Thermal image drift correction method based on composite compensation and related device

Abstract

The application discloses a thermal imaging image drift correction method based on composite compensation and related equipment, and the method comprises the following steps: receiving a shutter closing instruction, acquiring a shutter position parameter and collecting a transition frame image; collecting a shutter closing reference image after the shutter is closed; constructing a shutter blocking state model based on the shutter position parameter, and extracting a reflection non-uniform component based on the shutter blocking state model and the transition frame image; constructing a reflection radiation distribution model according to the reflection non-uniform component, and separating to obtain a correction bias image by combining the shutter closing reference image and the reflection radiation distribution model; determining an initial bias correction parameter based on the correction bias image; collecting a thermal imaging image, combining the initial bias correction parameter, the reflection radiation distribution model and the current thermal imaging image to determine a shutter-free correction compensation parameter; and correcting the thermal imaging image according to the shutter-free correction compensation parameter to obtain a target thermal imaging image. The application can improve the accuracy of thermal imaging image correction.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a thermal imaging image drift correction method and related equipment based on composite compensation. Background Technology

[0002] With the continuous development of uncooled infrared thermal imaging technology, thermal imaging equipment has been widely used in security monitoring, vehicle-mounted assisted driving, power line inspection, industrial testing, emergency rescue, and drone payloads. Because uncooled infrared focal plane detectors are susceptible to factors such as device temperature drift, thermal changes in the readout circuit, and the accumulation of fixed-mode noise during long-term operation, a composite compensation method combining shutter correction and no-shutter correction is needed to correct detector drift in real time, ensuring the temperature measurement accuracy and imaging quality of thermal images.

[0003] Currently, shutter correction technology typically involves acquiring a shutter background image when the shutter is closed and using this image as a detector bias reference for subsequent offset compensation. Existing technologies generally treat the shutter blades as ideal blackbodies, and assume that the image acquired when the shutter is fully closed accurately reflects the detector's current bias state. Therefore, the acquired shutter background image is directly used to generate corresponding correction parameters to achieve offset compensation in thermal imaging. However, in practical applications, the surface emissivity of the shutter blades is not that of an ideal blackbody, and the radiation received by the detector when the shutter is closed includes not only the thermal radiation from the shutter itself, but also radiation reflected from the shutter surface from the internal components of the lens barrel, the detector cavity, and the readout circuit. As the equipment continues to operate, the internal thermal environment of the lens barrel and detector constantly changes, and the aforementioned radiation components also change dynamically. This causes non-true bias information to be mixed into the background image acquired when the shutter is closed, resulting in cold rings, dark spots, or aperture-shaped non-uniform artifacts in the thermal imaging image, affecting the drift correction accuracy and imaging quality.

[0004] There is currently no good solution to the above problems. Summary of the Invention

[0005] This application provides a thermal imaging image drift correction method and related equipment based on composite compensation, which can improve the accuracy of thermal imaging image correction.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, a thermal imaging image drift correction method based on composite compensation is provided, the method comprising: In response to receiving a shutter closing command, during the transition period of shutter blade movement, shutter position parameters are acquired and multiple transition frame images corresponding to the shutter closing process are captured. After the shutter is fully closed, a reference image of the shutter closure is captured; A shutter occlusion state model is constructed based on shutter position parameters, and non-uniform reflection components are extracted based on the shutter occlusion state model and transition frame images. A reflection radiation distribution model is constructed based on the non-uniform reflection components, and the corrected bias image is obtained by separating the shutter closure reference image and the reflection radiation distribution model. Initial bias correction parameters are determined based on the corrected bias image; The thermal imaging image is acquired while the shutter is open, and the shutter-free compensation parameters are determined by combining the initial offset correction parameters, the reflected radiation distribution model, and the current thermal imaging image. The thermal image is corrected based on the shutter-free correction compensation parameters to obtain the target thermal image.

[0007] In one possible implementation of the first aspect, the method of constructing a shutter occlusion state model based on shutter position parameters and extracting non-uniform reflection components based on the shutter occlusion state model and the transition frame image includes: Determine the spatial distribution of pixels in the detector pixel array of the thermal imaging device; The shutter exposure state is determined by combining the preset shutter control timing signal and shutter position parameters; The light transmission state of each pixel in the pixel array is determined by combining the pixel spatial distribution and shutter exposure state, and the mapping relationship between the pixel position and the shutter exposure time is determined based on the light transmission state. A shutter occlusion state model is constructed based on the mapping relationship between pixel position and shutter exposure time; Establish the temporal correspondence between multiple transition frame images and shutter occlusion state models by combining multiple transition frame images and shutter occlusion state models; The pixel received signal components are obtained by decomposing the light transmission state in multiple transition frame images based on the time correspondence. The non-uniform reflection component is extracted by combining the pixel received signal components and the shutter occlusion state model.

[0008] In one possible implementation of the first aspect, the method for extracting the non-uniform reflection component by combining the pixel received signal component and the shutter occlusion state model includes: The occlusion and light transmission states of each pixel are determined by the shutter occlusion state model; Based on the occlusion state and the light transmission state, the received signal components of the pixel are spatiotemporally filtered to determine the target pixel that is in an occlusion state and has a reflection response. For any target pixel, the reflection response component is extracted from the received signal component of the pixel, and the consistency of the reflection response component is verified by combining the light transmission state change process. For target pixels that pass the consistency verification, determine the temporal relationship of the occlusion state, and use the temporal relationship of the occlusion state to perform time alignment on the reflection response component to obtain the time alignment result; Based on the time alignment results, the reflection response components of the target pixel are spatially interpolated and reconstructed to obtain a continuous spatial reflection field; Extracting the non-uniform reflection component from a continuous spatial reflection field.

[0009] In one possible implementation of the first aspect, the method of decomposing the light transmission states in multiple transition frame images based on time correspondence to obtain pixel received signal components includes: Based on the time correspondence, determine the switching frame in which the light-transmitting state of each pixel changes in the transition frame image; For any pixel, extract the first set of transition frame images with a preset number of frames before the switching frame and the second set of transition frame images with a preset number of frames after the switching frame, and determine the signal value sequence in each image in the first set of transition frame images and the second set of transition frame images; Calculate the difference between the signal value sequences of each image in the first transition frame image set and the second transition frame image set, and use the difference as the scene light component of the pixel; The target occlusion frames are selected from the second transition frame image set by filtering the continuous frames in the second transition frame image set whose pixels are in a state of continuous occlusion and whose signal fluctuation is lower than a preset threshold. Calculate the mean of the signal values ​​corresponding to the pixels in all target occlusion frames, and use the mean of the signal values ​​as the reflection response component of the pixel; The reflection response component and the scene light component are used as the pixel received signal components.

