Infrared target clustering identification method and system based on multi-feature fusion

Abstract

The invention provides an infrared target clustering recognition method and system based on multi-feature fusion. The infrared target clustering recognition method comprises the steps that 1, multi-feature extraction of target motion features and spectral features is completed based on a target observation radiation sequence; 2, target rough classification is carried out according to a multi-feature fusion function, and the target recognition range is narrowed; and step 3, target type identification is completed based on a KANN-DBSCAN clustering identification algorithm of multiband characteristics. According to the method, on the basis of traditional target recognition, recognition information of multi-dimensional features is introduced, the robustness of a target recognition algorithm is improved, meanwhile, the fault-tolerant capability of a clustering recognition model is fully utilized, and the defect that the recognition rate is low due to single recognition information is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared target recognition, and in particular to an infrared target cluster recognition method and system based on multi-feature fusion. Background Art

[0002] Infrared target recognition refers to the use of infrared images to detect a single target or a group of targets without human intervention and analyze the acquired information. In particular, infrared target clustering recognition based on multi-feature fusion can determine the target type in a relatively short time.

[0003] Patent document CN117095244A discloses an infrared target recognition method, device, equipment and medium, which obtains an infrared image data set collected by an infrared camera and calls a trained YOLOv5 target recognition model to recognize and process the data set. This patent is a target recognition method based on trained neural network learning.

[0004] Patent document CN117115599A discloses a small sample target recognition method for infrared images, which constructs a small sample infrared data set based on infrared target recognition data, combines the related category visible light target samples to form a visible light data set, constructs a Bayesian convolutional neural network as a target recognition network, uses the visible light data set to train the neural network until the loss converges, uses the infrared data set to adaptively adjust the Bayesian network parameters, and finally uses the Bayesian network to recognize the sample target. This patent is a target recognition method based on small sample transfer learning.

[0005] Patent document CN117197547A discloses a few-sample infrared image target recognition method based on concept learners. The recognition model includes a concept learner module and a similarity metric module. The concept learner module learns the unique concept embedding of each concept dimension. The similarity module compares the concept prototype with the concept embedding and assigns an importance score to each concept dimension as a basis for classification. This patent is a few-sample learning method based on different concept learners and concept prototype information.

[0006] Patent document CN116087974A discloses a space-based infrared target recognition system and a target recognition method thereof, which collects images of targets based on a space-based infrared remote sensing camera, sends a laser beam to a flying target based on a laser rangefinder and receives an echo signal, and determines the target type in combination with orbit information. This patent is a target classification method based on remote sensing camera imaging and laser ranging.

[0007] Patent document CN115422968A discloses an infrared target recognition method based on a multi-channel convolutional neural network, which recognizes the pre-processed target infrared radiation sequence based on a loss function by constructing a multi-channel convolutional neural network. This patent discloses a multi-channel target recognition method based on a convolutional neural network.

[0008] Space target recognition is a dynamic process from coarse to fine and gradually refined. The single-dimensional target recognition features disclosed in the above patent documents cannot effectively guarantee the recognition accuracy. Therefore, how to select the optimal classifier to identify the features is still a difficult problem for infrared target recognition. Summary of the invention

[0009] In view of the defects in the prior art, the object of the present invention is to provide an infrared target clustering recognition method and system based on multi-feature fusion.

[0010] The infrared target clustering recognition method based on multi-feature fusion provided by the present invention includes:

[0011] Step 1: Extract target motion features and spectral features based on the target observed radiation sequence;

[0012] Step 2: Roughly classify the target according to the multi-feature fusion function to narrow the target recognition range;

[0013] Step 3: KANN-DBSCAN clustering recognition algorithm based on multi-band features is used to complete target type recognition.

[0014] Preferably, the target motion characteristics include the target launch position, spatial three-dimensional trajectory, speed change characteristics and shutdown point characteristics; after the target takes off to a preset height, it is captured by the space-based detection system, and the launch point position is obtained by inverting the initial pixel coordinates of the target on the image plane. The target motion equation at any time is:

[0015]

[0016] in, μ e is the earth's gravitational constant, J2 is the second-order harmonic coefficient of the earth's gravitational field, r is the radius of the earth, [XYZV X V Y V Z ] is the position and velocity state vector of the target.

