A high-time-sensitive target recognition method combined with boost phase detection

Abstract

The present invention belongs to the field of target recognition in infrared image processing and provides a high-time-sensitive target recognition method combined with boost segment detection to solve the problem of difficult recognition of high-time-sensitive targets. The main scheme includes: extracting the infrared radiation energy curve and space-time characteristic curve of the high-time-sensitive target, and performing outlier processing on these two types of time series signals to obtain the time series signal to be identified. Subsequently, the space-time characteristic curve is used to perform a first round of preliminary screening of non-boost segment targets. The two types of time series signals are then input into the main process of boost segment detection for a second round of screening to obtain high-time-sensitive targets containing boost segments, thereby realizing the detection of boost segments. Subsequently, key point detection is performed on the target containing boost segments to facilitate the extraction of the boost segment portion. Finally, a method of calculating waveform similarity and matching is used to classify and identify the high-time-sensitive targets containing boost segments.

Description

Technical Field

[0001] The invention discloses a high-time-sensitive target recognition method combined with boost segment detection, which is used for high-time-sensitive target recognition in infrared images and belongs to the field of target recognition in infrared image processing. Background Art

[0002] With the development of infrared imaging technology, infrared imaging has been widely used in military, civilian and other related fields in various countries. Infrared small target detection plays a huge role in early warning systems, precision strike weapons, etc. Infrared imaging generally has the characteristics of long imaging distance and is easily affected by environmental factors. Therefore, infrared small targets are small in size and have no obvious shape features in the image. Infrared small target recognition has always been a difficult problem in the field of target recognition.

[0003] Infrared thermal imaging systems receive infrared radiation from both the target and the background, converting optical signals into electrical signals and transforming invisible infrared radiation into visible two-dimensional images. Due to the unique wavelength bands used in infrared imaging, they offer significant advantages in military applications. Compared to visible light imaging, infrared detection is less dependent on the environment. While visible light detection is typically limited to daytime, infrared detection imaging is not subject to this limitation and operates in all weather conditions. Furthermore, infrared detection offers advantages such as high sensitivity, strong resistance to electromagnetic interference, a wide dynamic range, and good environmental adaptability. Therefore, infrared detection can capture, lock onto, track, and identify targets even in complex backgrounds.

[0004] However, for highly time-sensitive targets, their fast movement, small size, and distance from the infrared thermal imaging system result in them occupying a small size in infrared images and lacking key information such as color and texture. However, identifying the specific type of highly time-sensitive targets can provide key information such as the target's motion characteristics and enable timely and effective countermeasures. Therefore, using infrared images to identify the type of highly time-sensitive targets has extraordinary research value and significance. Taking these factors into consideration, this paper proposes an algorithm for identifying the type of highly time-sensitive targets based on infrared image sequences. This algorithm detects the target's boost phase, then performs key point detection on it. Based on this boost phase, the target's type is identified. This algorithm achieves specific identification of highly time-sensitive targets and improves recognition accuracy and efficiency to a certain extent. Therefore, high-time-sensitive target recognition algorithms are of great research significance in practical applications. Summary of the Invention

[0005] The present invention aims to address the shortcomings of existing high-time-sensitive target recognition methods and to solve the following technical problems:

[0006] Traditional methods have low feature utilization: Existing technologies rely on full-time signal analysis and fail to focus on high-value features of the boost phase (engine operation phase) (such as strong infrared radiation, regular speed changes, etc.), resulting in limited recognition accuracy and low computational efficiency.

[0007] Insufficient false alarm suppression capability: Under complex backgrounds and noise interference, existing algorithms find it difficult to effectively distinguish real targets from false alarm signals. In particular, they lack error correction mechanisms for sudden changes in space-time characteristics or violations of physical laws, resulting in a high misjudgment rate.

[0008] Loss of high-frequency feature information: Conventional filtering and missing value processing methods over-smoothe the data, weakening the extreme or sudden change characteristics of key points in the boost phase (such as ignition and shutdown moments), affecting the detection accuracy of subsequent key events.

[0009] Limitations of template matching: Existing template library matching algorithms ignore the specificity of dynamic information of different target boosting segments when calculating waveform similarity, and fail to improve matching efficiency through efficient downsampling and path optimization, resulting in insufficient generalization capabilities for classification and recognition.

[0010] Through the technical solution proposed in the present invention, an effort is made to achieve accurate and efficient type recognition of highly time-sensitive targets, which is particularly suitable for real-time processing requirements under complex backgrounds.

[0011] The technical solution adopted in the present invention is as follows:

[0012] A high-time-sensitive target recognition method combined with boost phase detection includes the following steps:

[0013] Step 1: Extract the infrared radiation energy curve and space-time characteristic curve of the high time-sensitive target;

[0014] Step 2: Perform outlier processing on the original signal to reduce noise and remove outliers;

[0015] Step 3: Perform boost segment detection on the signal after data processing in step 2;

[0016] Step 4: Perform key point detection on the signal containing the boost segment detected in step 3, and intercept the boost segment according to the power-on / off point;

[0017] Step 5: The target signal including the boost segment obtained in step 4 is matched with the type template library using the template matching method to obtain the classification and recognition result of the specific type of the target.

[0018] In particular, the step 1 includes the following steps:

[0019] Step 1.1 extracts the infrared radiation energy characteristic curve of the target in the infrared image;

[0020] Since high-time-sensitive targets are generally located at high altitudes or even in outer space, they occupy very few pixels in the infrared images detected by the infrared detection system. In other words, high-time-sensitive targets are very "small" in the infrared images, which results in a lack of information such as shape, texture, and color, making them very difficult to identify. Therefore, we use infrared images to extract the target energy time series signal. Usually, when calculating the energy based on the infrared radiation signal of the target area, we use the following formula:

[0021] E=∑ (x，y) ∈ R I(x,y) (1-1)

[0022] In formula (1-1), I(x, y) is the pixel intensity value within the target area R (representing the radiation brightness).

[0023] Then, the total energy of the target area is calculated for each frame of infrared image to form a time series signal E(t) of energy changing with time, that is, the infrared radiation energy characteristic curve is obtained.

[0024] Step 1.2 extracts the space-time characteristic curve of the target in the infrared image;

[0025] To extract the space-time characteristic curve, it is usually necessary to combine the spatial position change and time change information of the target in the image sequence for calculation. First, the center of mass of the target in the infrared image needs to be calculated:

[0026]

[0027] In equations (1-2) and (1-3), R is the target area and I(x, y) is the pixel intensity. After calculating the target's center of mass coordinates, the target's discrete velocity can be calculated using the velocity calculation formula:

[0028]

[0029] In formula (1-4) and formula (1-5), x c (t) and y c (t) is the coordinate of the center of mass of the target in the tth frame, and Δt is the time interval between adjacent frames. For each frame, the speed of the target is calculated as:

[0030]

[0031] Then the space-time characteristic curve v(t) of the high time-sensitive target is drawn.