[0010] In one possible implementation of the first aspect, the method of constructing a reflection radiation distribution model based on the non-uniform reflection components and separating the shutter closure reference image and the reflection radiation distribution model to obtain a corrected bias image includes: The spatial response variation trend of each pixel is calculated based on the non-uniform reflection component, and the reflected radiation accumulation area is identified by the spatial response variation trend. The regional response fluctuation characteristics and spatial response continuity characteristics of the reflected radiation accumulation area are determined by the spatial response variation trend. The spatial structure confidence level of the reflected radiation accumulation region is determined by combining the regional response fluctuation characteristics and the spatial response continuity characteristics. The propagation results of reflected radiation are determined by combining the confidence level of spatial structure and the area of ​​concentrated reflected radiation. The intensity and spatial range of reflected radiation propagation are identified by the results of reflected radiation propagation, and a reflected radiation distribution model is constructed by combining the intensity and spatial range of reflected radiation propagation. The reflected radiation contribution value is calculated based on the reflected radiation distribution model, and the reflected radiation contribution value is used to generate a reflected radiation mapping image corresponding to the shutter closure reference image. The initial bias image is obtained by separating the pixel response of the reflected radiation mapping image using the shutter closure reference image as the original response image; The corrected bias image is obtained by performing a residual reflection separation step based on the initial bias image.

[0011] In one possible implementation of the first aspect, the method of obtaining a corrected bias image by performing a residual reflection separation step based on an initial bias image includes: Local bias response features are extracted from the initial bias image, and the bias change relationship between adjacent pixels is calculated using the local bias response features; Identify regions of anomalous bias by combining bias change relationships; A corrected bias image is obtained by performing local consistency correction on the initial bias image through bias anomaly regions. The corresponding offset correction residual is calculated by combining the corrected offset image and the initial offset image, and the residual reflection influence area is determined by the offset correction residual; The reflected radiation propagation intensity and spatial influence range in the reflected radiation distribution model are corrected based on the residual reflection influence area, and the reflected radiation mapping image is regenerated based on the corrected reflected radiation distribution model. The pixel response separation is re-performed using the regenerated reflectance-radiation map image, and the separation results are updated to the corrected bias image.

[0012] In one possible implementation of the first aspect, the method of feedback correction of the reflected radiation propagation intensity and spatial influence range in the reflected radiation distribution model based on the residual reflection influence area includes: Obtain the offset correction residual for each pixel within the residual reflection influence area and the reflected radiation contribution value within the residual reflection influence area; The reflection intensity deviation and spatial extension deviation are calculated based on the offset correction residual; The mean value of reflected radiation contribution in the residual reflection influence area is calculated, and the propagation intensity coefficient of the reflected radiation distribution model is determined by scaling the ratio of the reflection intensity deviation to the mean value of reflected radiation contribution according to a preset step size. The propagation intensity coefficient is used to represent the degree of retention of the reflection contribution of each pixel during the propagation of reflected radiation. The propagation spread coefficient of the reflected radiation distribution model is determined based on the spatial extension deviation and the spatial coverage area corresponding to the residual reflection influence area. The propagation spread coefficient is used to represent the adjustment range of the spatial influence range of reflected radiation in the pixel array. The intensity of reflected radiation propagation is updated based on the propagation intensity coefficient, and the spatial influence range is updated based on the propagation spread coefficient.

[0013] In one possible implementation of the first aspect, the method for determining shutterless compensation parameters by combining initial bias correction parameters, a reflected radiation distribution model, and the current thermal imaging image includes: Obtain the real-time shutter position parameters corresponding to the current acquisition of the thermal imaging image; The current reflection prediction map is generated based on the reflection radiation distribution model and real-time shutter position parameters; The initial bias correction parameters are mapped to the corresponding pixel positions in the thermal imaging image to obtain the current bias reference map; The non-uniform background response in the current thermal image is determined by combining the current bias reference map and the current reflection prediction map; Generate shutter-free compensation parameters based on the non-uniform background response.

[0014] Secondly, this application provides an electronic device, comprising: The memory is configured to store instructions; and The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the aforementioned composite compensation-based thermal imaging image drift correction method.

[0015] Thirdly, this application provides a machine-readable storage medium storing instructions that cause a machine to perform the above-described composite compensation-based thermal imaging image drift correction method.

[0016] By utilizing the continuous transition information during the shutter's transition from open to closed states, rather than relying solely on single-frame response data at the moment of complete closure, the aforementioned technical solutions can capture the gradual reflection and occlusion changes that occur during shutter movement. This increases the dimension of observable information and improves the accuracy of model construction. Acquiring the closed reference response under conditions where the detector is completely isolated from external scene radiation input ensures that the acquisition results primarily reflect the detector's internal bias, thermal offset, and system background response information. This reduces the impact of environmental scene changes on offset correction results and improves the stability of bias estimation. A correspondence between the shutter's spatial position and pixel exposure state is established, and the non-uniform reflection component is extracted using multi-frame response data, achieving preliminary decoupling between the reflection contribution and the true bias contribution in the pixel response. Compared to directly treating the entire response as the bias, this reduces the accumulation of reflection errors and improves the accuracy of bias identification. Modeling the reflection effect from a static interference quantity to a spatial propagation process, by describing the propagation intensity and spatial influence range of reflected radiation in the pixel array, avoids local reflection hotspots contaminating the overall bias estimation, improving the accuracy of true bias recovery and the spatial consistency of correction results. Using the corrected offset image after reflection separation as the calibration benchmark avoids introducing reflection noise into the subsequent compensation process, enabling the generated offset correction parameters to more accurately reflect the actual offset state of the detector and improving the reliability of the initial correction parameters. Joint modeling of historical real offset information with the current scene state reduces imaging interruptions caused by frequent shutter actions, improving the system's continuous observation capability and real-time performance. Using real-time updated shutterless correction compensation parameters to perform offset correction on the current thermal imaging image improves image uniformity, temperature measurement consistency, and target edge preservation, enhancing the output stability of the thermal imaging equipment under long-term continuous operation scenarios.