[0017] Preferably, the target spectral feature has fingerprint characteristics, and the spectral information of the object is analyzed as an important basis for obtaining the inherent attribute of the target; due to the surface material of the engine, the radiation spectrum distribution of different types of targets is different; under the infrared multi-band detection system, the dual-band colorimetric temperature measurement method is used to estimate the temperature of the point target, and the point target outside the atmosphere is regarded as a gray body. The target estimated temperature value T(λi ,λ j )for:

[0018]

[0019] Where: c2 is Planck's constant, I(λ i ,T) is the target temperature T in the spectrum λ i The radiation intensity generated, I(λ j ,T) is the target temperature T in the spectrum λ j the intensity of the radiation produced;

[0020] According to the estimated target temperature, the effective radiation area ε(λ i ,T) is:

[0021]

[0022] Among them: A p is the projected area of ​​the target in the observation direction, τ a (λ i ) is the spectrum band λ i The corresponding emissivity; the wavelength of the maximum energy of the target surface light radiation is determined by the emissivity and temperature of the object surface material, so different types of targets are identified by the spectral distribution characteristics of the spectral radiation.

[0023] Preferably, the multi-feature fusion function is:

[0024]

[0025] Among them, M i are two types of features: motion and spectrum, ω i is the weight coefficient, and the weights corresponding to motion features and spectral features are ω1 and ω2 respectively.

[0026] Preferably, the clustering process of the KANN-DBSCAN clustering recognition algorithm is as follows: select a data object point p from a given data set; if the selected data object point p is a core point for the parameters of the neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts, then find all the data object points that are density-reachable from p and create a cluster cluster with p as the core; if the selected data object point p is an edge point, then select another data object point until all points are processed;

[0027] Based on the KANN-DBSCAN clustering recognition algorithm, the spatial distribution characteristics of the data set are utilized, and a parameter list of two parameters is generated based on the K-average nearest neighbor algorithm and mathematical expectation. The K-average nearest neighbor algorithm is based on the closest distance between each point in the data set and the Kth nearest data point, and the K-average nearest distance of all points is averaged to obtain the K-average nearest neighbor distance of the data set.

[0028] First, calculate the Euclidean distance from each point to other points, which is the distance matrix Dist n×n ;

[0029] Dist n×n ={dist(i,j),1≤i≤n,1≤j≤n}

[0030] Where dist(i,j) is the distance between point i and point j, and n is the number of data points;

[0031] Secondly, Dist n×n Sort each row of elements and sort the sorted Dist n×n The mean of the Kth column is calculated. 1≤K≤n, forming the parameter list Eps_list of Eps:

[0032]

[0033] Calculate MinPts_list based on Eps_list:

[0034]

[0035] Among them, P i is the number of neighboring objects in the Eps neighborhood of the i-th object;

[0036] As the number of clusters in the clustering results converges, Eps_list and MinPts_list are continuously iterated to obtain the optimal neighborhood radius Eps opt And the threshold value of the number of data objects in the neighborhood MinPts opt When the results of three consecutive clustering clusters are consistent, the neighborhood radius and the threshold of the number of data objects in the neighborhood are determined to be Eps. opt and MinPts opt .

[0037] The infrared target clustering recognition system based on multi-feature fusion provided by the present invention includes:

[0038] Module M1: Extract target motion features and spectral features based on the target observed radiation sequence;

[0039] Module M2: Roughly classify the target according to the multi-feature fusion function to narrow the target recognition range;

[0040] Module M3: KANN-DBSCAN clustering recognition algorithm based on multi-band features to complete target type recognition.

[0041] Preferably, the target motion characteristics include the target launch position, spatial three-dimensional trajectory, speed change characteristics and shutdown point characteristics; after the target takes off to a preset height, it is captured by the space-based detection system, and the launch point position is obtained by inverting the initial pixel coordinates of the target on the image plane. The target motion equation at any time is:

[0042]

[0043] in, μ e is the earth's gravitational constant, J2 is the second-order harmonic coefficient of the earth's gravitational field, r is the radius of the earth, [XYZV X V Y V Z ] is the position and velocity state vector of the target.

