the probability of the current luminosity signal belonging to any of the known categories based on the prior distribution probability of the current luminosity signal, wherein the true category of the current luminosity signal is known; determining the score of the current feature subset based on the position of the probability of each luminosity signal belonging to the true category in the total probability ranking; determining the feature subset with the highest score as the optimal feature subset.

2. The method of claim 1, wherein, The preprocessing of the luminosity signal of the to-be-processed space target to obtain the preprocessed luminosity signal comprises the following steps: processing outliers in the luminosity signal of the to-be-processed space target; suppressing long-time noise in the luminosity signal that has been processed for outliers; mapping the luminosity signal after the long-time noise suppression to a preset observation distance to obtain the preprocessed luminosity signal.

3. The method of claim 1, wherein, The calculation of the preset number of accumulation quantity features of the luminosity signals with different time resolutions comprises the following steps: for each time resolution luminosity signal, calculating a preset order moment feature of the current time resolution luminosity signal; and based on the preset order moment feature, calculating the preset order accumulation quantity feature of the current time resolution luminosity signal; collecting the preset order accumulation quantity features of each time resolution luminosity signal to obtain the preset number of accumulation quantity features of the luminosity signals with different time resolutions.

4. The method of claim 1, wherein, The determination of the score of the current feature subset based on the position of the probability of each luminosity signal belonging to the true category in the total probability ranking comprises the following steps: determining the total ranking according to the size of the probability value, and determining the position of the probability of the current luminosity signal belonging to the true category in the total ranking; based on a preset position scoring criterion, scores of the current feature subsets in distinguishing the current photometric signals are determined; scores of the current feature subsets in distinguishing all photometric signals are added to obtain scores of the current feature subsets.

5. The method of claim 1, wherein, The classification of the to-be-processed space target based on the optimal feature subset comprises: calculating a prior distribution probability of the photometric signal of the to-be-processed space target based on the optimal feature subset; calculating a probability that the photometric signal belongs to any of the known category targets based on a Bayesian posterior probability formula based on the prior distribution probability; determining a category of the to-be-processed space target as a category with the highest probability.

6. An apparatus for feature extraction and classification of a spatial target photometric signal, the apparatus comprising: comprise: a preprocessing module configured to preprocess a photometric signal of a to-be-processed space target to obtain a preprocessed photometric signal; a sampling module configured to down-sample the preprocessed photometric signal based on a preset plurality of sampling intervals to obtain photometric signals with different time resolutions; a first calculation module configured to calculate a preset number of cumulative quantity features of the photometric signals with different time resolutions; an extraction module configured to extract an optimal feature subset from the preset number of cumulative quantity features, the optimal feature subset being calculated based on a photometric signal set of a plurality of known category targets, and the to-be-processed space target being one of the known categories; a determination module configured to classify the to-be-processed space target based on the optimal feature subset to determine a category of the to-be-processed space target; The optimal feature subset is calculated based on the following method: obtaining a photometric signal set of a plurality of known category targets; calculating a preset number of cumulative quantity features of each photometric signal in the photometric signal set; based on the preset number of cumulative quantity features of each photometric signal, determining a plurality of feature subsets, each of the feature subsets comprising at least one of the cumulative quantity features, and each of the feature subsets comprising different cumulative quantity features; for each photometric signal set corresponding to each of the known category targets, the following are performed: for each feature in the feature subset, a mean value, a variance, and a distribution function of a current feature of each photometric signal in the current category are calculated; based on the mean value, the variance, and the distribution function of each feature, a prior distribution probability of the current photometric signal is determined; based on the prior distribution probability of the current photometric signal, a probability that the current photometric signal belongs to any of the known categories is calculated, wherein the true category of the current photometric signal is known; based on a position of the probability that each photometric signal belongs to the true category in a total probability ranking, a score of the current feature subset is determined; the feature subset with the highest score is determined as the optimal feature subset. 7.An electronic device comprising a memory and a processor, the memory having stored therein a computer program, characterized in that, The processor executes the computer program to implement the method of any one of claims 1-5.

8. A storage medium having stored thereon a computer program, characterized in that When the computer program is executed in the computer, the computer is caused to execute the method of any one of claims 1-5.

Images