[0032] In particular, the step 2 includes the following steps:

[0033] Step 2.1: Process missing values of the extracted time series signal;

[0034] Due to the influence of multiple factors such as observation angle, observation accuracy, weather conditions, and obstruction by foreign objects, the extracted time series signal may have missing values. In order to facilitate subsequent calculations, we need to perform missing value processing operations.

[0035] This method uses linear interpolation to fill missing values, that is, linear interpolation is performed using the values of adjacent points. The calculation formula for the filling value of the missing point is as follows:

[0036]

[0037] In formula (2-1), x t Refers to the time when the missing value is located, x t-1 and x t+1 is the time adjacent to this time.

[0038] Step 2.2 uses a low-frequency filtering method guided by polynomial fitting to remove noise from the time series signal that has processed missing values;

[0039] Since there are sharply changing parts in the infrared radiation energy characteristic curve and space-time characteristic curve of high time-sensitive targets, and this part of the change trend represents the special motion characteristics of high time-sensitive targets, we need to retain the high-frequency part of the timing signal (such as spikes, rapid changes, etc.), so we use a filtering method that can retain the high-frequency part on the basis of smoothed data.

[0040] The core of this method is to smooth the signal by fitting a polynomial within a sliding window. Assume that the signal is x = {x1, x2, ..., x N}, this method selects the window size and polynomial order in each sliding window, fits the signal, and then uses the fitted value to replace the value of the window center point. The specific calculation formula used in this method is as follows:

[0041]

[0042] In formula (2-2), y t is the signal value after filtering, h i is the filter weight, x t+i is the value in the original signal, where the filter weight is determined by the polynomial order d and the window size W.

[0043] In particular, the specific steps of step 3 are:

[0044] Step 3.1: Input the time series signal processed in step 2 into the boost phase detection system (initial velocity screening module) and perform initial velocity screening based on the space-time characteristic curve;

[0045] First, considering that the infrared radiation characteristic curves of reentry targets and high-time-sensitivity targets are extremely similar, the difference is that the initial velocity of the space-time characteristic curve of the reentry type is generally larger, while the velocity value of the front part of the boost phase of the high-time-sensitivity target has not yet been fully accelerated, so the velocity value of the front part of the boost phase should be lower. Using this characteristic, we designed a space-time characteristic curve initial velocity screening module to exclude reentry types and other non-boost phase targets in space. The specific formula is as follows:

[0046]

[0047] In formula (3-1), δ p This is the state after the space-time characteristic curve initial velocity screening module. When the state is set to 1, it will enter the next step of the process. Otherwise, it will jump out of the detection process and return to the non-boost segment detection result. i0 Refers to the initial velocity value in the target space-time characteristic curve, v p Refers to the initial velocity threshold set by the screening module.

[0048] Step 3.2: Input the timing signal filtered in step 3.1 into the boosting segment detection module;

[0049] The boost phase refers to the period during which a highly time-sensitive target takes off and ascends under the influence of the engine. Subsequently, under the influence of thrust, aerodynamic forces, control forces, and the Earth's gravity, its flight speed, altitude, and distance gradually increase. This is reflected in the infrared energy radiation characteristic curve and the space-time characteristic curve as follows: 1. The space-time characteristic curve of a highly time-sensitive target continuously rises during the boost phase, with a rapid rate of increase (large acceleration). 2. The infrared radiation energy characteristic curve of a highly time-sensitive target remains above a certain threshold during the boost phase, and the infrared radiation energy characteristic curve exhibits a regular waveform. Therefore, we design a detection algorithm based on these characteristics.

[0050] First, the space-time characteristic curve characteristics of the high time-sensitive target, assuming that the space-time characteristic curve is x s ={x s1 , x s2 ,…,x sN}, we use the sliding window form to calculate the curve change trend in the window, tentatively set the length of the sliding window to be W, and the curve in the window to be x sw ={x st , x s(t+1) ,…,x s(t+w-1)}, for the curve within the window we calculate its first-order difference:

[0051] Δx st =x st -x s(t-1) , t≤W (3-2)

[0052] Then calculate the curve rise coefficient within the window:

[0053]

[0054] In formula (3-3) and formula (3-4), c t is the change trend between every two points in the window, c t When it is 1, it means that the space-time characteristic curve is on an upward trend in the second before the moment, otherwise it is on a downward trend or no change. W is the window length, θ T It is the unit of measurement for the proportion of the upward trend of the curve in the sliding window. When θ T When the target curve is greater than or equal to the rising proportion threshold of the space-time characteristic curve set in the boost segment detection module, the target curve will enter the next step of detection.

[0055] Step 3.3: further determine the rising range of the space-time characteristic curve if it meets the rising trend;

[0056] In order to avoid the impact of false alarm targets on the boost phase detection results, we also need to detect and determine the rise of the space-time characteristic curve. Only when the rise is greater than a certain level can we consider it to be acceleration caused by the engine, rather than a false alarm. Therefore, we will calculate the average rise coefficient θ of the space-time characteristic curve within the sliding window v ,θ v The specific calculation formula is as follows:

[0057]

[0058] In formula (3-5), W represents the length of the sliding window, Δx t is the velocity difference between each point in the window and the previous moment. v If the rise in the space-time characteristic curve exceeds the set threshold, we consider the target's space-time characteristic curve to be in an ascending phase, and the upward trend is caused by engine operation. Once the target meets the boost phase test on the space-time characteristic curve, we will further test the target's infrared radiation energy characteristic curve.

[0059] Step 3.4: Determine the infrared radiation energy intensity of the target whose space-time characteristic curve meets the boost phase characteristics;

[0060] Since the engine of a high-time-sensitive target continues to work during the boost phase, the target tail flame generates high heat and strong infrared radiation. Therefore, the infrared radiation energy characteristics of the high-time-sensitive target during this phase will continue to be higher than a certain threshold. Therefore, we design a continuous strong energy radiation detection algorithm to detect this feature. Assume that the infrared radiation energy characteristic curve is x I ={x I1 , x I2 ,…,x IN}, the sliding window framework is still used for detection, the window length is W, when the strong energy ratio coefficient θ e When the value in the window is greater than the threshold, we believe that the window is in a state of strong energy radiation. Strong energy ratio coefficient θ e The specific calculation formula is as follows:

[0061]

[0062] In formula (3-6), θ e , is the strong energy ratio, c t is a unit of measurement for strong energy, P e Set a threshold for strong energy, when x t Greater than or equal to P e When , it is considered that the target is in a strong energy radiation state at this point, c t Equal to 1, then calculate θ e , when θ e When it is greater than the set threshold, the target in the window is considered to be in a state of strong energy radiation.