[0044] Preferably, the target spectral feature has fingerprint characteristics, and the spectral information of the object is analyzed as an important basis for obtaining the inherent attribute of the target; due to the surface material of the engine, the radiation spectrum distribution of different types of targets is different; under the infrared multi-band detection system, the dual-band colorimetric temperature measurement method is used to estimate the temperature of the point target, and the point target outside the atmosphere is regarded as a gray body. The target estimated temperature value T(λ i ,λ j )for:

[0045]

[0046] Where: c2 is Planck's constant, I(λ i ,T) is the target temperature T in the spectrum λ i The radiation intensity generated, I(λ j ,T) is the target temperature T in the spectrum λ j the intensity of the radiation produced;

[0047] According to the estimated target temperature, the effective radiation area ε(λ i ,T) is:

[0048]

[0049] Among them: A p is the projected area of ​​the target in the observation direction, τ a (λ i ) is the spectrum band λ iThe corresponding emissivity; the wavelength of the maximum energy of the target surface light radiation is determined by the emissivity and temperature of the object surface material, so different types of targets are identified by the spectral distribution characteristics of the spectral radiation.

[0050] Preferably, the multi-feature fusion function is:

[0051]

[0052] Among them, M i are two types of features: motion and spectrum, ω i is the weight coefficient, and the weights corresponding to motion features and spectral features are ω1 and ω2 respectively.

[0053] Preferably, the clustering process of the KANN-DBSCAN clustering recognition algorithm is as follows: select a data object point p from a given data set; if the selected data object point p is a core point for the parameters of the neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts, then find all the data object points that are density-reachable from p and create a cluster cluster with p as the core; if the selected data object point p is an edge point, then select another data object point until all points are processed;

[0054] Based on the KANN-DBSCAN clustering recognition algorithm, the spatial distribution characteristics of the data set are utilized, and a parameter list of two parameters is generated based on the K-average nearest neighbor algorithm and mathematical expectation. The K-average nearest neighbor algorithm is based on the closest distance between each point in the data set and the Kth nearest data point, and the K-average nearest distance of all points is averaged to obtain the K-average nearest neighbor distance of the data set.

[0055] First, calculate the Euclidean distance from each point to other points, which is the distance matrix Dist n×n ;

[0056] Dist n×n ={dist(i,j),1≤i≤n,1≤j≤n}

[0057] Where dist(i,j) is the distance between point i and point j, and n is the number of data points;

[0058] Secondly, Dist n×n Sort each row of elements and sort the sorted Dist n×n The mean of the Kth column is calculated. 1≤K≤n, forming the parameter list Eps_list of Eps:

[0059]

[0060] Calculate MinPts_list based on Eps_list:

[0061]

[0062] Among them, P i is the number of neighboring objects in the Eps neighborhood of the i-th object;

[0063] As the number of clusters in the clustering results converges, Eps_list and MinPts_list are continuously iterated to obtain the optimal neighborhood radius Eps opt And the threshold value of the number of data objects in the neighborhood MinPts opt When the results of three consecutive clustering clusters are consistent, the neighborhood radius and the threshold of the number of data objects in the neighborhood are determined to be Eps. opt and MinPts opt .

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention provides an infrared target clustering recognition method based on multi-feature fusion. On the basis of traditional target recognition, the recognition information of multi-dimensional features is introduced to increase the robustness of the target recognition algorithm. The fault tolerance of the clustering recognition model is fully utilized to improve the defect of low recognition rate caused by single recognition information. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0067] Figure 1 It is a schematic diagram of the target clustering recognition process based on multi-feature fusion;

[0068] Figure 2 This is a schematic diagram of the target dual-band feature space clustering;

[0069] Figure 3 Identify test results for target classification. DETAILED DESCRIPTION

[0070] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0071] Example

[0072] The present invention provides an infrared target cluster recognition system based on multi-feature fusion, which includes three modules: a target feature extraction module, a target recognition coarse classification module, and a target cluster recognition fine classification module. On the basis of traditional target recognition, the present invention introduces the recognition information of multi-dimensional features to increase the robustness of the target recognition algorithm, and at the same time fully utilizes the fault tolerance of the cluster recognition model to improve the defect of low recognition rate caused by single recognition information.

[0073] The target feature extraction module includes two submodules, a target motion feature extraction module and a target spectrum feature extraction module. The target motion feature extraction module is used to realize motion trajectory feature calculation and narrow the target type range by using target flight features; the target spectrum feature extraction module is used to complete the target temperature and target effective radiation area estimation and assist in the rough identification of target types.

[0074] The target spectral feature extraction module considers that the target has radiation characteristics that are different from other attribute objects in the main atmospheric absorption window band. The wavelength of the target surface light radiation with maximum energy is determined by the emissivity and temperature of the surface material of the object; therefore, different types of targets can be identified by the spectral distribution characteristics of the spectral radiation.

