A method, apparatus and readable storage medium for small target recognition in infrared images

Abstract

This invention belongs to the field of target recognition technology, and relates to a method, apparatus, and readable storage medium for small target recognition in infrared images. The method involves georegistration of the infrared image and optical image of the area to be detected to obtain the target infrared image and target optical image; clustering the pixels in the target optical image to obtain multiple pixel clusters; extracting the boundary pixels of the pixel clusters and converting the boundary pixels from raster to vector to obtain the segmentation boundary coordinates, thereby segmenting the target infrared image; fitting the feature values ​​of the pixels in each segmented infrared sub-image using the generalized extreme value distribution method to obtain a generalized extreme value distribution model; optimizing the generalized extreme value distribution model and obtaining the background model of the infrared sub-image based on the optimized generalized extreme value distribution model; using the background model of the infrared sub-image to perform target recognition on the infrared sub-image; and obtaining the small target recognition result of the infrared image based on the target recognition results of all infrared sub-images.

Description

Technical Field

[0001] This invention relates to the field of target recognition technology, and in particular to a method, apparatus and computer-readable storage medium for recognizing small targets in infrared images. Background Technology

[0002] With the development of UAV remote sensing technology, UAVs equipped with infrared sensors are playing an important role in environmental monitoring, ground object identification, and other fields. Due to their unique imaging characteristics, small targets in UAV infrared images often lack definite shape and texture information. In complex and ever-changing ground environments, small targets can be submerged by background noise, making them difficult to identify. Therefore, accurately identifying small targets in UAV infrared images is a problem that urgently needs to be solved.

[0003] Traditional infrared image small target recognition methods mainly fall into two categories. The first category is based on image filtering, which enhances the original infrared image by filtering to increase the difference between small targets and the background, making the small targets more prominent. Feature extraction and target detection are then performed on the processed infrared image to identify the small target regions. The second category is based on pattern recognition, which treats small target recognition as a classification problem. It determines whether each pixel in the enhanced infrared image belongs to the target class or the background class to achieve small target detection and recognition. Both methods typically rely on image enhancement and filtering, based on the assumption that the contrast between small targets and the background can be clearly distinguished through filtering enhancement or feature extraction. However, small targets in infrared images often have low contrast, and the background may contain complex noise, textures, or dynamic background changes. Therefore, when there is strong background interference or the small target has a high similarity to the background, processing methods relying solely on image filtering and pixel classification are difficult to accurately identify small targets in the image.

[0004] To address the aforementioned issues, existing technologies have proposed a target recognition method based on background statistical information. This method selects pixel samples from infrared images to estimate their Gaussian distribution, calculates their mean vector and covariance matrix, and then constructs a multivariate Gaussian distribution function. This multivariate Gaussian distribution function is used as a background model to describe the probability distribution of background pixels in the infrared image. A probability threshold is then empirically set, and the feature values ​​of each pixel in the infrared image are substituted into this background model. The probability value of the pixel belonging to the background is output. If the output probability value is less than the probability threshold, the pixel is determined to be a small target, thus achieving small target recognition in infrared images. By grasping the statistical characteristics of the background as a whole and gaining a deeper understanding of the distribution of background pixels in the feature space, rather than simply filtering and enhancing the image or classifying pixels in isolation, this method achieves higher recognition accuracy compared to traditional small target recognition methods. However, as the resolution of UAV infrared images improves, the background of small targets becomes increasingly complex, with a significant increase in background detail. Background pixels no longer conform to a single probability distribution, and simple probability distribution models cannot accurately describe the distribution characteristics of background pixels. Consequently, the background models constructed using these models have low accuracy, thus affecting the accuracy of small target recognition.

[0005] In summary, existing methods for small target recognition in UAV infrared images suffer from the problem that the constructed background model cannot accurately describe the distribution characteristics of background pixels in the infrared image, resulting in low accuracy of small target recognition results based on this background model. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the background model constructed by the existing UAV infrared image small target recognition method cannot accurately describe the distribution characteristics of background pixels in the infrared image, resulting in low accuracy of the small target recognition result based on the background model.