[0017] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a thermal imaging image drift correction method based on composite compensation, provided for an embodiment of this application; Figure 2 This is a schematic diagram of a process for extracting non-uniform reflection components, provided as an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0021] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0022] Figure 1 The illustration schematically shows a flowchart of a thermal imaging image drift correction method based on composite compensation according to an embodiment of this application. Figure 1 As shown in the figure, this application provides a thermal imaging image drift correction method based on composite compensation, which may include the following steps.

[0023] S110. In response to receiving the shutter closing command, during the transition period of the shutter blade movement, acquire the shutter position parameters and capture the corresponding transition frame images during the shutter closing process of multiple frames. In this embodiment, shutter position parameters may include data such as shutter displacement, shutter opening angle, shutter occlusion ratio, and shutter speed. The transition frame image is the image from the moment the shutter blades begin to block the detector's optical path until the blades are fully closed and no scene light is incident on any of the detector's pixels. When the controller of the thermal imaging device sends a closing command to the shutter drive mechanism, a high-frequency timed interrupt service routine is initiated synchronously with the command to manage data acquisition during the transition period. Shutter position parameters can be acquired by integrating position sensors, such as Hall effect sensors or photoelectric encoders, onto the shutter mechanism. Simultaneously, during the shutter blade movement, the control acquisition module continuously acquires multiple frames of thermal imaging images at a preset sampling frequency as transition frame images. For example, during the process of the shutter gradually moving from 0% occlusion to 100% occlusion, N frames are continuously acquired at fixed time intervals, with each frame recording the shutter position parameters at the corresponding moment.

[0024] S120. After the shutter is fully closed, acquire a shutter closure reference image; Upon receiving the shutter closing command, the shutter drive mechanism is controlled to gradually move the shutter blades from the open state to the closed state. After the shutter is fully closed, the thermal imaging detector continues to acquire current image data. The acquired shutter closing reference image mainly includes the detector's own bias response, system thermal background response, internal optical structure reflection response, and residual thermal radiation effects. The shutter closing reference image refers to the detector's response image to the shutter blades' own radiation and its reflected internal stray radiation when the shutter is fully closed.

[0025] S130. Construct a shutter occlusion state model based on shutter position parameters, and extract the non-uniform reflection component based on the shutter occlusion state model and the transition frame image. First, the pixel array structure information of the thermal imaging device's detector is acquired, and the spatial distribution of each pixel in the detector's imaging plane is determined. Each pixel corresponds to a unique spatial coordinate, used to describe the pixel's arrangement and adjacency relationship in the imaging area. Then, during the shutter's closing action, a preset shutter control timing signal and corresponding shutter position parameters are acquired simultaneously. The shutter control timing signal characterizes the shutter drive start time, motion duration, and sampling time information; the shutter position parameters characterize the shutter blades' occlusion position, occlusion ratio, or closing progress at each sampling time. Based on the shutter position parameters and the shutter control timing signal, the shutter exposure state at each moment is determined. The shutter exposure state characterizes the spatial distribution of shutter occlusion in the detector's imaging area at the current moment, including fully illuminated, partially occluded, and completely occluded states. After determining the shutter exposure state, the spatial distribution of detector pixels is spatially matched with the shutter exposure state. The illumination state at each moment is determined pixel by pixel. The illumination state indicates whether the corresponding pixel receives external scene radiation input at the current sampling moment, and a mapping relationship between pixel position and shutter exposure moment is established based on the illumination state. Based on this mapping relationship, a shutter occlusion state model is constructed. Then, multiple acquired transition frame images are time-synchronized with the shutter occlusion state model to establish a temporal correspondence between each transition frame image and its corresponding shutter occlusion state, ensuring that the response value of each pixel in each sampling frame can be associated with the occlusion state at the corresponding moment. Based on the established temporal correspondence, the pixel responses in multiple transition frame images are decomposed into illumination state components, dividing the pixel received signal into scene radiation response components, reflection response components, and detector background response components. Finally, by combining the pixel received signal components with the shutter occlusion state model, the response region that continues to exist and has spatial continuity during the occlusion state change is identified, and this part of the response is determined as the non-uniform reflection component, thereby achieving a preliminary separation between the reflection effect and the true bias response.

[0026] S140. Construct a reflection radiation distribution model based on the non-uniform reflection components, and separate the shutter closure reference image and the reflection radiation distribution model to obtain a corrected bias image. Based on the aforementioned extracted non-uniform reflection components, spatial statistical analysis is performed on the response changes of each pixel in the detector pixel array to calculate the spatial response change trend of each pixel within its neighborhood. Further, by analyzing the consistency of the response change direction of adjacent pixels and the degree of response enhancement within local regions, regions with continuously enhanced and locally concentrated reflection responses are identified and designated as reflected radiation clustering regions. Subsequently, regional features are extracted from the identified reflected radiation clustering regions, and the regional response fluctuation characteristics resulting from spatial changes in the response values ​​within these regions are calculated. Simultaneously, the degree of continuous response expansion within the regions is calculated to obtain spatial response continuity characteristics. After obtaining the regional response fluctuation characteristics and spatial response continuity characteristics, the credibility of the clustering regions is evaluated to determine the corresponding spatial structure confidence level. This confidence level is used to assess whether the current clustering region conforms to the reflection propagation formation law, thus avoiding random noise, local anomalous responses, or isolated hotspots being misidentified as reflection source regions. Then, the reflection propagation relationship is established by combining the spatial structure confidence level and the reflected radiation clustering regions to determine the reflection radiation propagation results. Based on the reflection radiation propagation results, the propagation intensity and spatial influence range of reflected radiation in the pixel array are further calculated. Subsequently, using the constructed reflected radiation distribution model, the reflection contribution of each pixel under shutter closure is estimated, the reflected radiation contribution value is calculated, and a reflected radiation mapping image corresponding to the shutter closure reference image is generated based on the reflected radiation contribution value. The reflected radiation mapping image is used to characterize the distribution of the theoretical reflection response in the pixel space under shutter closure. Then, using the shutter closure reference image as the original response image and the reflected radiation mapping image as the reflection estimation result, the response of each pixel is separated. After deducting the reflected radiation contribution, an initial bias image is obtained, which mainly retains the detector's true bias response and the system background response. Further, using the initial bias image as input, a residual reflection separation step is performed. By analyzing local bias anomaly regions and residual reflection characteristics, the reflected radiation distribution model is corrected, and the reflection contribution estimation and pixel response separation are re-performed, finally yielding a corrected bias image.

