A processing method and system for infrared image recognition

Abstract

This invention provides a processing method for infrared image recognition, comprising: acquiring an infrared image; preprocessing the infrared image to generate a score image; performing multi-scale segmentation processing on the score image to generate a high signal-to-noise ratio (SNR) image; performing local thresholding processing on the high SNR image to generate a target region; and performing filtering processing on the target region to generate a target contour. The infrared image recognition processing method and system disclosed in this invention can improve the efficiency and accuracy of image recognition.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a processing method and system for infrared image recognition. Background Technology

[0002] Current methods for infrared image processing heavily rely on the recognition threshold of each sample. The instruments are usually installed in large public places and need to be run for a long time. Complex background factors in different usage environments can affect the accuracy of image recognition. In unevenly distributed scenarios, the accuracy of image segmentation is insufficient, resulting in a high false alarm rate. Furthermore, they can only recognize existing images and cannot provide early warning functions for unknown images. Summary of the Invention

[0003] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide an infrared image recognition processing method and system that can improve the efficiency and accuracy of image recognition.

[0004] To achieve the above and other related objectives, the present invention provides a processing method for infrared image recognition, comprising:

[0005] Acquire infrared images;

[0006] The infrared image is preprocessed to generate a score image;

[0007] The scored image is subjected to multi-scale segmentation to generate a high signal-to-noise ratio image;

[0008] The high signal-to-noise ratio image is locally thresholded to generate the target region;

[0009] The target region is filtered to generate a target contour.

[0010] In one embodiment of the present invention, the infrared image includes:

[0011] Low-frequency images correspond to smooth regions;

[0012] High-frequency images correspond to detailed areas.

[0013] In one embodiment of the present invention, the smooth region includes the target area, the background area and other interference areas, and the detail area includes the edge area, the noise area and the bad pixel area.

[0014] In one embodiment of the present invention, the step of preprocessing the infrared image to generate a score image includes:

[0015] The infrared image is registered according to the image registration coefficient of the instrument;

[0016] The infrared image is calibrated for light intensity according to the instrument's light intensity calibration coefficient.

[0017] The infrared image is processed according to the registration coefficient and the light intensity calibration coefficient to generate a score image.

[0018] In one embodiment of the present invention, the step of performing multi-scale segmentation processing on the scored image to generate a high signal-to-noise ratio image includes:

[0019] The scored image is subjected to watershed segmentation to generate a water basin image;

[0020] The image of the water collection basin is subjected to background stretching to generate a difference image;

[0021] The difference image is subjected to secondary segmentation, and the signal-to-noise ratio of the score image and the secondary segmented image on the score image is calculated to generate a high signal-to-noise ratio image.

[0022] In one embodiment of the present invention, before the step of stretching the background of the water basin image to generate a difference image, the method further includes selecting a multi-scale sliding window to perform a top-hat transformation or a bottom-hat transformation on the water basin image.

[0023] In one embodiment of the present invention, the target area includes a real area and a fake area.

[0024] In one embodiment of the present invention, the step of filtering the target region to generate the target contour includes:

[0025] Morphological filtering is applied to the target region to generate a real region;

[0026] The real region is subjected to temporal filtering to generate the target contour.

[0027] In one embodiment of the present invention, the step of performing morphological filtering on the target region to generate a real region includes:

[0028] The target area is subjected to erosion and expansion processes in sequence to generate a false area;

[0029] The fake region is then subjected to expansion and erosion processes in sequence to generate the real region.

[0030] The present invention also provides an infrared image recognition processing system, comprising:

[0031] The image acquisition module is used to acquire infrared images;

[0032] The preprocessing module is used to preprocess the infrared image to generate a score image;

[0033] The segmentation module is used to perform multi-scale segmentation processing on the scored image to generate a high signal-to-noise ratio image.

[0034] The thresholding module is used to perform local thresholding processing on the high signal-to-noise ratio image to generate the target region;

[0035] The target generation module is used to filter the target region and generate the target contour.