[0007] To address the aforementioned technical problems, this invention provides a method for recognizing small targets in infrared images, comprising:

[0008] Acquire an infrared image and an optical image of the area to be detected, and geo-register the infrared image and the optical image to obtain a target infrared image and a target optical image;

[0009] The pixels in the target optical image are clustered to obtain multiple pixel clusters; the boundary pixels of each pixel cluster are extracted using an edge detection algorithm, and the boundary pixels of each pixel cluster are converted from raster to vector to obtain the segmentation boundary coordinates corresponding to each pixel cluster.

[0010] The target infrared image is segmented based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images;

[0011] The feature values ​​of all pixels in each infrared sub-image are fitted using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image. The parameter likelihood function of the generalized extreme value distribution model is constructed based on the maximum likelihood estimation method. The generalized extreme value distribution model is optimized based on the parameter likelihood function. The background model of the infrared sub-image is obtained based on the optimized generalized extreme value distribution model.

[0012] The background model of each infrared sub-image is used to perform target recognition on each pixel in the infrared sub-image, and the small target recognition result of the infrared image is obtained based on the target recognition results of all infrared sub-images.

[0013] Preferably, the generalized extreme value distribution model of the infrared sub-image is expressed as:

[0014] ,

[0015] in, Indicates the first A generalized extreme value distribution model for each infrared sub-image; The feature value of a pixel; This represents the location parameters of the generalized extreme value distribution model; The scale parameter represents the generalized extreme value distribution model; Represents the shape parameters of the generalized extreme value distribution model;

[0016] The parameter likelihood function of the generalized extreme value distribution model is expressed as:

[0017] ,

[0018] in, Represents the generalized extreme value distribution model The parameter likelihood function; ; Indicates the first The number of pixels in each infrared sub-image; .

[0019] Preferably, optimizing the generalized extreme value distribution model based on the parameter likelihood function includes:

[0020] Calculate the first partial derivative of the parametric likelihood function with respect to the location parameters of the generalized extreme value distribution model, and set the first partial derivative to 0 to obtain the location parameter equation;

[0021] Calculate the second partial derivative of the parametric likelihood function with respect to the scaling parameters of the generalized extreme value distribution model, and set the second partial derivative to 0 to obtain the scaling parameter equation;

[0022] Calculate the third partial derivative of the parametric likelihood function with respect to the shape parameters of the generalized extreme value distribution model, and set the third partial derivative to 0 to obtain the shape parameter equation;

[0023] Based on the position parameter equation, the scale parameter equation, and the shape parameter equation, a set of parametric equations is constructed, and the Newton-Raphson method is used to solve the set of parametric equations to obtain the target values ​​of the position parameter, the scale parameter, and the shape parameter, thereby optimizing the generalized extreme value distribution model.

[0024] Preferably, after obtaining the background model of each infrared sub-image, the process further includes:

[0025] Calculate the KL distance between the probability density function of the background model of each infrared sub-image and the probability distribution function corresponding to the background histogram of the infrared sub-image, and use the KL distance as the fitting accuracy of the background model of the infrared sub-image;

[0026] Determine the fitting accuracy of the background model for each infrared sub-image and the value of the preset threshold. If the... If the fitting accuracy of the background model of the infrared sub-image is greater than the preset threshold, then in the target optical image, obtain the model that matches the first infrared sub-image. The set of pixels with the same spatial position in each infrared sub-image is obtained;

[0027] Clustering the set of pixels yields multiple target pixel clusters, thereby enabling the targeting of the first pixel based on these multiple target pixel clusters. The infrared sub-images are segmented to obtain multiple new infrared sub-images.

[0028] Preferably, the formula for calculating the KL distance between the probability density function of the background model of the infrared sub-image and the probability distribution function corresponding to the background histogram of the infrared sub-image is as follows:

[0029] ,

[0030] in, Indicates the first The KL distance between the probability density function of the background model of an infrared sub-image and the probability distribution function corresponding to the background histogram of that infrared sub-image; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The probability distribution function corresponding to the background histogram of each infrared sub-image.

[0031] Preferably, target recognition is performed on each pixel in the infrared sub-image using the background model of each infrared sub-image, and the small target recognition result of the infrared image is obtained based on the target recognition results of all infrared sub-images, including:

[0032] The detection threshold of each infrared sub-image is calculated using a binary search method based on the probability density function of the background model of each infrared sub-image and the preset false alarm probability.

[0033] The feature value of each pixel in the infrared sub-image is compared with the detection threshold, and the pixel with the feature value greater than the detection threshold is identified as a small target pixel in the infrared sub-image.