[0075] The target recognition coarse classification module is based on the target feature extraction module, assigns different weight coefficients to multi-dimensional features, and completes coarse classification in combination with feature fusion function.

[0076] The target clustering recognition fine classification module is based on the target recognition coarse classification module and uses the KANN-DBSCAN clustering recognition algorithm based on multi-band features to complete target type recognition.

[0077] The KANN-DBSCAN clustering recognition algorithm based on multi-band features is a density-based clustering algorithm based on high-density connected areas. It can divide areas with sufficiently high density into clusters and find clusters of arbitrary shapes in noisy data. The purpose of DBSCAN is to find the largest set of density-connected objects.

[0078] like Figure 1 to Figure 3 As shown, the specific implementation steps of the present invention are as follows:

[0079] Step 1: Extract target motion features and spectral features based on the target observed radiation sequence. Target motion features include target launch position, spatial three-dimensional trajectory, velocity change features, and shutdown point features. After the target takes off to a certain height, it is captured by the space-based detection system. The launch point position is obtained by inverting the initial pixel coordinates of the target on the image plane. The target motion equation at any time is:

[0080]

[0081] in, μ e is the earth's gravitational constant, and J2 is the second-order harmonic coefficient of the earth's gravitational field. X V Y V Z ] is the position and velocity state vector of the target.

[0082] The target spectral characteristics have fingerprint characteristics. By analyzing the spectral information of the object, it can be used as an important basis for obtaining the inherent properties of the target. Due to factors such as the surface material of the engine, the radiation spectrum distribution of different types of targets is different. In the infrared multi-band detection system, the dual-band colorimetric temperature measurement method is an effective method for estimating the temperature of point targets; considering the point target outside the atmosphere as a gray body, the target estimated temperature value T(λ i ,λ j )for:

[0083]

[0084] Where: c2 is Planck's constant, I(λ i ,T) is the target temperature T in the spectrum λ i The radiation intensity generated, I(λ j ,T) is the target temperature T in the spectrum λ j The intensity of the radiation produced.

[0085] According to the estimated target temperature, the effective radiation area ε(λ i ,T) is:

[0086]

[0087] Among them: A p is the projected area of ​​the target in the observation direction, τ a (λ i ) is the spectrum band λ i The wavelength of the target surface light radiation with maximum energy is determined by the emissivity and temperature of the surface material of the object; therefore, different types of targets can be identified by the spectral distribution characteristics of the spectral radiation.

[0088] Step 2: Roughly classify the target according to the multi-feature fusion function to narrow the target recognition range. According to the feature fusion function:

[0089]

[0090] Among them, M i are two types of features (motion, spectrum), ω i is the weight coefficient, and the weights corresponding to motion features and spectral features are ω1 and ω2 respectively;

[0091] Step 3: KANN-DBSCAN clustering recognition algorithm based on multi-band features is used to complete target type recognition. The DBSCAN algorithm clustering process is as follows:

[0092] 1) Select a data object point p from a given data set;

[0093] 2) If the selected data object point p is a core point for the parameters of the neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts, then find all the data object points that are density-reachable from p and create a cluster with p as the core;

[0094] 3) If the selected data object point p is an edge point, select another data object point;

[0095] 4) Repeat steps 2) and 3) until all points are processed.

[0096] Since the neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts have a great influence on the final calculation result of the algorithm, in order to avoid the uncertainty caused by human selection, the present invention is based on the KANN-DBSCAN algorithm, utilizes the spatial distribution characteristics of the data set itself, and generates a parameter list of the two parameters based on the K-average nearest neighbor algorithm and mathematical expectation.

[0097] The K-average nearest neighbor algorithm is based on the nearest distance between each point in the data set and the Kth nearest data point, and the average of the nearest distances of all points is used to obtain the K-average nearest neighbor distance of the data set.

[0098] First, calculate the Euclidean distance from each point to other points, which is the distance matrix Dist n×n ;

[0099] Dist n×n ={dist(i,j),1≤i≤n,1≤j≤n}

[0100] Among them, dist(i,j) is the distance between point i and point j, and n is the number of data points.

[0101] Secondly, Dist n×n Sort each row of elements and sort the sorted Dist n×n The mean of the Kth column (1≤K≤n) is calculated. Form the parameter list Eps_list of Eps:

[0102]

[0103] Accordingly, MinPts_list is calculated based on Eps_list:

[0104]

[0105] Among them, P i is the number of neighboring objects in the Eps neighborhood of the i-th object.