[0036] As described above, the present invention provides an infrared image recognition processing method and system. By utilizing the consistency of the background region for multi-scale image segmentation, the problem of over-segmentation and under-segmentation caused by uneven illumination is solved. By converting the scored image into a high signal-to-noise ratio image, the basis for identifying samples is changed from a threshold to an adaptive signal-to-noise ratio value of the target region. This can provide early warning information for unknown samples outside the existing database, thereby improving the efficiency and accuracy of image recognition. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 The flowchart shown is a processing method for infrared image recognition according to the present invention;

[0039] Figure 2 yes Figure 1 The flowchart of step S120 in the illustrated embodiment is shown in an exemplary embodiment;

[0040] Figure 3 yes Figure 1 The flowchart of step S130 in the illustrated embodiment is shown in an exemplary embodiment;

[0041] Figure 4 yes Figure 1 The flowchart of step S150 in the illustrated embodiment is shown in an exemplary embodiment;

[0042] Figure 5 yes Figure 4 The flowchart of step S510 in the illustrated embodiment is shown in an exemplary embodiment;

[0043] Figure 6 This is a schematic diagram of the first image acquired by camera A;

[0044] Figure 7 This is a schematic diagram of the second image acquired by camera B;

[0045] Figure 8 This is a schematic diagram of bad pixel detection in the first image;

[0046] Figure 9 This is a schematic diagram of the bad pixel detection image in the second image;

[0047] Figure 10 This is a schematic diagram of highlight detection in the first image;

[0048] Figure 11 This is a schematic diagram of highlight detection in the second image;

[0049] Figure 12 This is a schematic diagram of the preprocessed score image;

[0050] Figure 13 This is a schematic diagram of the water collection basin after watershed segmentation.

[0051] Figure 14 This is a schematic diagram of the water collection basin image after removing non-uniform illumination.

[0052] Figure 15 This is a schematic diagram of the difference image after background stretching.

[0053] Figure 16 This is a schematic diagram of a high signal-to-noise ratio image after secondary segmentation processing;

[0054] Figure 17 This is a schematic diagram of the target region for local thresholding processing of a high signal-to-noise ratio image;

[0055] Figure 18 This is a schematic diagram of the target contour generated after filtering the target area.

[0056] Figure 19 This is a schematic diagram of the actual region generated after the target outline has undergone dilation and erosion processing.

[0057] Figure 20 The diagram shown is a schematic of an infrared image recognition processing system according to the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] First, it should be noted that current infrared image processing methods heavily rely on the recognition threshold of each sample. Instruments are typically installed in large public places and require long-term operation. Even slight temperature changes in the instrument can cause spectral drift in the infrared band, leading to threshold drift in the database and resulting in false alarms. Existing infrared image processing methods require high instrument precision, making them difficult to use in non-laboratory environments. Furthermore, complex background factors in different usage environments can affect the accuracy of image recognition. In unevenly distributed scenarios, insufficient image segmentation precision leads to a high false alarm rate, and these methods can only recognize existing images, failing to provide warnings for unknown images. An infrared image recognition processing system can be used in non-laboratory environments, improving image recognition accuracy and reducing false alarms caused by insufficient image segmentation precision. In other application scenarios, the infrared image recognition processing methods and systems for specific objects can be configured according to actual conditions; the embodiments of this invention do not limit this.

[0060] Please see Figure 1 As shown, Figure 1 This is a schematic flowchart illustrating an infrared image recognition processing method according to an exemplary embodiment of the present invention. It should be understood that this method can also be applied to other exemplary implementation environments and specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0061] like Figure 1 In an exemplary embodiment, the infrared image recognition processing method includes at least steps S110 to S150, which are described in detail below:

[0062] Step S110: Acquire infrared image;

[0063] Step S120: Preprocess the infrared image to generate a score image;

[0064] Step S130: Perform multi-scale segmentation on the scored image to generate a high signal-to-noise ratio image;

[0065] Step S140: Perform local thresholding on the high signal-to-noise ratio image to generate the target region;

[0066] Step S150: Perform time filtering on the target area to generate the target contour.

[0067] like Figure 1In an exemplary embodiment, when step S110 is performed, an infrared image is acquired. It should first be noted that the infrared image is generated within a spectral imaging system. The infrared image can include low-frequency and high-frequency images, where the low-frequency image corresponds to smooth regions and the high-frequency image corresponds to detailed regions. Smooth regions can include target areas, background areas, and other interfering areas, while detailed regions can include edge areas, noise areas, and bad pixel areas.