[0063] At this point, the target's space-time characteristic curve and infrared radiation energy characteristic curve both meet the boost phase characteristics and pass the boost phase detection. The boost phase detection module ends and returns the detection results.

[0064] Step 3.5: Conduct targeted testing on highly time-sensitive special targets;

[0065] There are some special cases among high-time-sensitive targets, whose infrared radiation energy characteristic curve and space-time characteristic curve are different. Although the number of these special cases is very small, considering the strategic value and importance of high-time-sensitive targets, we need to further detect these special targets. Since these special targets have special and different performances in infrared radiation energy characteristic curve and space-time characteristic curve, each type of special target needs to design a targeted detection algorithm to detect its boosting stage. Finally, the detection algorithms of special targets are integrated into a special target detection algorithm library. in Represents the detection algorithm for each type of special target. The special target detection algorithm library will be used to detect special boost segment targets in step 3.5.

[0066] Step 3.6 Design monitoring module to optimize algorithm performance;

[0067] If the first sliding window fails to detect a boost phase, the window continues sliding backward along the timeline. The step size d of each slide can be adjusted according to specific requirements. In this method, the step size d is set to 1, achieving real-time detection every one second based on the input. However, considering the practical significance of the boost phase, which generally occurs at the beginning of the entire motion phase of a highly time-sensitive target, if a target remains in a low energy state or low speed for a long period after being detected, it can be considered a non-boost phase target. Based on this characteristic, this method designs an energy-speed monitoring module to optimize the boost phase detection algorithm, reducing the computation time required and optimizing algorithm performance.

[0068] The energy and speed monitoring module monitors the energy and speed levels of the target in real time. When the signal in the sliding window does not meet the boost stage state, it will slide backward. At this time, the energy and speed monitoring module will record the energy level λ of the center point in the window at this time. e With speed level λ v .

[0069]

[0070] In formula (3-8), x et Refers to the infrared radiation energy characteristic curve value at the center of the window, θ T and θ v is the rising proportion coefficient and rising amplitude coefficient of the space-time characteristic curve mentioned above, P e 、P T With P v are the thresholds set in the boost segment detection module. e With speed level λ v When the accumulation reaches a certain threshold, it is considered that the target is in a low energy level or low speed level stage for a long time, and the target is determined to be a non-boost segment target, and the non-boost segment detection result is returned.

[0071] Step 3.7: Perform a final error correction check on the detected boost phase targets to ensure the accuracy of boost phase recognition.

[0072] After the above steps, the boost phase target has been detected. The boost phase detection can be completed by the error correction module for the detected boost phase target. Due to objective laws and facts, the space-time characteristic curve of a high-time-sensitive target will not undergo a sudden change between two points, while the space-time characteristic curve of a false alarm will often undergo a sudden change. Therefore, this method uses this characteristic to design an error correction module to perform error correction review on the detected boost phase target.

[0073] Traverse the target space-time characteristic curve within the sliding window and calculate its acceleration:

[0074]

[0075] Get the acceleration set A in the window = {a1, a2, ..., a t If there is a value exceeding the threshold value in the acceleration set, it is considered that the target space-time characteristic curve violates the objective physical law and is judged as a false alarm.

[0076] In particular, the specific steps of step 4 are:

[0077] Step 4.1: Detect key points of the detected boost segment signal;

[0078] Key points refer to the moments when important events occur during the flight of a highly time-sensitive target, such as ignition, shutdown, and deflagration. By detecting key points, we can better confirm the flight motion process of a highly time-sensitive target and facilitate subsequent extraction and processing of more valuable data. The following details the method for key point detection:

[0079] First define a function Its expansion and translation forms are:

[0080]

[0081] Where a is the scale factor, which is used to control the frequency of the function, and b is the translation factor, which is used to control the time of the function.

[0082] We use high-pass and low-pass filters to decompose the signal x(t):

[0083] c a，n =∑ t x(t)·φ a，n (t) (4-2)

[0084]

[0085] where c a，n is the approximate coefficient, indicating the low-frequency component; c d，n is the detail coefficient, indicating the high-frequency component; φ a，n (t) is the scaling function, This is the custom function mentioned above.

[0086] Decompose the signal into J layers:

[0087]

[0088] in is the approximate coefficient of the J-th layer, is the detail coefficient of the jth layer.

[0089] Then calculate the magnitude of the detail coefficient:

[0090]

[0091] when When , we consider this point as a mutation point. Where θ is the threshold, which can be set based on historical data or statistical characteristics.

[0092] Satisfy all The time t is found and arranged in chronological order to detect the on / off points of the target data of the boosting segment.

[0093] In particular, the specific steps of step 5 are:

[0094] Step 5.1: Crop the data that has been key point detected and only keep the boost segment;

[0095] Generally speaking, the starting and ending points of the boost segment are the first power-on point and the last power-off point in the key point detection, and the target boost segment contains rich semantic information. Different types of targets have different infrared radiation energy characteristic curves and space-time characteristic curves in the boost segment, including the peak and waveform of the infrared radiation energy characteristic curve, the duration of the boost segment, the trend and peak of the space-time characteristic curve, etc. It is feasible and more accurate to identify the target type based on the boost segment data containing rich information. Therefore, this method extracts the boost segment of the data according to the detected key points for type recognition.

[0096] Step 5.2, calculating the waveform similarity between the boost segment target data and the data in the type template library;

[0097] The boost phase target data extracted in step 5.1 is matched against the target type template library. Each data entry in the target type template library represents a type, and the data in the template library are the most representative data of each type. Therefore, the extracted boost phase target data and the data in the template library are used to calculate the similarity of the infrared radiation energy characteristic curve waveform.