[0034] The small target pixels of the infrared image are obtained based on the small target pixels in all infrared sub-images.

[0035] Preferably, the formula for calculating the detection threshold of the infrared sub-image is:

[0036] ,

[0037] in, Indicates the preset false alarm probability; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The detection threshold for each infrared sub-image.

[0038] Preferably, acquiring an infrared image of the area to be detected includes:

[0039] Multiple images of the area to be detected are acquired using a drone, and geometric correction is performed on each image based on the pose information of the drone.

[0040] The corrected images are stitched together to obtain an infrared image of the area to be detected.

[0041] The present invention also provides an infrared image small target recognition device, comprising:

[0042] The image acquisition and registration module is used to acquire the infrared image and the optical image of the area to be detected, and to perform georegistration on the infrared image and the optical image to obtain the target infrared image and the target optical image.

[0043] The segmentation boundary coordinate acquisition module is used to cluster the pixels in the target optical image to obtain multiple pixel clusters; extract the boundary pixels of each pixel cluster using an edge detection algorithm, and perform a raster-to-vector operation on the boundary pixels of each pixel cluster to obtain the segmentation boundary coordinates corresponding to each pixel cluster.

[0044] The infrared image segmentation module is used to segment the target infrared image based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images;

[0045] The background model construction module is used to fit the feature values ​​of all pixels in each infrared sub-image using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image, and to construct the parameter likelihood function of the generalized extreme value distribution model based on the maximum likelihood estimation method; to optimize the generalized extreme value distribution model based on the parameter likelihood function, and to obtain the background model of the infrared sub-image based on the optimized generalized extreme value distribution model.

[0046] The small target recognition module is used to perform target recognition on each pixel in the infrared sub-image using the background model of each infrared sub-image, and to obtain the small target recognition result of the infrared image based on the target recognition results of all infrared sub-images.

[0047] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the infrared image small target recognition method described above.

[0048] The infrared image small target recognition method provided in this application simultaneously acquires optical and infrared images of the area to be detected. By matching the two images in geographic space, a target infrared image and a target optical image with the same geographic coordinate system and spatial reference are obtained. Since the pixel feature values ​​of the same type of object are relatively similar, segmenting the target infrared image based on the same type of pixels can obtain infrared sub-images of different background types, thus achieving background type classification. However, since infrared images mainly reflect the thermal radiation information of objects, they cannot clearly describe the shape and texture details of buildings or other objects in the background. Optical images can not only show the shape of each object in the background but also contain the color information of the objects. Therefore, this application clusters the pixels in the target optical image to obtain multiple pixel clusters. By extracting the boundary pixels of each pixel cluster and performing a raster-to-vector conversion operation on the boundary pixels of each pixel cluster, the segmentation boundary coordinates corresponding to each pixel cluster are obtained, thereby segmenting the target infrared image and obtaining... Multiple infrared sub-images with different background types are used. Since the distribution characteristics of background pixels differ in infrared sub-images with different background types, the parameters of their corresponding background models also differ. Therefore, this application performs targeted background modeling for each infrared sub-image and uses the background model of each infrared sub-image to perform small target recognition. This avoids the problem of low small target recognition accuracy caused by the background model of an infrared image constructed under complex backgrounds failing to fully fit the pixel distribution characteristics of various backgrounds in the image. Furthermore, since the increase in infrared image resolution causes the distribution characteristics of background pixels to deviate from a Gaussian distribution, this application uses the generalized extreme value distribution method to construct the background model of each infrared sub-image. This not only better fits the asymmetric distribution and extreme value distribution of background pixels in high-resolution infrared sub-images but also more accurately captures the shape and features of the background pixel distribution, enabling the constructed background model to accurately reflect the background pixel distribution characteristics of the infrared sub-image, thereby improving the accuracy of small target recognition. Attached Figure Description

[0049] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein...

[0050] Figure 1 Flowchart of the infrared image small target recognition method provided in this application;

[0051] Figure 2 This is a schematic diagram of the infrared image small target recognition device provided in this application. Detailed Implementation

[0052] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0053] Please see Figure 1 , Figure 1 The flowchart shown is a method for recognizing small targets in infrared images provided in this application. The method includes:

[0054] S10: Acquire the infrared image and optical image of the area to be detected, and perform georegistration on the infrared image and the optical image to obtain the target infrared image and the target optical image.