[0068] like Figure 2 In an exemplary embodiment, when step S120 is performed, the infrared image is preprocessed to generate a scored image. Specifically, step S120 may include steps S210 to S230, which are described in detail below:

[0069] Step S210: Perform image registration on the infrared image according to the image registration coefficient of the instrument;

[0070] Step S220: Perform light intensity calibration on the infrared image according to the instrument's light intensity calibration coefficient;

[0071] Step S230: Based on the image registration coefficient and light intensity calibration coefficient, perform shadow detection, highlight detection, blind pixel detection and mean filtering calculation on the infrared image in sequence to generate a score image.

[0072] In an exemplary embodiment, image registration involves first extracting feature points from two images, finding matching feature point pairs through similarity measurement, obtaining image space coordinate transformation parameters from the matching feature point pairs, and finally performing image registration using these parameters. Feature extraction is crucial in registration technology, and accurate feature extraction ensures successful feature matching. Image registration methods can include relative registration and absolute registration. Relative registration involves selecting one image from multiple images as a reference image and registering the other related images with it; the coordinate system is arbitrary. Absolute registration involves defining a control grid, and registering all images relative to this grid, essentially performing geometric correction on each component image to achieve coordinate system unification. Mean filtering is a linear filtering algorithm that applies a template to the target pixel in the image, including its neighboring pixels, and then replacing the original pixel value with the average value of all pixels in the template. Neighboring pixels refer to the pixels surrounding the target pixel, forming a filtering template that removes the target pixel itself.

[0073] like Figure 3 In an exemplary embodiment, when step S130 is executed, the scored image undergoes multi-scale segmentation processing to generate a high signal-to-noise ratio image. Specifically, step S130 may include steps S310 to S330, which are described in detail below:

[0074] Step S310: Perform watershed segmentation on the scored image to generate a water basin image;

[0075] Step S320: Perform background stretching processing on the water collection basin image to generate a difference image;

[0076] Step S330: Perform secondary segmentation on the difference image, and calculate the signal-to-noise ratio of the score image and the image after secondary segmentation on the score image to generate a high signal-to-noise ratio image.

[0077] In an exemplary embodiment, watershed segmentation refers to segmenting the scored image using a watershed algorithm. Each segmented image is a catchment basin image, and the boundary points of the catchment basin image are the watersheds, representing the maxima of the scored image. In real-world scenarios, the scores in the scored image may exhibit unevenness due to the non-uniformity of the active light source, causing the values ​​of the same background region to appear concave and convex, affecting image segmentation and resulting in oversegmentation and undersegmentation of the same background region. Oversegmentation refers to dividing the same region into two or more regions, while undersegmentation refers to dividing multiple different background regions into the same region. Oversegmentation and undersegmentation can cause false alarms in image recognition, reducing the accuracy of image recognition. Therefore, the purpose of watershed segmentation in infrared images is to correctly divide the region into blocks without changing the score values ​​of the corresponding regions. However, it is not limited to this; watershed segmentation can also employ other image segmentation techniques, including multi-threshold segmentation techniques, region growing segmentation algorithms, edge detection-based segmentation methods, image segmentation algorithms combined with specific tools, image segmentation based on genetic algorithms, and segmentation methods based on active contour models, etc.

[0078] In an exemplary embodiment, before step S320, a multi-scale sliding window is selected to perform a top-hat transformation or a bottom-hat transformation on the water basin image. Both top-hat and bottom-hat transformations are used to correct images with uneven illumination. The top-hat transformation corrects bright objects on a dark background, while the bottom-hat transformation corrects dark objects on a bright background. Uneven illumination causes an uneven image background, resulting in over-segmentation. Therefore, background stretching is performed on the entire background region of the water basin image based on the smoothness of the background area, flattening the background region. This allows the background to be segmented as a whole during image segmentation, removing the uneven illumination and obtaining a difference image. Secondary segmentation refers to detecting whether each region of the difference image has under-segmentation. When the variance exceeds a certain threshold, under-segmentation occurs, requiring secondary segmentation of smaller regions. After secondary segmentation of all regions, an under-segmentation check is performed on the segmented image obtained. The under-segmentation check determines whether secondary segmentation is needed based on the similarity of pixel values ​​in connected regions. The signal-to-noise ratio (SNR) of an image is the ratio of the power spectrum of the signal to that of the noise. In an image with high SNR, the SNR of the target region is at its maximum, and the higher the SNR value, the fewer impurities are in the target region.