[0106] As the number of clusters in the clustering results converges, Eps_list and MinPts_list are continuously iterated to obtain the optimal neighborhood radius Eps opt And the threshold value of the number of data objects in the neighborhood MinPts opt This paper stipulates that when the results of three consecutive clustering clusters are consistent, the neighborhood radius and the threshold of the number of data objects in the neighborhood of K are considered to be Eps. opt and MinPts opt .

[0107] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0108] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for infrared target clustering recognition based on multi-feature fusion, characterized in that: include: Step 1: Extract target motion features and spectral features based on the target observed radiation sequence; Step 2: Roughly classify the target according to the multi-feature fusion function to narrow the target recognition range; Step 3: KANN-DBSCAN clustering recognition algorithm based on multi-band features is used to complete target type recognition.

2. The infrared target clustering recognition method based on multi-feature fusion according to claim 1 is characterized in that: The target motion characteristics include the target launch position, three-dimensional trajectory in space, speed change characteristics and shutdown point characteristics. After the target takes off to a preset height, it is captured by the space-based detection system. The launch point position is obtained by inverting the initial pixel coordinates of the target on the image plane. The target motion equation at any time is: in, μ e is the earth's gravitational constant, J2 is the second-order harmonic coefficient of the earth's gravitational field, r is the radius of the earth, [XYZV X V Y V Z ] is the position and velocity state vector of the target.

3. The infrared target clustering recognition method based on multi-feature fusion according to claim 1 is characterized in that: The target spectrum features have fingerprint characteristics. By analyzing the spectral information of the object, it is used as an important basis for obtaining the inherent properties of the target. Due to the surface material factors of the engine, the radiation spectrum distribution of different types of targets is different. In the infrared multi-band detection system, the dual-band colorimetric temperature measurement method is used to estimate the temperature of the point target. The point target outside the atmosphere is regarded as a gray body, and the target estimated temperature value T(λ i ,λ j )for: Where: c2 is Planck's constant, I(λ i ,T) is the target temperature T in the spectrum λ i The radiation intensity generated, I(λ j ,T) is the target temperature T in the spectrum λ j the intensity of the radiation produced; According to the estimated target temperature, the effective radiation area ε(λ i ,T) is: Among them: A p is the projected area of ​​the target in the observation direction, τa(λi) is the spectrum λ i The corresponding emissivity; the wavelength of the maximum energy of the target surface light radiation is determined by the emissivity and temperature of the object surface material, so different types of targets are identified by the spectral distribution characteristics of the spectral radiation.

4. The infrared target clustering recognition method based on multi-feature fusion according to claim 1 is characterized in that: The multi-feature fusion function is: Among them, M i are two types of features: motion and spectrum, ω i is the weight coefficient, and the weights corresponding to motion features and spectral features are ω1 and ω2 respectively.

5. The infrared target clustering recognition method based on multi-feature fusion according to claim 1 is characterized in that: The clustering process of the KANN-DBSCAN clustering recognition algorithm is as follows: select a data object point p from a given data set; if the selected data object point p is a core point for the parameter neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts, then find all data object points that are density-reachable from p and create a cluster with p as the core; if the selected data object point p is an edge point, then select another data object point until all points are processed; Based on the KANN-DBSCAN clustering recognition algorithm, the spatial distribution characteristics of the data set are utilized, and the parameter list of the two parameters is generated based on the K-average nearest neighbor algorithm and mathematical expectation. The K-average nearest neighbor algorithm is based on the closest distance between each point in the data set and the Kth nearest data point, and the average of the nearest distances of all points is used to obtain the K-average nearest neighbor distance of the data set; First, calculate the Euclidean distance from each point to other points, which is the distance matrix Dist n×n ; Dist n×n ={dist(i,j),1≤i≤n,1≤j≤n} Where dist(i,j) is the distance between point i and point j, and n is the number of data points; Secondly, Dist n×n Sort each row of elements and sort the sorted Dist n×n The mean of the Kth column is calculated. 1≤K≤n, forming the parameter list Eps_list of Eps: Calculate MinPts_list based on Eps_list: Among them, P i is the number of neighboring objects in the Eps neighborhood of the i-th object; As the number of clusters in the clustering results converges, Eps_list and MinPts_list are continuously iterated to obtain the optimal neighborhood radius Eps opt And the threshold value of the number of data objects in the neighborhood MinPts opt When the results of three consecutive clustering clusters are consistent, the neighborhood radius and the threshold of the number of data objects in the neighborhood are determined to be Eps. opt and MinPts opt .