[0027] S150. Determine the initial bias correction parameters based on the corrected bias image; Next, the response value of each detector pixel in the image is read. This response value is the intrinsic dark level output value of that pixel under conditions of no external radiation input and elimination of shutter component reflection interference. The intrinsic dark level output values ​​of all valid pixels are statistically averaged to calculate a global mean. This global mean represents the ideal offset center level of the entire detector focal plane array in the current operating state, providing a unified baseline for subsequent correction and preventing an overall shift in the absolute brightness level of the corrected image. For each pixel position in the focal plane array, its intrinsic dark level output value is subtracted from the global mean. The resulting difference is the initial offset correction parameter for that pixel. This parameter directly quantifies the offset of that specific pixel relative to the array's average level, forming an initial offset correction coefficient matrix that corresponds one-to-one with each detector pixel. The generated initial offset correction coefficient matrix is ​​written into the non-volatile storage unit of the thermal imaging device, serving as a fixed reference lookup table for subsequent shutter-opening phases without shutter offset compensation. At this point, the initial offset correction parameters are determined.

[0028] S160. Acquire a thermal imaging image while the shutter is open, and determine the shutterless compensation parameters by combining the initial offset correction parameters, the reflected radiation distribution model, and the current thermal imaging image. Thermal imaging images are acquired while the shutter is open, and the real-time shutter position parameters corresponding to the current moment are simultaneously obtained. Further, combining a pre-constructed reflection radiation distribution model and the real-time shutter position parameters, the influence of internal reflection radiation under the current imaging conditions is predicted, generating a current reflection prediction map. This map represents the spatial distribution of the theoretical reflection response in the detector pixel array at the current moment and the contribution of each pixel to the reflection effect. Subsequently, the previously determined initial offset correction parameters are mapped according to the pixel spatial position corresponding to the current thermal imaging image to obtain the current offset reference map. The current offset reference map and the current reflection prediction map are jointly analyzed to determine the non-uniform background response in the current thermal imaging image. The non-uniform background response characterizes the non-target response components in the thermal imaging image formed by detector offset offset, internal reflection effects, and background thermal disturbances. After obtaining the non-uniform background response, it undergoes time-series analysis and background stability assessment. The background response is recursively updated based on historical offset states to generate the shutterless compensation parameters corresponding to the current moment. These shutterless compensation parameters describe the real-time offset compensation amount of the detector in the shutterless state.

[0029] S170. Correct the thermal imaging image according to the shutterless correction compensation parameters to obtain the target thermal imaging image.

[0030] For each pixel location in the current frame's raw thermal image, the target pixel value is obtained by subtracting the initial offset correction parameter and the shutterless correction compensation parameter from the raw pixel value. The raw pixel value is the original response value directly output by the detector, containing scene radiation, intrinsic offset, and shift. The initial offset correction parameter is the fixed background offset corresponding to that pixel, responsible for eliminating the detector's inherent fixed-mode noise. The shutterless correction compensation parameter is the time-varying shift correction corresponding to that pixel, responsible for compensating for offset shift and reflection environment changes since shutter correction. Subsequently, the target thermal image is obtained using the target pixel value, resulting in a clean thermal image that reflects only the true thermal radiation distribution of the scene, stripped of intrinsic offset, time-varying shift, and reflected radiation interference components.

[0031] By employing the aforementioned technical solutions, the continuous transition information during the shutter's opening and closing process replaces the traditional single-frame closure response acquisition method. This captures the reflection and occlusion changes formed during shutter movement, increasing the dimension of observable information and improving model construction accuracy. Acquiring the shutter closure reference response under conditions where the detector is isolated from external scene radiation ensures the response primarily reflects internal bias, thermal offset, and system background information, reducing environmental interference and improving bias estimation stability. Furthermore, establishing the correspondence between shutter position and pixel exposure state, and extracting the non-uniform reflection component by combining multi-frame responses, decouples reflection contribution from true bias contribution, reducing reflection error accumulation. Constructing a spatial propagation model of reflection radiation suppresses the influence of local reflection hotspots on bias estimation, improving the accuracy and spatial consistency of true bias recovery. Generating correction parameters based on the true bias and continuously generating shutterless correction compensation parameters in conjunction with the current scene reduces shutter interruptions while achieving long-term stable correction of thermal imaging images, improving image uniformity, temperature consistency, and continuous observation capabilities.

[0032] Figure 2 This application provides a schematic diagram of a process for extracting non-uniform reflection components, as illustrated in the embodiments of this application. Figure 2 As shown, in one embodiment of this example, constructing a shutter occlusion state model based on shutter position parameters, and extracting the non-uniform reflection component based on the shutter occlusion state model and the transition frame image includes: S210. Determine the spatial distribution of pixels in the detector pixel array of the thermal imaging device; S220: Determine the shutter exposure state by combining the preset shutter control timing signal and shutter position parameters; S230. Combine the pixel spatial distribution and shutter exposure state to determine the light transmission state of each pixel in the pixel array, and determine the mapping relationship between the pixel position and the shutter exposure time based on the light transmission state. S240. Construct a shutter occlusion state model based on the mapping relationship between pixel position and shutter exposure time; S250. Establish the temporal correspondence between the transition frame images and the shutter occlusion state model by combining multiple transition frame images and the shutter occlusion state model; S260. Based on the time correspondence, the light transmission state in multiple transition frame images is decomposed to obtain the pixel received signal components. S270. Extract the non-uniform reflection component by combining the pixel received signal component and the shutter occlusion state model.

[0033] Determining the pixel spatial distribution of the detector pixel array in a thermal imaging device involves acquiring the physical position of each pixel in the detector's focal plane array on a two-dimensional plane, the geometric relationships between adjacent pixels, and the overall array topology. In this embodiment, the pixel spatial distribution is used to represent the positional and neighborhood relationships of each pixel in the detector's imaging plane.