[0055] In some embodiments, acquiring an infrared image of the area to be detected includes:

[0056] Multiple images of the area to be detected are acquired using a drone, and geometric correction is performed on each image based on the drone's pose information.

[0057] The corrected images are stitched together to obtain an infrared image of the area to be detected.

[0058] Furthermore, the geographical location of each object in the target optical image is the same as the geographical location of that object in the target infrared image, that is, the geographical location of each pixel in the target optical image corresponds one-to-one with the geographical location of each pixel in the target infrared image.

[0059] S20: Cluster the pixels in the target optical image to obtain multiple pixel clusters; use an edge detection algorithm to extract the boundary pixels of each pixel cluster, and perform a raster-to-vector operation on the boundary pixels of each pixel cluster to obtain the segmentation boundary coordinates corresponding to each pixel cluster.

[0060] Specifically, in one embodiment, when clustering pixels in a target optical image, the number of building categories in the target optical image can be roughly determined based on the function or structure of the buildings, and the number of categories can be used as the number of cluster centers. A pixel is randomly selected in the area covered by each building, and the feature value of the pixel is used as the cluster center of that category.

[0061] Furthermore, the feature values ​​of a pixel can be its RGB value, grayscale value, or HSV value.

[0062] S30: Segment the target infrared image based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images.

[0063] S40: Fit the feature values ​​of all pixels in each infrared sub-image using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image, and construct the parameter likelihood function of the generalized extreme value distribution model based on the maximum likelihood estimation method; optimize the generalized extreme value distribution model based on the parameter likelihood function, and obtain the background model of the infrared sub-image based on the optimized generalized extreme value distribution model.

[0064] S50: Use the background model of each infrared sub-image to perform target recognition on each pixel in the infrared sub-image, and obtain the small target recognition result of the infrared image based on the target recognition results of all infrared sub-images.

[0065] The infrared image small target recognition method provided in this application simultaneously acquires optical and infrared images of the area to be detected. By matching the two images in geographic space, a target infrared image and a target optical image with the same geographic coordinate system and spatial reference are obtained. Since the pixel feature values ​​of the same type of object are relatively similar, segmenting the target infrared image based on the same type of pixels can obtain infrared sub-images of different background types, thus achieving background type classification. However, since infrared images mainly reflect the thermal radiation information of objects, they cannot clearly describe the shape and texture details of buildings or other objects in the background. Optical images, on the other hand, can not only show the shape of each object in the background but also contain the color information of the objects. Therefore, this application performs pixel segmentation on the target optical image. Clustering is used to obtain multiple pixel clusters. By extracting the boundary pixels of each pixel cluster and performing a raster-to-vector conversion operation on the boundary pixels of each pixel cluster, the segmentation boundary coordinates corresponding to each pixel cluster are obtained, thereby segmenting the target infrared image and obtaining multiple infrared sub-images with different background types. Since the distribution characteristics of background pixels in infrared sub-images with different background types are different, the parameters of their corresponding background models are also different. Therefore, this application performs targeted background modeling for each infrared sub-image and uses the background model of each infrared sub-image to perform small target recognition for that infrared sub-image. This avoids the problem that the background model of the infrared image constructed under complex backgrounds cannot fully fit the pixel distribution characteristics of various backgrounds in the image, resulting in low accuracy of small target recognition.

[0066] Specifically, after the resolution of infrared images is improved, the detailed features of ground objects in the images are more prominent, and the background of the images is more complex. According to the central limit theorem, the improvement of resolution will cause the statistical distribution of background pixels in the image to deviate from the Gaussian distribution. Based on this, this application uses a non-Gaussian generalized extreme value distribution for background modeling. Since the generalized extreme value distribution has a higher degree of freedom, it can fit the background pixel distribution in the infrared image very well. The background model constructed can also more accurately reflect the distribution characteristics of background pixels in the infrared sub-image.

[0067] Specifically, in some embodiments of this application, the generalized extreme value distribution model of the infrared sub-image is represented as:

[0068] ,

[0069] in, Indicates the first A generalized extreme value distribution model for each infrared sub-image; The feature value of a pixel; This represents the location parameters of the generalized extreme value distribution model; The scale parameter represents the generalized extreme value distribution model; Represents the shape parameters of the generalized extreme value distribution model;

[0070] The parameter likelihood function of the generalized extreme value distribution model is expressed as:

[0071] ,

[0072] in, Represents the generalized extreme value distribution model The parameter likelihood function; ; Indicates the first The number of pixels in each infrared sub-image; .