[0079] like Figure 1 In an exemplary embodiment, when step S140 is executed, a local thresholding process is performed on the high signal-to-noise ratio image to generate a target region. The local thresholding process assumes that the illumination within a certain area of ​​the image is relatively similar. It uses a sliding window to scan the image and compares the brightness of the center point of the sliding window with the brightness of its neighborhood area. If the brightness of the center point is higher than that of the neighborhood area, the center point is marked as white; otherwise, it is marked as black. The target region includes both real and false regions.

[0080] like Figure 4 In an exemplary embodiment, when step S150 is executed, the target region is filtered to generate a target contour. Specifically, step S150 may include steps S510 to S520, which are described in detail below:

[0081] Step S510: Perform morphological filtering on the target region to generate the real region;

[0082] Step S520: Perform time filtering on the real area to generate the target contour.

[0083] like Figure 5 In one exemplary embodiment, step S510 may include:

[0084] Step S511: Perform erosion and expansion processes on the target area sequentially to generate a false area;

[0085] Step S512: Perform expansion and erosion processing on the fake regions in sequence to generate real regions.

[0086] Morphological filtering is used to eliminate false regions, while dilation is a process of finding a local maximum. It involves convolving the image to be processed with a structured element, selecting an aiming point, and calculating the maximum area that the structured element can cover centered on that aiming point. Erosion is the opposite; it also selects a structured element and an aiming point, but the operation becomes finding the maximum area that the structured element can eliminate centered on that aiming point.

[0087] like Figure 6 and Figure 7 In an exemplary embodiment, camera A is used to acquire a first image, and camera B is used to acquire a second image. In both images, the black box in the lower right corner represents the target object, and the other black boxes represent interference. The two images are then converted into two infrared images using a low-pass filter. The low-pass filter formula satisfies:

[0088]

[0089] Where u and v are the pixels of the sliding window, M is the width occupied by the pixels of the first image, N is the width occupied by the pixels of the second image, and D is the adaptive cutoff frequency. Infrared images can include low-frequency and high-frequency images. Low-frequency images mainly correspond to smooth regions. For example, smooth regions can include target areas, background areas, and other interference areas, but are not limited to these; smooth regions can also include other areas. High-frequency images mainly correspond to the detailed areas of the image. For example, detailed regions can include edge areas, noise areas, and bad pixel areas. In the bad pixel display image, a height value of 1 represents a bad pixel, and a height value of 0 represents a non-bad pixel.

[0090] like Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 , Figure 13 , Figure 14 , Figure 15 , Figure 16 , Figure 17 , Figure 18 and Figure 19As shown, in an exemplary embodiment, image registration processing is performed on two infrared images using an affine transformation matrix based on the instrument's registration coefficients. Then, light intensity calibration processing is performed on the infrared images based on the instrument's light intensity calibration coefficients. Finally, shadow detection, highlight detection, blind pixel detection, and mean filtering calculations are performed sequentially on the infrared images based on the image registration coefficients and light intensity calibration coefficients to generate a score image. Subsequently, watershed segmentation processing is performed on the score image to generate a catchment image, where the black-framed area represents the segmented region of the target object. A top-hat transform is then performed on the catchment image, i.e., non-uniform illumination is removed, followed by background stretching to generate a difference image. A secondary segmentation process is then performed on the difference image, and the signal-to-noise ratio (SNR) of the score image and the difference image after secondary segmentation is calculated to generate a high SNR image. Each region in the high SNR image undergoes undersegmentation and oversegmentation checks, followed by local thresholding processing to generate the target region. Local thresholding processing assumes that the illumination received by the images within a certain region is relatively similar. It uses a sliding window to scan the image and compares the brightness of the center point of the sliding window with the brightness of its neighborhood area. If the brightness of the center point is higher than that of the neighborhood area, the center point is marked as white; otherwise, it is marked as black. The target region includes both real and spoof regions. Finally, morphological filtering is applied to the target region, involving erosion and dilation to generate spoof regions. These spoof regions are then subjected to dilation and erosion to generate the real region. Temporal filtering is then applied to the real region to generate the target contour.

[0091] Please see Figure 20 As shown, Figure 20 This is a schematic diagram of an infrared image recognition processing system according to an exemplary embodiment of the present invention. In some embodiments, the infrared image recognition processing system may include an image acquisition module 610, a preprocessing module 620, a segmentation module 630, a thresholding module 640, and a target generation module 650. The image acquisition module 610 is used to acquire infrared images. The infrared images can be acquired by a spectral imaging system, and at least two infrared images are available. The infrared images may include low-frequency images and high-frequency images, where the low-frequency images correspond to smooth regions and the high-frequency images correspond to detailed regions. Smooth regions may include target areas, background areas, and other interference areas, while detailed regions may include edge areas, noise areas, and bad pixel areas.