[0034] In this embodiment, the preset shutter control timing signal is generated by the device's main controller, defining the timing rhythm of each frame image acquisition. This includes a frame synchronization signal, i.e., the start time of each frame image acquisition cycle; exposure trigger signals for each row of pixels; and exposure duration, i.e., the duration of each pixel's integration from start to end. After the shutter closing command is issued, the shutter blades begin to move. The system continuously acquires the shutter position parameters of the shutter blades through position sensors or stepper motor pulse counting. Next, the exposure time window of the pixel in the current frame is determined. Based on the row exposure trigger signal of the pixel's row, the precise start time of integration can be determined; and based on the preset exposure duration, the precise end time of integration can be calculated. The time interval between these two moments is the exposure time window of that pixel. Next, based on the shutter position parameter sequence, the precise occlusion time when the edge of the shutter blade sweeps across the physical location of the pixel is reconstructed. This moment is the time point when the blade edge reaches the pixel's coordinate position. Finally, by comparing the relative relationship between the occlusion time and the exposure time window, a conclusion on the exposure state is drawn. If the occlusion occurs later than the end of the exposure, or if the edge of the blade never reaches the pixel's position during the frame, it means the pixel was never occluded by the blade throughout the exposure, and its exposure state is determined to be fully illuminated. If the occlusion occurs earlier than the start of the exposure, it means the blade completely covers the pixel before integration begins, and the pixel is occluded throughout the exposure, its exposure state is determined to be fully occluded. If the occlusion occurs exactly between the start and end of the exposure, it means the pixel transitioned from illuminated to occluded during the exposure. In this case, the frame is marked as a transition frame, and the pixel's exposure state is determined to be partially occluded.

[0035] Next, the exposure state determination result of each pixel in each frame during the shutter closing process is read, that is, whether the pixel is completely blocked, fully illuminated, or partially blocked during the exposure time of each frame. For each pixel in the array, along the time axis of shutter closing, its exposure state in each frame is checked sequentially, including the first type of pixel, whose exposure state is fully illuminated in all frames during the entire shutter closing process, indicating that the blade never reaches the position of the pixel, and its illumination state is determined to be continuously illuminated; the second type of pixel, whose exposure state is completely blocked in all frames during the entire shutter closing process, indicating that the blade has covered the pixel before the process begins, and its illumination state is determined to be continuously blocked; the third type of pixel, whose exposure state changes from fully illuminated to partially blocked and finally to completely blocked in a certain frame during the process, indicating that the pixel has experienced a switch from illumination to blocking, and its illumination state is determined to have switched. For a pixel identified as having undergone a switch, the moment when the edge of the shutter blade exactly reaches the pixel's physical coordinate position is found from the shutter position parameter sequence. This moment is the occlusion time for that pixel. This establishes a one-to-one correspondence between the pixel's row and column position and the occlusion time. This process is performed on all pixels that have undergone a switch, ultimately forming a complete spatiotemporal mapping set, containing: the pixel located in row and column X, and at what millisecond after the shutter closing command was issued, that it was occluded by the blade.

[0036] Extract all switched pixel data from the mapping relationship between pixel position and shutter exposure time, including the pixel's row and column coordinates, and the occlusion time when the shutter blade edge reaches that pixel position. Then, using surface fitting or interpolation methods, establish a continuous time function. The input of the function is the pixel's row and column coordinates, and the output is the occlusion time when the shutter blade edge reaches that position. For shutter mechanisms using simple linear motion, the occlusion boundary is usually approximated as a straight line, and the function can be simplified to a linear relationship: the occlusion time depends only on the pixel's coordinates in a certain direction, for example, linearly related to the row coordinates, but independent of the column coordinates. Based on the fitted time function, calculate the corresponding occlusion occurrence time for each pixel position in the detector array. Compare this time with the time axis of the shutter closing process to determine the occlusion state of the pixel at any given time. That is, for any pixel and any query time, compare the timing of the query time with the occlusion occurrence time of that pixel. If the query time is earlier than the occlusion occurrence time, the pixel is in a lit state at that time; if the query time is later than or equal to the occlusion occurrence time, the pixel is in an occluded state at that time. Integrating the results of the three steps forms the final shutter occlusion state model, which means that given any point in time during the shutter closing process, it can output the occlusion state distribution map of the entire detector array at that moment, clearly marking the boundary between the occluded area and the lit area, as well as whether each pixel is occluded or not.

[0037] Next, the acquisition timestamp of the transition frame image is determined. When the shutter closing command is issued, the system records the corresponding timer reading at the start of exposure for each frame, which serves as the timestamp for that frame. Then, the time reference for the shutter occlusion state model is determined. The time axis of the shutter occlusion state model uses the moment the shutter closing command is issued as the time zero point, or the moment the shutter blades begin to enter the detector's field of view as the time zero point. Regardless of the reference used, it must be consistent with the reference used for the transition frame timestamp, or have a known fixed offset for conversion. Then, using the exposure start time as the corresponding time, a correspondence is established between each transition frame and the occlusion state distribution map in the model at that moment. This method is suitable for cases with short exposure times and slow blade movement speeds, where the occlusion state change during exposure can be approximated as minimal. After performing the above correspondence operation on each frame in the transition frame image sequence, a complete frame time correspondence table is formed.