[0073] Furthermore, optimization of the generalized extreme value distribution model based on the parametric likelihood function includes:

[0074] Calculate the first partial derivative of the parametric likelihood function with respect to the location parameters of the generalized extreme value distribution model, and set the first partial derivative to 0 to obtain the location parameter equation;

[0075] Calculate the second partial derivative of the parametric likelihood function with respect to the scaling parameter of the generalized extreme value distribution model, and set the second partial derivative to 0 to obtain the scaling parameter equation;

[0076] Calculate the third partial derivative of the parametric likelihood function with respect to the shape parameter of the generalized extreme value distribution model, and set the third partial derivative to 0 to obtain the shape parameter equation;

[0077] A set of parametric equations is constructed based on the position parameter equation, scale parameter equation, and shape parameter equation. The Newton-Raphson method is used to solve the set of parametric equations to obtain the target values ​​of the position parameter, scale parameter, and shape parameter, thereby optimizing the generalized extreme value distribution model.

[0078] To determine the accuracy of the background model for each infrared sub-image, this application uses the KL distance to measure the fitting accuracy between the background model and the background histogram of the infrared sub-image. Specifically, after obtaining the background model for each infrared sub-image, the following steps are also included:

[0079] Calculate the KL distance between the probability density function of the background model of each infrared sub-image and the probability distribution function corresponding to the background histogram of that infrared sub-image, and use the KL distance as the fitting accuracy of the background model of that infrared sub-image;

[0080] Determine the fitting accuracy of the background model for each infrared sub-image and the value of the preset threshold. If the... If the fitting accuracy of the background model of the infrared sub-image is greater than a preset threshold, then in the target optical image, obtain the model that matches the first infrared sub-image. i The set of pixels with the same spatial position in each infrared sub-image is obtained;

[0081] Clustering the set of pixels yields multiple target pixel clusters, which in turn enable the analysis of the first pixel based on these clusters. The infrared sub-images are segmented to obtain multiple new infrared sub-images.

[0082] Specifically, the formula for calculating the KL distance between the probability density function of the background model of the infrared sub-image and the probability distribution function corresponding to the background histogram of the infrared sub-image is as follows:

[0083] ,

[0084] in, Indicates the first The KL distance between the probability density function of the background model of an infrared sub-image and the probability distribution function corresponding to the background histogram of that infrared sub-image; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The probability distribution function corresponding to the background histogram of each infrared sub-image.

[0085] In a specific example of this application, the preset threshold for judging the fitting accuracy is 0.2. If the fitting accuracy is greater than 0.2, it indicates that the constructed background model cannot accurately fit the background of the infrared sub-image. The infrared sub-image is further segmented, and the background model is reconstructed on the newly segmented infrared sub-images until the fitting accuracy of the background model of all infrared sub-images reaches the preset value.

[0086] By verifying the fit between the background model and the background histogram of the infrared sub-image, the accuracy of the background model was improved, which in turn improved the accuracy of the recognition results when small target recognition was performed based on the background model.

[0087] Optionally, in some embodiments of this application, when performing small target identification on each infrared sub-image in step S50, a probability threshold can be set based on experience. Then, the pixels in the infrared sub-image are input into the corresponding background model, and the probability value of the pixel belonging to the background is output. The probability value is compared with the probability threshold to determine whether the pixel belongs to the background or a small target.

[0088] In other embodiments of this application, the detection threshold for each infrared sub-image can also be solved using a bisection method, the specific recognition process of which includes:

[0089] The detection threshold of each infrared sub-image is calculated using a binary search method based on the probability density function of the background model of each infrared sub-image and the preset false alarm probability.

[0090] The feature value of each pixel in the infrared sub-image is compared with the detection threshold, and the pixel with the feature value greater than the detection threshold is identified as a small target pixel in the infrared sub-image.

[0091] The small target pixels in the infrared image are obtained based on the small target pixels in all infrared sub-images.

[0092] Specifically, the formula for calculating the detection threshold of the infrared sub-image is:

[0093] ,

[0094] in, Indicates the preset false alarm probability; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The detection threshold for each infrared sub-image.