[0098] First, assume that the extracted boost phase target infrared radiation energy characteristic curve sequence is x=[x1, x2,…, x t ], the dataset in the target type template library is Φ={Y1,Y2,…,Y n}, where Y1 to Y n Refers to different types of data templates in the template library, mathematically expressed as Y = [y1, y2, ..., y t ], first recursively downsample the two time series:

[0099] X lo w=[x1,x 1+k , x 1+2k ,…] (5-1)

[0100] Y lo w=[y1,y 1+k ,y 1+2k ,…] (5-2)

[0101] where Y low is the template data after recursive downsampling, X low is the infrared radiation energy characteristic curve sequence after recursive downsampling, k refers to the recursive downsampling time interval, and the waveform similarity is calculated at low resolution after recursive downsampling of the time series. Here, the downsampling sequence X is calculated. low and Y low The optimal alignment path Path low First, define the cumulative distance D(i, j) low :

[0102] D(i, j) low =d(x i ,y i ) low +min{D(i-1, j) low , D(i, j-1) low , D(i-1, j-1) low}(5-3)

[0103] where d(x i ,y j ) low =|x i -y j | 2 is the Euclidean local distance. Assuming that the two time series have n and m points respectively after downsampling, a n×m cumulative distance matrix is constructed, where each element represents x i and y j The distance, D(i, j) low is the value of the i, j element in the cumulative distance matrix, which represents the minimum path cost from the starting point to (i, j). The boundary conditions of the search process are:

[0104] D(0,0) low =0, D(i, 0) low =∞, D(0, j) low =∞(i, j>0) (5-4)

[0105] When the lower right corner (n, m) of the cumulative distance matrix is searched, the Path in the low-resolution case is obtained. low At this time, the path expansion operation is performed to expand the low-resolution path Path lowMapping back to high resolution, each path is expanded into a set of neighborhood points. For example, if the low-resolution path point is (i, j), it is expanded to {(ki, kj), (ki+1, kj), (ki, kj+1), (ki+1, kj+1)} in high resolution, where k is the recursive downsampling interval. D(i, j) is recalculated in the expanded high-resolution path neighborhood to calculate the waveform similarity between the two time series.

[0106] Step 5.3: Match and identify target types based on waveform similarity;

[0107] When the waveform similarity between the target and a certain type of template is higher than the set threshold, the type is included in the matching template group with high similarity. After the waveform similarity calculation between the target and all types of templates is completed, the type with the highest waveform similarity in the matching template group with high similarity is taken as the type recognition result and output.

[0108] If the waveform similarity calculation results between the target and all templates are lower than the set threshold, a new type of template is created and added to the target type template library.

[0109] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0110] 1. Using boost segment detection as an auxiliary means can effectively improve the accuracy of high-time-sensitive target type recognition;

[0111] 2. During boost phase detection, the present invention proposes a real-time boost phase detection method that combines space-time characteristics with radiation characteristics. This effectively eliminates interference from false alarm sources and enables efficient and accurate boost phase detection, laying the foundation for subsequent target type identification.

[0112] 3. When processing outliers, we consider that the important information of the data is contained in the peak and high-frequency parts. We use special filtering to retain the high-frequency characteristics of the data to the greatest extent possible, laying the foundation for subsequent boost segment detection and type identification.

[0113] 4. Before type recognition, the data boosting segments are extracted separately to remove redundant information and retain high-value data segments, which improves the accuracy of type recognition while also improving the efficiency of the system.

[0114] 5. A detection method is proposed to detect the occurrence of key events during the target movement, laying the foundation for the subsequent extraction of boost segment data. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0116] Figure 1 This is a flow chart of the high time-sensitive infrared target type recognition system of the present invention;

[0117] Figure 2 It is a schematic diagram of the overall process of the present invention; DETAILED DESCRIPTION

[0118] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0119] Example 1

[0120] The present invention provides a high-time-sensitive target recognition method combined with boost phase detection, comprising the following steps:

[0121] The step 1 includes the following:

[0122] Step 1.1 extracts the infrared radiation energy characteristic curve of the target in the infrared image;

[0123] Since high-time-sensitive targets are generally located at high altitudes or even in outer space, they occupy very few pixels in the infrared images detected by the infrared detection system. In other words, high-time-sensitive targets are very "small" in the infrared images, which results in a lack of information such as shape, texture, and color, making them very difficult to identify. Therefore, we use infrared images to extract the target energy time series signal. Usually, when calculating the energy based on the infrared radiation signal of the target area, we use the following formula:

[0124] E=∑ (x，y)∈R I(x,y) (1-1)

[0125] In formula (1-1), I(x, y) is the pixel intensity value within the target area R (representing the radiation brightness).

[0126] Then, the total energy of the target area is calculated for each frame of infrared image to form a time series signal E(t) of energy changing with time, that is, the infrared radiation energy characteristic curve is obtained.

[0127] Step 1.2 extracts the space-time characteristic curve of the target in the infrared image;

[0128] To extract the space-time characteristic curve, it is usually necessary to combine the spatial position change and time change information of the target in the image sequence for calculation. First, the center of mass of the target in the infrared image needs to be calculated:

[0129]

[0130] In equations (1-2) and (1-3), R is the target area and I(x, y) is the pixel intensity. After calculating the target's center of mass coordinates, the target's discrete velocity can be calculated using the velocity calculation formula:

[0131]

[0132] In formula (1-4) and formula (1-5), x c (t) and y c (t) is the coordinate of the center of mass of the target in the tth frame, and Δt is the time interval between adjacent frames. For each frame, the velocity of the target is calculated as:

[0133]

[0134] Then the space-time characteristic curve v(t) of the high time-sensitive target is drawn.

[0135] The step 2 includes the following:

[0136] Step 2.1: Process missing values of the extracted time series signal;

[0137] Due to the influence of multiple factors such as observation angle, observation accuracy, weather conditions, and obstruction by foreign objects, the extracted time series signal may have missing values. In order to facilitate subsequent calculations, we need to perform missing value processing operations.

[0138] This method uses linear interpolation to fill missing values, that is, linear interpolation is performed using the values of adjacent points. The calculation formula for the filling value of the missing point is as follows:

[0139]

[0140] In formula (2-1), x t Refers to the moment where the missing value is located, x t-1 and x t+1 is the time adjacent to this time.

[0141] Step 2.2 uses a low-frequency filtering method guided by polynomial fitting to remove noise from the time series signal that has processed missing values;

[0142] Since there are sharply changing parts in the infrared radiation energy characteristic curve and space-time characteristic curve of high time-sensitive targets, and this part of the change trend represents the special motion characteristics of high time-sensitive targets, we need to retain the high-frequency part of the timing signal (such as spikes, rapid changes, etc.), so we use a filtering method that can retain the high-frequency part on the basis of smoothed data.