[0038] Based on the established temporal correspondence, the corresponding time in the shutter occlusion state model for each transition frame image is queried, and the occlusion state distribution map of the entire detector array at that time is extracted. The distribution map indicates which pixels are in an occluded state and which pixels are in an illuminated state at the exposure time of that frame image. According to the occlusion state changes of each pixel in the transition frame sequence, all pixels are divided into always-illuminated pixels, always-occluded pixels, and state-switching pixels. Always-illuminated pixels are never occluded by the blades throughout the entire transition period, and their signals always contain scene radiation and detector background, but not blade reflected radiation. Always-occluded pixels are completely covered by the blades before the start of the transition period, and their signals always do not contain scene radiation, but only blade reflected radiation and detector background. State-switching pixels undergo a change from illuminated to occluded, and their signal composition changes over time: before the switch, they contain scene radiation, blade reflected radiation, and detector background; after the switch, they only contain blade reflected radiation and detector background. For the third type of state-switching pixels, several frames before and after the state-switching moment are selected, and the average signal value of the pixel in these two sets of frames is calculated. Since the only change before and after the switch is the scene radiation from present to absent, while the blade reflected radiation and detector background remain stable for a short period of time, the difference between the two sets of average values ​​is the estimated value of the scene radiation component received by the pixel. For the second type of always-occluded pixels and the third type of state-switching pixels, several consecutive frames in which the pixel is in an occluded state and the blade position is stable are selected, and the average signal value of the pixel in these frames is calculated. Since the scene radiation has been blocked at this time, this average value represents the sum of the blade reflected radiation component and the detector background component. Furthermore, for the first type of always-lit pixels, since they have never been occluded, their signals lack the contribution of blade reflected radiation, and the scene radiation component obtained in the third step can be directly subtracted from their signal value to obtain the detector background component. The components obtained above are integrated to output a complete signal decomposition result for each pixel. The results obtained from the above processing are integrated to output the corresponding signal decomposition result for each pixel. The signal decomposition result includes at least the scene light component and the reflection response component. The scene light component is used to characterize the effective radiation contribution generated by the target scene, and the reflection response component is used to characterize the response information formed by non-scene radiation. The reflection response component includes the response contribution formed by the shutter blade reflection radiation and the detector background bias.

[0039] To extract the non-uniform reflection component, the shutter occlusion state model is used to first determine the state information of each pixel during the current transition exposure stage. Further, based on the occlusion and illumination states of each pixel, the pixel received signal components are spatiotemporally filtered. These components include scene light components and reflection response components. For pixels in an occluded state, since scene light input is suppressed, their remaining response is mainly formed by reflection response. Therefore, pixels in an occluded state with significant response changes are selected as target pixels to reduce the interference of scene radiation on the reflection extraction process. For any target pixel, the reflection response component is extracted from the pixel received signal components at the corresponding time. The consistency of the extracted reflection response component is verified by combining the signal change pattern during the pixel's transition from an illumination state to an occluded state. After obtaining the time alignment result, spatial interpolation is performed on the reflection response component corresponding to the target pixel to reconstruct the continuous distribution of the reflection response within the detector array space, resulting in a continuous spatial reflection field. Bilinear interpolation can be used for the spatial interpolation process. Finally, local response differences and inconsistencies in the spatial distribution are extracted from the continuous spatial reflection field and identified as non-uniform reflection components. These non-uniform reflection components are used to characterize the spatial distribution characteristics of shutter reflection path and detector response differences.

[0040] By extracting the non-uniform reflection component, the traditional method avoids the problems of reflection component aliasing, spatiotemporal misalignment, and non-uniformity correction residuals caused by ignoring the dynamic occlusion process of the shutter, thus improving the quality of infrared images.

[0041] In one embodiment of this invention, extracting the non-uniform reflection component by combining the pixel received signal component and the shutter occlusion state model includes: S310. Determine the occlusion and light transmission states of each pixel using the shutter occlusion state model; S320. Based on the occlusion state and the light transmission state, perform spatiotemporal filtering on the received signal components of the pixel to determine the target pixel that is in an occlusion state and has a reflection response. S330. For any target pixel, extract the reflection response component from the received signal component of the pixel, and verify the consistency of the reflection response component in combination with the light transmission state change process. S340. For target pixels that pass the consistency verification, determine the temporal relationship of the occlusion state, and use the temporal relationship of the occlusion state to perform time alignment on the reflection response component to obtain the time alignment result. S350. Based on the time alignment results, the reflection response components of the target pixel are spatially interpolated and reconstructed to obtain a continuous spatial reflection field. S360, Extract the non-uniform reflection component from the continuous spatial reflection field.

[0042] Based on the target transition frame image to be analyzed, the specific time corresponding to the frame image is found from the previously established time correspondence. For each pixel in the detector array, the occlusion occurrence time corresponding to that pixel position is read from the shutter occlusion state model. The occlusion occurrence time refers to the precise time point when the edge of the shutter blade advances to the physical position of the pixel. This information has been calculated and stored when the shutter occlusion state model is constructed and can be directly retrieved. The query time obtained in the first step is compared with the occlusion occurrence time of the pixel obtained in the second step. If the query time is earlier than the occlusion occurrence time of the pixel, it means that the edge of the blade has not yet reached the pixel position, and the pixel is still in the light-transmitting state; if the query time is later than or equal to the occlusion occurrence time of the pixel, it means that the blade has covered the pixel, and the pixel is in the occluded state. The above comparison and judgment are performed on all pixels in the detector array one by one, and the judgment results of each pixel are summarized to determine whether each pixel position is marked as occluded or in the light-transmitting state.

[0043] Based on the shutter occlusion state model, the occlusion state change process of each pixel during the shutter closure transition period is examined. Specifically, the pixel must have undergone a transition from a lit state to an occluded state during the transition period, or it must have been in an occluded state throughout. The duration of the occlusion state must be long enough to ensure that the occlusion state has stabilized and the influence of factors such as blade vibration has been eliminated. Pixels that are always in a lit state are excluded because they are never occluded by the blades and do not contain blade reflection responses. Among the candidate pixels, further screening is performed based on spatial location. The occlusion area formed by the shutter blades on the detector surface has a certain spatial continuity. Therefore, if a candidate pixel is spatially isolated and surrounded by unoccluded pixels, it indicates that the pixel may be near the occlusion boundary and subject to diffraction effects or boundary aliasing, potentially resulting in inaccurate reflection responses. Specifically, for each candidate pixel, the occlusion state of its surrounding neighboring pixels is checked. If all neighboring pixels within a certain range of a given pixel are also determined to be occluded, the pixel is considered to be located within the occluded area and spatially reliable. Conversely, if a large number of pixels in the neighborhood are in a state of light transmission, it indicates that the pixel is close to the occlusion boundary and is therefore discarded. For candidate pixels that have passed both temporal and spatial screening, further examination is conducted to determine whether a valid reflection response component actually exists in their received signal. From the pixel's received signal components obtained in the previous steps, the reflection response component value of the pixel is extracted and compared with a preset signal strength threshold. If the reflection response component is below the threshold, it indicates that although the pixel is occluded, the signal generated by the blade reflection on it is too weak, resulting in an insufficient signal-to-noise ratio, making it unsuitable as a sample for reflection analysis. Only pixels with a reflection response component above the threshold are confirmed to have a valid reflection response, thus obtaining the target pixel. The preset signal strength threshold can be set based on the detector noise level. Select multiple consecutive frames of images in a stable state after the shutter is closed, calculate the root mean square value of the time noise of each pixel signal, and set the threshold to a multiple of the root mean square value of the noise to ensure that the effective reflection response is higher than the noise floor.