[0095] This application uses a method based on mathematical principles and statistical laws to calculate the detection threshold of infrared sub-images. The preset false alarm threshold takes into account the probability of errors that can occur during the recognition process and can be set according to the requirements of the actual task. At the same time, it combines the bisection method to gradually approach the optimal detection threshold by continuously dividing the interval, avoiding the problem of the detection threshold being too high or too low due to human factors when setting the detection threshold based on experience, which leads to low accuracy in small target recognition.

[0096] In one specific implementation, when identifying small targets in each infrared sub-image, each pixel can be marked as either a target or background using 1 or 0. Finally, the region where the small target is located in the infrared image is obtained based on the markings of each pixel in all infrared sub-images.

[0097] Based on the infrared image small target recognition method provided in the above embodiments, this application also provides an infrared image small target recognition device, such as... Figure 2 As shown, the device includes:

[0098] The image acquisition and registration module 10 is used to acquire the infrared image and the optical image of the area to be detected, and to perform georegistration on the infrared image and the optical image to obtain the target infrared image and the target optical image.

[0099] The segmentation boundary coordinate acquisition module 20 is used to cluster the pixels in the target optical image to obtain multiple pixel clusters; the boundary pixels of each pixel cluster are extracted using an edge detection algorithm, and the boundary pixels of each pixel cluster are converted from raster to vector to obtain the segmentation boundary coordinates corresponding to each pixel cluster.

[0100] The infrared image segmentation module 30 is used to segment the target infrared image based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images;

[0101] Background model construction module 40 is used to fit the feature values ​​of all pixels in each infrared sub-image using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image, and to construct the parameter likelihood function of the generalized extreme value distribution model based on the maximum likelihood estimation method; to optimize the generalized extreme value distribution model based on the parameter likelihood function, and to obtain the background model of the infrared sub-image based on the optimized generalized extreme value distribution model.

[0102] The small target recognition module 50 is used to perform target recognition on each pixel in the infrared sub-image using the background model of each infrared sub-image, and to obtain the small target recognition result of the infrared image based on the target recognition results of all infrared sub-images.

[0103] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the infrared image small target recognition method described above.

[0104] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for recognizing small targets in infrared images, characterized in that, include: Acquire an infrared image and an optical image of the area to be detected, and geo-register the infrared image and the optical image to obtain a target infrared image and a target optical image; The pixels in the target optical image are clustered to obtain multiple pixel clusters; the boundary pixels of each pixel cluster are extracted using an edge detection algorithm, and the boundary pixels of each pixel cluster are converted from raster to vector to obtain the segmentation boundary coordinates corresponding to each pixel cluster. The target infrared image is segmented based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images; The feature values ​​of all pixels in each infrared sub-image are fitted using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image. The parameter likelihood function of the generalized extreme value distribution model is constructed based on the maximum likelihood estimation method. The generalized extreme value distribution model is optimized based on the parameter likelihood function. The background model of the infrared sub-image is obtained based on the optimized generalized extreme value distribution model. The background model of each infrared sub-image is used to perform target recognition on each pixel in the infrared sub-image, and the small target recognition result of the infrared image is obtained based on the target recognition results of all infrared sub-images.

2. The infrared image small target recognition method according to claim 1, characterized in that, The generalized extreme value distribution model of the infrared sub-image is expressed as: ， in, Indicates the first A generalized extreme value distribution model for each infrared sub-image; The feature value of a pixel; This represents the location parameters of the generalized extreme value distribution model; The scale parameter represents the generalized extreme value distribution model; Represents the shape parameters of the generalized extreme value distribution model; The parameter likelihood function of the generalized extreme value distribution model is expressed as: ， in, Represents the generalized extreme value distribution model The parameter likelihood function; ; Indicates the first The number of pixels in each infrared sub-image; .

3. The infrared image small target recognition method according to claim 2, characterized in that, Optimizing the generalized extreme value distribution model based on the parametric likelihood function includes: Calculate the first partial derivative of the parametric likelihood function with respect to the location parameters of the generalized extreme value distribution model, and set the first partial derivative to 0 to obtain the location parameter equation; Calculate the second partial derivative of the parametric likelihood function with respect to the scaling parameters of the generalized extreme value distribution model, and set the second partial derivative to 0 to obtain the scaling parameter equation; Calculate the third partial derivative of the parametric likelihood function with respect to the shape parameters of the generalized extreme value distribution model, and set the third partial derivative to 0 to obtain the shape parameter equation; Based on the position parameter equation, the scale parameter equation, and the shape parameter equation, a set of parametric equations is constructed, and the Newton-Raphson method is used to solve the set of parametric equations to obtain the target values ​​of the position parameter, the scale parameter, and the shape parameter, thereby optimizing the generalized extreme value distribution model.