[0143] The core of this method is to smooth the signal by fitting a polynomial within a sliding window. Assume that the signal is x = {x1, x2, ..., x N}, this method selects the window size and polynomial order in each sliding window, fits the signal, and then uses the fitted value to replace the value of the window center point. The specific calculation formula used in this method is as follows:

[0144]

[0145] In formula (2-2), y t is the signal value after filtering, h i is the filter weight, x t+i is the value in the original signal, where the filter weight is determined by the polynomial order d and the window size W.

[0146] The step 3 includes the following:

[0147] Step 3.1 Input the timing signal processed in step 2 into the boost phase detection system and perform initial velocity screening based on the space-time characteristic curve;

[0148] First, considering that the infrared radiation characteristic curves of reentry targets and high-time-sensitivity targets are extremely similar, the difference is that the initial velocity of the space-time characteristic curve of the reentry type is generally larger, while the velocity value of the front part of the boost phase of the high-time-sensitivity target has not yet been fully accelerated, so the velocity value of the front part of the boost phase should be lower. Using this characteristic, we designed a space-time characteristic curve initial velocity screening module to exclude reentry types and other non-boost phase targets in space. The specific formula is as follows:

[0149]

[0150] In formula (3-1), δ p This is the state after the space-time characteristic curve initial velocity screening module. When the state is set to 1, it will enter the next step of the process. Otherwise, it will jump out of the detection process and return to the non-boost segment detection result. i0 Refers to the initial velocity value in the target space-time characteristic curve, v p Refers to the initial velocity threshold set by the screening module.

[0151] Step 3.2: Input the timing signal filtered in step 3.1 into the boosting segment detection module;

[0152] The boost phase refers to the period during which a highly time-sensitive target takes off and ascends under the influence of the engine. Subsequently, under the influence of thrust, aerodynamic forces, control forces, and the Earth's gravity, its flight speed, altitude, and distance gradually increase. This is reflected in the infrared energy radiation characteristic curve and the space-time characteristic curve as follows: 1. The space-time characteristic curve of a highly time-sensitive target continuously rises during the boost phase, with a rapid rate of increase (large acceleration). 2. The infrared radiation energy characteristic curve of a highly time-sensitive target remains above a certain threshold during the boost phase, and the infrared radiation energy characteristic curve exhibits a regular waveform. Therefore, we design a detection algorithm based on these characteristics.

[0153] First, the space-time characteristic curve characteristics of the high time-sensitive target, assuming that the space-time characteristic curve is x s ={x s1 , x s2 ,…,x sN}, we use the sliding window form to calculate the curve change trend in the window, tentatively set the length of the sliding window to be W, and the curve in the window to be x sw ={x st , x s(t+1) ,…,x s(t+w-1)}, for the curve within the window we calculate its first-order difference:

[0154] Δx st =x st -x s(t-1) , t≤W (3-2)

[0155] Then calculate the curve rise coefficient within the window:

[0156]

[0157] In formula (3-3) and formula (3-4), c t is the change trend between every two points in the window, c t When it is 1, it means that the space-time characteristic curve is on an upward trend in the second before the moment, otherwise it is on a downward trend or no change. W is the window length, θ T It is the unit of measurement for the proportion of the upward trend of the curve in the sliding window. When θ T When the target curve is greater than or equal to the rising proportion threshold of the space-time characteristic curve set in the boost segment detection module, the target curve will enter the next step of detection.

[0158] Step 3.3: further determine the rising range of the space-time characteristic curve if it meets the rising trend;

[0159] In order to avoid the impact of false alarm targets on the boost phase detection results, we also need to detect and determine the rise of the space-time characteristic curve. Only when the rise is greater than a certain level can we consider it to be acceleration caused by the engine, rather than a false alarm. Therefore, we will calculate the average rise coefficient θ of the space-time characteristic curve within the sliding window v ,θ v The specific calculation formula is as follows:

[0160]

[0161] In formula (4-5), W represents the length of the sliding window, Δx t is the velocity difference between each point in the window and the previous moment. v If the rise in the space-time characteristic curve exceeds the set threshold, we consider the target's space-time characteristic curve to be in an ascending phase, and the upward trend is caused by engine operation. Once the target meets the boost phase test on the space-time characteristic curve, we will further test the target's infrared radiation energy characteristic curve.

[0162] Step 3.4: Determine the infrared radiation energy intensity of the target whose space-time characteristic curve meets the boost phase characteristics;

[0163] Since the engine of a high-time-sensitive target continues to work during the boost phase, the target tail flame generates high heat and strong infrared radiation. Therefore, the infrared radiation energy characteristics of the high-time-sensitive target during this phase will continue to be higher than a certain threshold. Therefore, we design a continuous strong energy radiation detection algorithm to detect this feature. Assume that the infrared radiation energy characteristic curve is x = {x1, x2, ..., x N}, the sliding window framework is still used for detection, the window length is W, when the strong energy ratio coefficient θ e When the value in the window is greater than the threshold, we believe that the window is in a state of strong energy radiation. Strong energy ratio coefficient θ e The specific calculation formula is as follows:

[0164]

[0165] In formula (3-6), θ e , is the strong energy ratio, c t is a unit of measurement for strong energy, P e Set a threshold for strong energy, when x t Greater than or equal to P e When , it is considered that the target is in a strong energy radiation state at this point, c t Equal to 1, then calculate θ e , when θ e When it is greater than the set threshold, the target in the window is considered to be in a state of strong energy radiation.

[0166] At this point, the target's space-time characteristic curve and infrared radiation energy characteristic curve both meet the boost phase characteristics and pass the boost phase detection. The boost phase detection module ends and returns the detection results.

[0167] Step 3.5: Conduct targeted testing on highly time-sensitive special targets;

[0168] There are some special cases among high-time-sensitive targets, whose infrared radiation energy characteristic curve and space-time characteristic curve are different. Although the number of these special cases is very small, considering the strategic value and importance of high-time-sensitive targets, we need to further detect these special targets. Since these special targets have special and different performances in infrared radiation energy characteristic curve and space-time characteristic curve, each type of special target needs to design a targeted detection algorithm to detect its boosting stage. Finally, the detection algorithms of special targets are integrated into a special target detection algorithm library. in Represents the detection algorithm for each type of special target. The special target detection algorithm library will be used to detect special boost segment targets in step 3.5.