[0044] For any selected target pixel, the reflection response component values ​​of that pixel during the shutter closure transition period are extracted from its received signal components. Then, the moment when the target pixel's light-transmitting state switches is extracted from the shutter occlusion state model. According to physical laws, before and after the light-transmitting state switch, the reflection response component should meet the stability and monotonicity expectations. That is, after the light-transmitting state switch, the shutter blades have completely covered the pixel, and the blade position is relatively stable, so the value of the reflection response component should tend to be stable. As the shutter blades gradually approach the pixel, the distance between the blades and the pixel gradually decreases, and the reflected radiation signal usually shows a monotonically increasing trend until it stabilizes after the blades completely cover the pixel. The extracted actual reflection response time series is compared item by item with the established expected features. Several consecutive frames after the light-transmitting state switch, when the blades are completely stable, are selected, and the fluctuation amplitude of the reflection response component during this time period is calculated, which can be measured by the standard deviation or the difference between the maximum and minimum values. If the fluctuation amplitude is lower than the preset stability threshold, the stability verification is passed. The preset stability threshold can be set using a fixed empirical value. Based on the response characteristics of the detector pixels, the threshold is directly set as a certain percentage of the average value of the reflection response components, serving as the upper limit of the allowable deviation for determining stability. Next, monotonicity verification is performed. This checks whether the reflection response components show a monotonically increasing trend as the blade gradually approaches the pixel. If there is a significant decrease or disordered jump in the sequence, the monotonicity verification fails. For target pixels that pass both the stability and monotonicity verifications, their consistency verification is considered successful. The reflection response components of this pixel are marked as reliable data and retained for subsequent temporal alignment and spatial interpolation reconstruction.

[0045] For each target pixel that passes the consistency verification, the occlusion occurrence time corresponding to that pixel is queried from the shutter occlusion state model. This time is used as the time alignment reference point for that pixel. For each target pixel, a time series is extracted from its reflection response component data, centered on the occlusion occurrence time and covering a certain time range before and after it. The extracted range includes several frames before and after the occlusion occurrence. The absolute timestamp corresponding to each frame in the time series of each pixel is subtracted from the occlusion occurrence time of that pixel to obtain the time offset of each frame relative to the occlusion occurrence time. After this transformation, the reflection response time series of all target pixels are unified into the same relative time coordinate system, where time zero uniformly represents the instant when each pixel is occluded by the blade. The aligned reflection response time series of all target pixels are summarized to form the time alignment result.

[0046] From the time alignment results, the row and column coordinates of each target pixel in the detector array, as well as the reflection response value of that pixel at a specified relative time after time alignment, are extracted. Next, the spatial range and resolution of the interpolation are determined. The spatial interpolation range covers the entire effective pixel area of ​​the detector array; the target resolution of the interpolation is consistent with the original resolution of the detector array, meaning the final generated continuous spatial reflection field has the same number of rows and columns as the detector array. Based on the spatial distribution characteristics of the target pixels and the physical properties of the reflection response, the radial basis function interpolation method is selected. Using the radial basis function as the interpolation kernel, the reflection response values ​​of all target pixels are globally fitted to generate a smooth and continuous surface. For each location in the detector array not covered by target pixels, the estimated reflection response value is calculated using the interpolation method and the known reflection response values ​​of the target pixels. For locations that are originally target pixels, their original reflection response values ​​are directly retained. After completing the calculations for all locations, a continuous spatial reflection field with the same resolution as the detector array is obtained.

[0047] The reflection response values ​​of all valid pixels are read from the continuous spatial reflection field. These values ​​are statistically averaged to obtain a scalar value, which serves as the global reference level for the entire reflection field. For each pixel in the detector array, its reflection response value in the continuous spatial reflection field is subtracted from the global reference level calculated in the first step; this subtraction yields the initial value of the reflection non-uniformity component for that pixel. Next, a spatial continuity check is performed on the obtained array-wide deviation to eliminate local anomalous jump points. The array-wide deviation, verified and corrected for spatial continuity, is then output as the final reflection non-uniformity component. This component is a two-dimensional data set with the same resolution as the detector array, where each pixel location corresponds to a value representing the deviation of the reflected radiation at that pixel relative to the global reference.

[0048] By extracting the non-uniform reflection component, the accuracy and robustness of shutter blade reflection interference separation are significantly improved, providing a high-quality reflection distribution benchmark for subsequent recovery of offset images and shutter-off compensation.

[0049] In one embodiment of this invention, the process of decomposing the light transmission states in multiple transition frame images based on time correspondence to obtain pixel received signal components includes: S410. Determine the switching frame in which the light-transmitting state of each pixel changes in the transition frame image according to the time correspondence. S420. For any pixel, extract the first set of transition frame images with a preset number of frames before the switching frame and the second set of transition frame images with a preset number of frames after the switching frame, and determine the signal value sequence in each image in the first set of transition frame images and the second set of transition frame images. S430. Calculate the difference between the signal value sequences of each image in the first transition frame image set and the second transition frame image set, and use the difference as the scene light component of the pixel; S440. Select the continuous frames in the second transition frame image set whose pixels are in a continuous occlusion state and whose signal fluctuation is lower than a preset threshold as target occlusion frames based on the light transmission status of the second transition frame image set. S450. Calculate the mean value of the signal value corresponding to the pixel in all target occlusion frames, and use the mean value of the signal value as the reflection response component of the pixel. S460, The reflection response component and the scene light component are used as the pixel received signal components.

[0050] Based on the established temporal correspondence between transition frame images and shutter occlusion state models, the acquisition time and shutter position state corresponding to each transition frame image are determined. This temporal correspondence describes the mapping between image sampling time and shutter motion state. Based on the detector pixel spatial distribution and shutter exposure state, the illumination state of each pixel at each sampling time is determined, i.e., whether the corresponding pixel receives external scene radiation input, which can be divided into illumination state and occlusion state. Subsequently, a temporal analysis is performed on the illumination state of each pixel in consecutive transition frames to identify the time nodes when pixel states change. When a pixel is detected to change from a continuous illumination state to an occlusion state, or from an occlusion state back to an illumination state, the transition frame corresponding to the state change is determined as the switching frame for that pixel.