6. An infrared target clustering recognition system based on multi-feature fusion, characterized in that: include: Module M1: Extract target motion features and spectral features based on the target observed radiation sequence; Module M2: Roughly classify the target according to the multi-feature fusion function to narrow the target recognition range; Module M3: KANN-DBSCAN clustering recognition algorithm based on multi-band features to complete target type recognition.

7. The infrared target clustering recognition system based on multi-feature fusion according to claim 6 is characterized in that: The target motion characteristics include the target launch position, three-dimensional trajectory in space, speed change characteristics and shutdown point characteristics. After the target takes off to a preset height, it is captured by the space-based detection system. The launch point position is obtained by inverting the initial pixel coordinates of the target on the image plane. The target motion equation at any time is: in, μ e is the earth's gravitational constant, J2 is the second-order harmonic coefficient of the earth's gravitational field, r is the radius of the earth, [XYZV X V Y V Z ] is the position and velocity state vector of the target.

8. The infrared target clustering recognition system based on multi-feature fusion according to claim 6 is characterized in that: The target spectrum features have fingerprint characteristics. By analyzing the spectral information of the object, it is used as an important basis for obtaining the inherent properties of the target. Due to the surface material factors of the engine, the radiation spectrum distribution of different types of targets is different. In the infrared multi-band detection system, the dual-band colorimetric temperature measurement method is used to estimate the temperature of the point target. The point target outside the atmosphere is regarded as a gray body, and the target estimated temperature value T(λ i ,λ j )for: Where: c2 is Planck's constant, I(λ i ,T) is the target temperature T in the spectrum λ i The radiation intensity generated, I(λ j ,T) is the target temperature T in the spectrum λ j the intensity of the radiation produced; According to the estimated target temperature, the effective radiation area ε(λ i ,T) is: Among them: A p is the projected area of ​​the target in the observation direction, τa(λi) is the spectrum λ i The corresponding emissivity; the wavelength of the maximum energy of the target surface light radiation is determined by the emissivity and temperature of the object surface material, so different types of targets are identified by the spectral distribution characteristics of the spectral radiation.

9. The infrared target clustering recognition system based on multi-feature fusion according to claim 6 is characterized in that: The multi-feature fusion function is: Among them, M i are two types of features: motion and spectrum, ω i is the weight coefficient, and the weights corresponding to motion features and spectral features are ω1 and ω2 respectively.

10. The infrared target clustering recognition system based on multi-feature fusion according to claim 6 is characterized in that: The clustering process of the KANN-DBSCAN clustering recognition algorithm is as follows: select a data object point p from a given data set; if the selected data object point p is a core point for the parameter neighborhood radius Eps and the threshold value of the number of data objects in the neighborhood MinPts, then find all data object points that are density-reachable from p and create a cluster with p as the core; if the selected data object point p is an edge point, then select another data object point until all points are processed; Based on the KANN-DBSCAN clustering recognition algorithm, the spatial distribution characteristics of the data set are utilized, and the parameter list of the two parameters is generated based on the K-average nearest neighbor algorithm and mathematical expectation. The K-average nearest neighbor algorithm is based on the closest distance between each point in the data set and the Kth nearest data point, and the average of the nearest distances of all points is used to obtain the K-average nearest neighbor distance of the data set; First, calculate the Euclidean distance from each point to other points, which is the distance matrix Dist n×n ; Dist n×n ={dist(i,j),1≤i≤n,1≤j≤n} Where dist(i,j) is the distance between point i and point j, and n is the number of data points; Secondly, Dist n×n Sort each row of elements and sort the sorted Dist n×n The mean of the Kth column is calculated. 1≤K≤n, forming the parameter list Eps_list of Eps: Calculate MinPts_list based on Eps_list: Among them, P i is the number of neighboring objects in the Eps neighborhood of the i-th object; As the number of clusters in the clustering results converges, Eps_list and MinPts_list are continuously iterated to obtain the optimal neighborhood radius Eps opt And the threshold value of the number of data objects in the neighborhood MinPts opt When the results of three consecutive clustering clusters are consistent, the neighborhood radius and the threshold of the number of data objects in the neighborhood are determined to be Eps. opt and MinPts opt .

Images