[0169] Step 3.6 Design monitoring module to optimize algorithm performance;

[0170] If the first sliding window fails to detect a boost phase, the window continues sliding backward along the timeline. The step size d of each slide can be adjusted according to specific requirements. In this method, the step size d is set to 1, achieving real-time detection every one second based on the input. However, considering the practical significance of the boost phase, which generally occurs at the beginning of the entire motion phase of a highly time-sensitive target, if a target remains in a low energy state or low speed for a long period after being detected, it can be considered a non-boost phase target. Based on this characteristic, this method designs an energy-speed monitoring module to optimize the boost phase detection algorithm, reducing the computation time required and optimizing algorithm performance.

[0171] The energy and speed monitoring module monitors the energy and speed levels of the target in real time. When the signal in the sliding window does not meet the boost phase state, it will slide backward. At this time, the energy and speed monitoring module will record the energy level λ of the center point in the window at this time. e With speed level λ v .

[0172]

[0173] In formula (3-8), x et Refers to the infrared radiation energy characteristic curve value at the center of the window, θ T and θ v is the rising proportion coefficient and rising amplitude coefficient of the space-time characteristic curve mentioned above, P e、P T With P v are the thresholds set in the boost segment detection module. e With speed level λ v When the accumulation reaches a certain threshold, it is considered that the target is in a low energy level or low speed level stage for a long time, and the target is determined to be a non-boost segment target, and the non-boost segment detection result is returned.

[0174] Step 3.7: Perform a final error correction check on the detected boost phase targets to ensure the accuracy of boost phase recognition.

[0175] After the above steps, the boost phase target has been detected. The boost phase detection can be completed by the error correction module for the detected boost phase target. Due to objective laws and facts, the space-time characteristic curve of a high-time-sensitive target will not undergo a sudden change between two points, while the space-time characteristic curve of a false alarm will often undergo a sudden change. Therefore, this method uses this characteristic to design an error correction module to perform error correction review on the detected boost phase target.

[0176] Traverse the target space-time characteristic curve within the sliding window and calculate its acceleration:

[0177]

[0178] Get the acceleration set A in the window = {a1, a2, ..., a t If there is a value exceeding the threshold value in the acceleration set, it is considered that the target space-time characteristic curve violates the objective physical law and is judged as a false alarm.

[0179] The step 4 includes the following:

[0180] Step 4.1: Detect key points of the detected boost segment signal;

[0181] Key points refer to the moments when important events occur during the flight of a highly time-sensitive target, such as ignition, shutdown, and deflagration. By detecting key points, we can better confirm the flight motion process of a highly time-sensitive target and facilitate subsequent extraction and processing of more valuable data. The following details the method for key point detection:

[0182] First define a function Its expansion and translation forms are:

[0183]

[0184] Where a is the scale factor, which is used to control the frequency of the function, and b is the translation factor, which is used to control the time of the function.

[0185] We use high-pass and low-pass filters to decompose the signal x(t):

[0186] c a，n =∑ t x(t)·φ a，n (t)(4-2)

[0187]

[0188] where c a，n is the approximate coefficient, indicating the low-frequency component; c d，n is the detail coefficient, indicating the high-frequency component; φ a，n (t) is the scaling function, This is the custom function mentioned above.

[0189] Decompose the signal into J layers:

[0190]

[0191] in is the approximate coefficient of the J-th layer, is the detail coefficient of the jth layer.

[0192] Then calculate the magnitude of the detail coefficient:

[0193] when When , we consider this point as a mutation point. Where θ is the threshold, which can be set based on historical data or statistical characteristics.

[0194] Satisfy all The time t is found and arranged in chronological order to detect the on / off points of the target data of the boosting segment.

[0195] The step 5 includes the following:

[0196] Step 5.1: Crop the data that has been key point detected and only keep the boost segment;

[0197] Generally speaking, the starting and ending points of the boost segment are the first power-on point and the last power-off point in the key point detection, and the target boost segment contains rich semantic information. Different types of targets have different infrared radiation energy characteristic curves and space-time characteristic curves in the boost segment, including the peak and waveform of the infrared radiation energy characteristic curve, the duration of the boost segment, the trend and peak of the space-time characteristic curve, etc. It is feasible and more accurate to identify the target type based on the boost segment data containing rich information. Therefore, this method extracts the boost segment of the data according to the detected key points for type recognition.

[0198] Step 5.2, calculating the waveform similarity between the boost segment target data and the data in the type template library;

[0199] The boost phase target data extracted in step 5.1 is matched against the target type template library. Each data entry in the target type template library represents a type, and the data in the template library are the most representative data of each type. Therefore, the extracted boost phase target data and the data in the template library are used to calculate the similarity of the infrared radiation energy characteristic curve waveform.

[0200] First, assume that the extracted boost phase target infrared radiation energy characteristic curve sequence is x=[x1, x2,…, x t ], the dataset in the target type template library is Φ={Y1,Y2,…,Y n}, where Y1 to Y n Refers to different types of data templates in the template library, mathematically expressed as Y = [y1, y2, ..., y t ], first recursively downsample the two time series:

[0201] X low =[x1,x 1+k , x 1+2k ,…] (5-1)

[0202] Y lo w=[y1,y 1+k ,y 1+2k ,…] (5-2)

[0203] where Y low is the template data after recursive downsampling, X low is the infrared radiation energy characteristic curve sequence after recursive downsampling, k refers to the recursive downsampling time interval, and the waveform similarity is calculated at low resolution after recursive downsampling of the time series. Here, the downsampling sequence X is calculated. low and Y low The optimal alignment path Path low First, define the cumulative distance D(i, j) low :

[0204] D(i, j) low =d(x i ,y i ) low +min{D(i-1, j) low , D(i, j-1) low , D(i-1, j-1) low}(5-3)

[0205] where d(x i ,y j ) low =|x i -y j |2 is the Euclidean local distance. Assuming that the two time series have n and m points respectively after downsampling, a n×m cumulative distance matrix is constructed, where each element represents x i and y j The distance, D(i, j) low is the value of the i, j element in the cumulative distance matrix, which represents the minimum path cost from the starting point to (i, j). The boundary conditions of the search process are:

[0206] D(0,0) low =0, D(i, 0) low =∞, D(0, j) low =∞(i, j>0) (5-4)

[0207] When the lower right corner (n, m) of the cumulative distance matrix is searched, the Path in the low-resolution case is obtained. low At this time, the path expansion operation is performed to expand the low-resolution path Path low Mapping back to high resolution, each path is expanded into a set of neighborhood points. For example, if the low-resolution path point is (i, j), it is expanded to {(ki, kj), (ki+1, kj), (ki, kj+1), (ki+1, kj+1)} in high resolution, where k is the recursive downsampling interval. D(i, j) is recalculated in the expanded high-resolution path neighborhood to calculate the waveform similarity between the two time series.