4. The infrared image small target recognition method according to claim 1, characterized in that, After obtaining the background models of each infrared sub-image, the following is also included: Calculate the KL distance between the probability density function of the background model of each infrared sub-image and the probability distribution function corresponding to the background histogram of the infrared sub-image, and use the KL distance as the fitting accuracy of the background model of the infrared sub-image; Determine the fitting accuracy of the background model for each infrared sub-image and the value of the preset threshold. If the... If the fitting accuracy of the background model of the infrared sub-image is greater than the preset threshold, then in the target optical image, obtain the model that matches the first infrared sub-image. i The set of pixels with the same spatial position in each infrared sub-image is obtained; Clustering the set of pixels yields multiple target pixel clusters, thereby enabling the targeting of the first pixel based on these multiple target pixel clusters. i The infrared sub-images are segmented to obtain multiple new infrared sub-images.

5. The infrared image small target recognition method according to claim 4, characterized in that, The formula for calculating the KL distance between the probability density function of the background model of the infrared sub-image and the probability distribution function corresponding to the background histogram of the infrared sub-image is as follows: ， in, Indicates the first The KL distance between the probability density function of the background model of an infrared sub-image and the probability distribution function corresponding to the background histogram of that infrared sub-image; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The probability distribution function corresponding to the background histogram of each infrared sub-image.

6. The infrared image small target recognition method according to claim 1, characterized in that, Using the background model of each infrared sub-image, target recognition is performed on each pixel in the infrared sub-image, and the small target recognition result of the infrared image is obtained based on the target recognition results of all infrared sub-images, including: The detection threshold of each infrared sub-image is calculated using a binary search method based on the probability density function of the background model of each infrared sub-image and the preset false alarm probability. The feature value of each pixel in the infrared sub-image is compared with the detection threshold, and the pixel with the feature value greater than the detection threshold is identified as a small target pixel in the infrared sub-image. The small target pixels of the infrared image are obtained based on the small target pixels in all infrared sub-images.

7. The infrared image small target recognition method according to claim 6, characterized in that, The formula for calculating the detection threshold of infrared sub-images is: ， in, Indicates the preset false alarm probability; Indicates the first The probability density function of the background model of each infrared sub-image; Indicates the first The detection threshold for each infrared sub-image.

8. The infrared image small target recognition method according to claim 1, characterized in that, Acquiring an infrared image of the area to be detected includes: Multiple images of the area to be detected are acquired using a drone, and geometric correction is performed on each image based on the pose information of the drone. The corrected images are stitched together to obtain an infrared image of the area to be detected.

9. An infrared image small target recognition device, characterized in that, include: The image acquisition and registration module is used to acquire the infrared image and the optical image of the area to be detected, and to perform georegistration on the infrared image and the optical image to obtain the target infrared image and the target optical image. The segmentation boundary coordinate acquisition module is used to cluster the pixels in the target optical image to obtain multiple pixel clusters; extract the boundary pixels of each pixel cluster using an edge detection algorithm, and perform a raster-to-vector operation on the boundary pixels of each pixel cluster to obtain the segmentation boundary coordinates corresponding to each pixel cluster. The infrared image segmentation module is used to segment the target infrared image based on the segmentation boundary coordinates corresponding to all pixel clusters to obtain multiple infrared sub-images; The background model construction module is used to fit the feature values ​​of all pixels in each infrared sub-image using the generalized extreme value distribution method to obtain the generalized extreme value distribution model of the infrared sub-image, and to construct the parameter likelihood function of the generalized extreme value distribution model based on the maximum likelihood estimation method; to optimize the generalized extreme value distribution model based on the parameter likelihood function, and to obtain the background model of the infrared sub-image based on the optimized generalized extreme value distribution model. The small target recognition module is used to perform target recognition on each pixel in the infrared sub-image using the background model of each infrared sub-image, and to obtain the small target recognition result of the infrared image based on the target recognition results of all infrared sub-images.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the infrared image small target recognition method according to any one of claims 1-8.

Images