[0092] Please see Figure 20As shown, in an exemplary embodiment, the preprocessing module 620 is used to preprocess the infrared image to generate a scored image. The preprocessing may include image registration of the infrared image, light intensity calibration of the infrared image, and shadow detection, highlight detection, blind pixel detection, and mean filtering calculation of the infrared image sequentially based on the image registration coefficient and light intensity calibration coefficient. Image registration refers to first extracting feature points from two images, finding matching feature point pairs through similarity measurement, obtaining image space coordinate transformation parameters through the matching feature point pairs, and finally performing image registration using the coordinate transformation parameters. Feature extraction is crucial in registration technology; accurate feature extraction ensures successful feature matching. Image registration methods may include relative registration and absolute registration. Relative registration involves selecting one image from multiple images as a reference image and registering the other related images to it; the coordinate system is arbitrary. Absolute registration involves first defining a control grid, and registering all images relative to this grid, that is, performing geometric correction on each component image to achieve coordinate system unification. Mean filtering is a linear filtering algorithm that applies a template to a target pixel in an image, including its neighboring pixels. The original pixel value is then replaced by the average value of all pixels in the template. Neighboring pixels refer to the pixels surrounding the target pixel, forming a filtering template that essentially removes the target pixel itself.

[0093] Please see Figure 20As shown, in an exemplary embodiment, the segmentation module 630 is used to perform multi-scale segmentation processing on the scored image to generate a high signal-to-noise ratio (SNR) image. The multi-scale segmentation processing may include watershed cutting of the scored image to generate a catchment image. Background stretching of the catchment image generates a difference image. Secondary segmentation is performed on the difference image, and the SNR of corresponding regions in the scored image and the secondary segmented image is calculated. Watershed cutting refers to using a watershed algorithm to segment the scored image; each segmented image is a catchment image, and the boundary points of the catchment image are the watersheds, representing the maxima of the scored image. In real-world scenarios, the unevenness of illumination from active light sources can cause the numerical values ​​of the same background region to exhibit concavity and convexity, affecting image segmentation and resulting in oversegmentation and undersegmentation of the same background region. Oversegmentation refers to dividing the same region into two or more regions, while undersegmentation refers to classifying multiple different background regions into the same region. Oversegmentation and undersegmentation can cause false alarms in image recognition, reducing the accuracy of image recognition. Therefore, the purpose of watershed segmentation in infrared images is to correctly divide the region into blocks without changing the score value of the corresponding region. However, it is not limited to this; watershed segmentation can also employ other image segmentation techniques, including multi-threshold segmentation, region growing segmentation algorithms, edge detection-based segmentation methods, image segmentation algorithms combined with specific tools, image segmentation based on genetic algorithms, and segmentation methods based on active contour models.

[0094] Please see Figure 20 As shown, in an exemplary embodiment, the thresholding module 640 is used to perform local thresholding processing on a high signal-to-noise ratio image to generate a target region. The local thresholding processing assumes that the illumination within a certain area of ​​the image is relatively similar. It scans the image using a sliding window and compares the brightness of the center point of the sliding window with the brightness of its neighborhood area. If the brightness of the center point is higher than the neighborhood brightness, the center point is marked as white; otherwise, it is marked as black. The target region includes both real and false regions.

[0095] Please see Figure 20 As shown, in an exemplary embodiment, the target generation module 650 is used to perform temporal filtering on the target region to generate a target contour. The filtering process may include morphological filtering and temporal filtering. Morphological filtering is used to eliminate false regions. Dilation is a process of finding a local maximum; the image to be processed is convolved with a structured element, an aiming point is selected, and the maximum area that the structured element can cover centered on that aiming point is calculated. Erosion is the opposite; it also selects a structured element and an aiming point, but the operation becomes finding the maximum area that the structured element can eliminate centered on that aiming point.