[0208] Step 5.3: Match and identify target types based on waveform similarity;

[0209] When the waveform similarity between the target and a certain type of template is higher than the set threshold, the type is included in the matching template group with high similarity. After the waveform similarity calculation between the target and all types of templates is completed, the type with the highest waveform similarity in the matching template group with high similarity is taken as the type recognition result and output.

[0210] If the waveform similarity calculation results between the target and all templates are lower than the set threshold, a new type of template is created and added to the target type template library.

[0211] The proposed method for identifying high-time-sensitive targets combined with boost-phase guidance significantly improves the accuracy and efficiency of identifying high-time-sensitive targets in complex scenarios through multi-level feature fusion, dynamic detection mechanisms, and physical law verification. This method offers the following advantages:

[0212] 1. Boosting feature focus and efficient detection

[0213] To address the redundancy of traditional full-time signal analysis methods, this method uses a dual-mode time series signal collaborative analysis of infrared radiation energy curves and space-time characteristic curves. This method focuses on the unique characteristics of strong infrared radiation persistence during the boost phase (such as energy intensity thresholds), regular velocity variation characteristics (such as initial velocity screening and acceleration trend determination), and physical rationality verification (acceleration traversal verification), significantly improving the reliability of boost phase detection. Experiments have shown that compared to traditional full-time analysis methods, the false alarm rate is reduced, and boost phase positioning accuracy is improved to the millisecond level, providing a high-quality data foundation for subsequent classification.

[0214] 2. High-frequency feature preservation and noise suppression

[0215] A low-frequency filtering method guided by polynomial fitting balances noise removal with high-frequency retention through dynamic weight adjustment within a sliding window. This effectively suppresses background noise while fully preserving the sudden changes in key boost phase points (such as ignition and shutdown). Compared to conventional Gaussian filtering, this method improves key point detection accuracy and reduces signal waveform distortion, providing high-fidelity input for waveform similarity matching.

[0216] 3. Multi-level false alarm suppression and dynamic optimization mechanism

[0217] A four-fold false alarm filtering system is constructed through initial velocity threshold screening (excluding re-entry targets), dual-criteria detection of space-time characteristic curves (joint determination of rising trend proportion and amplitude), continuous energy intensity monitoring (strong energy window proportion threshold) and physical law error correction module (acceleration rationality verification). Combined with the energy-velocity accumulation monitoring module, the invalid detection process is terminated in real time, the system misjudgment rate is reduced, the average algorithm time consumption is shortened, and the real-time requirements of highly time-sensitive scenarios are met.

[0218] 4. Dynamic waveform matching and template generalization capabilities

[0219] A hybrid matching strategy based on recursive downsampling and dynamic time warping (DTW) algorithms addresses the sensitivity of traditional template libraries to timing scaling and phase shifts through low-resolution path screening and high-resolution neighborhood optimization. Combined with a self-expansion mechanism for unknown template types, the system improves its ability to generalize and classify new types of targets, increasing matching efficiency in measured data and adapting to the diverse targets found in complex battlefield environments.

[0220] 5. Accurate detection of key events and data optimization

[0221] Using multi-scale wavelet decomposition and detail coefficient mutation detection, high-frequency component analysis was used to locate the on / off points of the boost segment, achieving a key point detection error of less than 2 frames (approximately 0.1 seconds). This boost segment data cropping strategy eliminated 70% of redundant data, allowing the classification algorithm to focus on high-value feature segments, improving recognition accuracy while reducing computing resource consumption.

[0222] 6. Special target adaptive detection capability

[0223] The system integrates a scalable library of special target detection algorithms, dynamically invoking targeted detection modules through feature parameter matching to address the issue of missed detection of heterogeneous targets (such as those with variable thrust engines and those with stealth coatings). Field tests have shown that the detection coverage of known special targets has increased from 68% with traditional methods to 95%, supporting online algorithm updates and rapid adaptation to tactical scenarios.

[0224] In summary, through four core innovations: feature focusing, multi-level verification, dynamic optimization, and adaptive expansion, the present invention achieves accurate and efficient processing of the entire process of "detection-positioning-classification" of highly time-sensitive targets in a complex infrared interference environment, providing reliable technical support for key tasks such as anti-missile warning and target interception.

Claims

1. A highly time-sensitive target recognition method combined with boost phase detection, characterized by: The steps include: Step 1: Extract the infrared radiation energy time series signal and space-time characteristic time series signal of the high time-sensitive target from the infrared image sequence; Step 2: Perform outlier processing and noise filtering on the time series signal obtained in step 1 to retain the high-frequency characteristics of the signal; Step 3: Based on the initial velocity screening of the space-time characteristic curve and the elimination of non-boost phase targets, the upward trend, amplitude and infrared radiation energy intensity of the space-time characteristic curve are analyzed through a sliding window to complete the boost phase detection; Step 3.1 Perform initial velocity screening on the time series signal processed in step 2 according to the space-time characteristic curve to distinguish between reentry targets and high-time-sensitive targets; Step 3.2 Perform boost segment detection on the timing signal filtered in step 3.1; Step 3.3: The space-time characteristic curve meets the target of upward trend and the upward range is further determined; Step 3.5: Conduct targeted testing on highly time-sensitive special targets; Step 3.6 Design monitoring module to optimize algorithm performance; Step 3.7: Design an error correction module to perform a final error correction check on the detected boost phase targets to ensure the accuracy of boost phase recognition. Step 4: Perform key point detection on the signal containing the boost segment detected in step 3. Subsequently, it is necessary to intercept the boost segment data based on the power-on point and power-off point for identification; Step 5: Utilize recursive downsampling and dynamic time warping algorithms to calculate the waveform similarity between the boost segment data and the data in the type template library, and output the target type recognition result based on the highest similarity.

2. The high-time-sensitive target recognition method combined with boost phase detection according to claim 1, characterized in that: The step 1 comprises the following steps: Calculate the sum of pixel intensities in the target area of the infrared image to form an infrared radiation energy time series signal, that is, obtain the infrared radiation energy characteristic curve ; Calculate the target center of mass coordinates frame by frame, deduce the target speed based on the changes in the center of mass coordinates of adjacent frames, and generate a space-time characteristic curve ; The calculation formula for infrared radiation energy is: in The target area The pixel intensity value within represents the radiance; The calculation formula of space-time characteristic curve speed is: The center of mass of the target in the infrared image is ( ), and The goal is The coordinates of the center of mass of the frame, is the time interval between adjacent frames.

3. The high-time-sensitive target recognition method combined with boost phase detection according to claim 1, characterized in that: The step 2 comprises the following steps: Step 2.1: Process missing values of the extracted time series signal; Step 2.2 uses the low-frequency filtering method guided by polynomial fitting to remove noise from the time series signal that has processed missing values. The signal is , after selecting the window size and polynomial order in each sliding window, the signal is fitted, and then the fitted value is used to replace the value of the window center point. The calculation formula is as follows: In the formula is the signal value after filtering, is the filter weight, is the value in the original signal, where the filter weights are given by the polynomial order With window size decided.

4. The high-time-sensitive target recognition method combined with boost phase detection according to claim 1 is characterized in that: Initial velocity screening in step 3.1 includes: Set the initial velocity threshold , if the target initial velocity , it is determined to be a non-boost segment target; otherwise, it enters the boost segment detection process; Step 3.2 specifically includes: First, the space-time characteristic curve characteristics of the high time-sensitive target. Assuming that the space-time characteristic curve is , the curve change trend within the window is calculated in the form of a sliding window, and the length of the sliding window is tentatively set to , the curve in the window is , for the curve within the window we calculate its first-order difference: Then calculate the curve rise coefficient within the window: when Greater than or equal to the space-time characteristic curve rising ratio threshold set in the boost segment detection module When , the target curve will enter the next step of detection; Step 3.3 specifically includes: Detect and determine the rising amplitude of the space-time characteristic curve , only when the increase is greater than the threshold It is considered that this is acceleration caused by the engine, not a false alarm, and the infrared radiation energy intensity is determined; Step 3.4 specifically includes: The infrared radiation energy characteristic curve is , the sliding window framework is used for detection, and the window length is , when the strong energy ratio Greater than the threshold within the window When the window is considered to be in a state of strong energy radiation, the strong energy ratio coefficient The specific calculation formula is as follows: In the formula , is the strong energy ratio, A unit of measurement for strong energy. Set a threshold for strong energy when Greater than or equal to When , it is considered that the target is in a strong energy radiation state at this point. Equal to 1, then calculate ,when Greater than the set threshold When , the target in the window is considered to be in a strong energy radiation state. At this time, the target's space-time characteristic curve and infrared radiation energy characteristic curve have met the boost phase characteristics and passed the boost phase detection, and the detection result is returned; Step 3.5 specifically includes: The boost phase detection in step 3 further includes targeted detection of special targets, and the specific steps are as follows: Step 3.5.

1. Pre-establish a special target detection algorithm: For highly time-sensitive targets with special infrared radiation or motion characteristics, pre-design a corresponding targeted detection algorithm and integrate the algorithm into the special target detection algorithm library. Among them Indicates the Detection algorithm for special targets; Step 3.5.

2. Call the algorithm library for detection: For the targets screened in step 3.1, match the corresponding detection algorithm in the special target detection algorithm library based on the characteristic parameters of its infrared radiation energy characteristic curve and space-time characteristic curve, and call the detection algorithm in real time to perform boost phase detection; Step 3.5.

3. Verification and Result Determination: If the detection results output by the matched detection algorithm meet the preset boost phase characteristics, the target is determined to be a special type boost phase target. Otherwise, proceed to step 3.2 to match other algorithms or continue with the subsequent detection process. Step 3.6 specifically includes: Step 3.6.

1. Real-time monitoring of energy and speed levels: When the signal within the sliding window does not meet the boost phase state, slide the window backward at a preset step size and record the energy and speed levels at the current window center point, where: Refers to the infrared radiation energy characteristic curve value at the center point of the window. and are the rising proportion coefficient and rising amplitude coefficient of the space-time characteristic curve mentioned above, 、 and The thresholds set in the boost segment detection module respectively; Step 3.6.

2. Cumulative judgment and process termination: statistics and When any of the following conditions is met, the target is determined to be a non-boost segment target and the detection process is terminated: The cumulative number of times indicates that the target is continuously The energy level in the window is lower than ;or The cumulative number of times indicates that the target is continuously The speed level within the window does not meet the characteristics of the boost phase; Step 3.7 specifically includes: Step 3.7.1 Acceleration traversal calculation: For the target determined to be in the boost phase, extract the continuous velocity value of its space-time characteristic curve within the sliding window and calculate the instantaneous acceleration at each moment. in, For the window The speed value of the frame, is the time interval between adjacent frames; Step 3.7.2 Verification of physical laws: Traverse all accelerations in the window Forming an acceleration set , if there is any ,in is a preset physical rationality threshold, the target is determined to be a false alarm with abnormal space-time characteristics; Step 3.7.3 Correction of detection results: Exclude the boost phase detection results for targets determined to be false alarms and return to step 3.6 to re-execute the detection process.

5. The high-time-sensitive target recognition method combined with boost phase detection according to claim 1 is characterized in that: The step 4 comprises the following steps: Step 4.1 Multi-scale signal decomposition: Perform multi-scale decomposition on the detected boost segment signal and define the function in is the scale parameter, The signal is decomposed into approximate coefficients using high-pass and low-pass filters. Low frequency components and detail coefficients High frequency components, and complete Layer decomposition: Indicates the upper limit of decomposition levels, Indicates the actual number of decomposition layers; Step 4.2 Detail coefficient mutation point detection: calculate the amplitude of each layer detail coefficient when Time, determine the moment is the mutation point, where It is a threshold value preset based on historical data statistics or experience; Step 4.3: Extract key points: Based on the temporal distribution of the mutation points, select the boost segment power-on and power-off points that form highly time-sensitive targets, and intercept the boost segment data based on the power-on and power-off points.

6. The high-time-sensitive target recognition method combined with boost phase detection according to claim 1, characterized in that: The step 5 comprises the following steps: Step 5.1 Boost segment data clipping: Based on the key points detected in step 4, the boost segment data from the first power-on point to the last power-off point in the infrared radiation energy characteristic curve is intercepted; Step 5.2, recursive downsampling alignment: recursively downsample the boost segment data with the template data in the type template library to generate a low-resolution sequence and , where the downsampling interval is ; Step 5.3 Dynamic Time Warping Matching: Calculate the cumulative distance matrix on the low-resolution sequence , through the recursive formula: Determine the optimal alignment path , and extend the path to the high-resolution neighborhood for optimization to obtain the waveform similarity between the target data and the template data; Step 5.4 Type identification and template update: If the waveform similarity is higher than the preset threshold, the template type with the highest matching degree is output as the identification result; if the similarity of all templates is lower than the threshold, the boosting segment data is added as an unknown type template in the template library.

Images