[0096] It should be noted that the infrared image recognition processing system and the infrared image recognition processing method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the infrared image recognition processing system provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0097] In summary, the infrared image recognition processing method and system provided by this invention solves the problems of over-segmentation and under-segmentation caused by uneven illumination by using the consistency of the background region for multi-scale image segmentation. By converting the scored image into a high signal-to-noise ratio image and changing the basis for identifying samples from a threshold to an adaptive signal-to-noise ratio value of the target region, it can provide early warning information for unknown samples outside the existing database, thereby improving the efficiency and accuracy of image recognition.

[0098] In the description of this specification, the references to terms such as "this embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0099] The embodiments of the present invention disclosed above are merely illustrative of the invention. The embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A processing method for infrared image recognition, characterized in that, include: Acquire infrared images; The infrared image is preprocessed to generate a score image; The scored image is subjected to multi-scale segmentation to generate a high signal-to-noise ratio image; The high signal-to-noise ratio image is subjected to local thresholding to generate a target region; wherein, local thresholding refers to the process of comparing the brightness of a certain region of the high signal-to-noise ratio image with the brightness of its neighboring regions. The target region is filtered to generate a target contour; The step of preprocessing the infrared image to generate a score image includes: The infrared image is registered according to the image registration coefficient of the instrument; The infrared image is calibrated for light intensity according to the instrument's light intensity calibration coefficient. The infrared image is processed according to the registration coefficient and the light intensity calibration coefficient to generate a scored image; The step of performing multi-scale segmentation on the scored image to generate a high signal-to-noise ratio image includes: The scored image is segmented using a watershed algorithm. Each segmented image is a catchment basin image, and the boundary points of the catchment basin image are the watersheds. The watersheds represent the maximum points of the scored image. The image of the water collection basin is subjected to background stretching to generate a difference image; The difference image is subjected to secondary segmentation. On the score image, the signal-to-noise ratio of the secondary segmented image and its surrounding adjacent regions is calculated to generate a high signal-to-noise ratio image.

2. The infrared image recognition processing method according to claim 1, characterized in that, The infrared image includes: Low-frequency images correspond to smooth regions; High-frequency images correspond to detailed areas.

3. The infrared image recognition processing method according to claim 2, characterized in that, The smoothed region includes the target area, the background area, and other interfering areas, while the detailed region includes the edge area, the noise area, and the bad pixel area.

4. The infrared image recognition processing method according to claim 1, characterized in that, Before the step of stretching the background of the water collection basin image to generate a difference image, the method further includes selecting a multi-scale sliding window to perform a top-hat transformation or a bottom-hat transformation on the water collection basin image.

5. The infrared image recognition processing method according to claim 1, characterized in that, The target area includes real areas and fake areas.

6. The infrared image recognition processing method according to claim 5, characterized in that, The step of filtering the target region to generate the target contour includes: Morphological filtering is applied to the target region to generate a real region; The real region is subjected to temporal filtering to generate the target contour.

7. The infrared image recognition processing method according to claim 6, characterized in that, The step of performing morphological filtering on the target region to generate a real region includes: The target area is subjected to erosion and expansion processes in sequence to generate a false area; The dummy region is then subjected to expansion and erosion processes in sequence to generate the real region.

8. A processing system for infrared image recognition, characterized in that, include: The image acquisition module is used to acquire infrared images; The preprocessing module is used to preprocess the infrared image to generate a score image; The segmentation module is used to perform multi-scale segmentation processing on the scored image to generate a high signal-to-noise ratio image. The thresholding module is used to perform local thresholding processing on the high signal-to-noise ratio image to generate a target region; wherein, local thresholding processing refers to the process of comparing the brightness of a certain region of the high signal-to-noise ratio image with the brightness of its adjacent regions. The target generation module is used to filter the target region and generate a target outline. The step of preprocessing the infrared image to generate a score image includes: The infrared image is registered according to the image registration coefficient of the instrument; The infrared image is calibrated for light intensity according to the instrument's light intensity calibration coefficient. The infrared image is processed according to the registration coefficient and the light intensity calibration coefficient to generate a scored image; The step of performing multi-scale segmentation on the scored image to generate a high signal-to-noise ratio image includes: The scored image is segmented using a watershed algorithm. Each segmented image is a catchment basin image, and the boundary points of the catchment basin image are the watersheds. The watersheds represent the maximum points of the scored image. The image of the water collection basin is subjected to background stretching to generate a difference image; The difference image is subjected to secondary segmentation. On the score image, the signal-to-noise ratio of the secondary segmented image and its surrounding adjacent regions is calculated to generate a high signal-to-noise ratio image.

Images