[0051] Next, based on the previously determined pixel illumination state switching results, the corresponding switching frame for the target pixel is obtained. The switching frame represents the transition frame position when the target pixel changes from an illuminated state to an occluded state, or from an occluded state to an illuminated state. Using the switching frame as the central time node, multiple consecutive transition frames are extracted before the switching frame according to a preset number of frames to form a first transition frame image set. After the switching frame, multiple consecutive transition frames are extracted according to the same or different preset number of frames to form a second transition frame image set. The preset number of frames can be set according to the detector's temporal noise characteristics. The selection of the number of frames must ensure that the averaging operation of the signal value sequence can suppress random noise to the desired level, and the number of frames is not less than the statistical sample size required for noise suppression, for example, sixteen or thirty-two frames. The first transition frame image set corresponds to the images acquired while the target pixel is still in an illuminated state, and is used to reflect the signal changes when the pixel is simultaneously affected by external scene radiation and internal response. The second transition frame image set corresponds to the images after the target pixel enters an occluded state, and is used to reflect the changes in the remaining response after the scene input gradually disappears. Subsequently, the response values ​​at the corresponding positions of the target pixels are read from the first transition frame image set and the second transition frame image set, and arranged according to the image acquisition order to form a sequence of signal values ​​corresponding to the target pixels.

[0052] A first transition frame image set and a second transition frame image set corresponding to the target pixel are acquired, and the response values ​​of the target pixel in each transition frame are extracted respectively, forming a first signal value sequence and a second signal value sequence in chronological order. The first signal value sequence corresponds to the response change sequence received when the pixel is in the light-transmitting state; the second signal value sequence corresponds to the response change sequence after the pixel enters the occluded state. The first and second signal value sequences are time-aligned to establish a matching relationship between the two sets of signals at corresponding sampling positions. Subsequently, the response difference between the first and second signal value sequences is calculated time-by-time to obtain the response change of the target pixel between the light-transmitting and occluded states. Since the external scene radiation input is weakened or blocked in the occluded state, while the detector bias response and reflection response still exist, the difference between the two sets of signals is used to characterize the effective response contribution of the pixel formed by external scene radiation. Furthermore, the obtained difference results are subjected to time-domain statistical processing, and the scene light component corresponding to the target pixel is generated by calculating the mean, weighted mean, or median.

[0053] The state of consecutive frames in the second transition frame image set is traversed to filter image frames where the target pixel is continuously occluded. Then, the response value corresponding to the target pixel is extracted from consecutive frames that meet the continuous occlusion condition, and the signal fluctuation between consecutive frames is calculated, which can be determined by the difference between adjacent frames and the local standard deviation. The signal fluctuation result is compared with a preset threshold. When the signal fluctuation corresponding to the consecutive frame is lower than the preset threshold, it is determined that the target pixel response has entered a stable stage, and the consecutive frames that meet the condition are identified as target occluded frames. The preset threshold can be set based on the steady-state statistics of the reflection response components. The standard deviation of the signal value sequence of all target pixels after occlusion stabilization is statistically analyzed, and the median or upper quartile of this statistical distribution is taken as the threshold, so that the threshold can adaptively reflect the normal fluctuation level under the current device state.

[0054] The target occlusion frame set obtained through the aforementioned screening is acquired, and the response values ​​of the target pixels in each target occlusion frame are extracted, i.e., the pixel response signal values ​​output by the target pixels in the thermal imaging detector at the acquisition time of the corresponding target occlusion frame. Further, the signal values ​​of the target pixels in each target occlusion frame are read in chronological order to form a stable occlusion response sequence corresponding to the target pixels. Since the target pixels are essentially isolated from external scene radiation input in the target occlusion frames, the pixel response at this stage mainly reflects the detector's internal background response, thermal offset response, and the influence of reflected radiation. Subsequently, the stable occlusion response sequence is statistically processed, and the mean of the signal values ​​corresponding to all target occlusion frames is calculated to obtain the stable response estimate corresponding to the target pixels. The stable response estimate is determined as the reflection response component corresponding to the target pixels, used to represent the response contribution of the target pixels formed by internal reflection and background coupling under conditions without external scene input.

[0055] The reflection response component and the scene light component are used as the pixel received signal components. The scene light component is used to characterize the contribution of external target scene radiation received by the target pixel in the light-transmitting state; the reflection response component is used to characterize the response contribution formed by internal reflection of the detector, background thermal radiation, and system coupling effects during shutter occlusion. Expressing the scene light component and the reflection response component as components of the target pixel's received signal forms the corresponding received signal component results for the target pixel, which are used to describe the response composition relationship of the pixel during the transition phase.

[0056] By determining the received signal components of the pixel, the traditional method avoids the problems of scene light and reflection component mixing, signal component extraction distortion, and non-uniformity correction baseline shift caused by blurred light state determination and improper selection of transition frames during shutter dynamic switching. It also improves the accuracy of reflection response recognition, pixel response decomposition accuracy, and the reliability of reflection non-uniformity extraction.

[0057] In one embodiment of this invention, constructing a reflection radiation distribution model based on the non-uniform reflection components, and then separating the shutter closure reference image and the reflection radiation distribution model to obtain a corrected bias image includes: S510. Calculate the spatial response variation trend of each pixel based on the non-uniform reflection component, and identify the reflected radiation accumulation area through the spatial response variation trend. S520. Determine the regional response fluctuation characteristics and spatial response continuity characteristics of the reflected radiation accumulation area by the spatial response change trend; S530. Determine the spatial structure confidence level of the reflected radiation accumulation area by combining the regional response fluctuation characteristics and the spatial response continuity characteristics. S540. Combine the confidence level of spatial structure and the reflected radiation accumulation area to determine the reflected radiation propagation results; S550. Identify the intensity and spatial range of reflected radiation propagation through the results of reflected radiation propagation, and construct a reflected radiation distribution model by combining the intensity and spatial range of reflected radiation propagation. S560. Calculate the reflected radiation contribution value based on the reflected radiation distribution model, and use the reflected radiation contribution value to generate a reflected radiation mapping image corresponding to the shutter closure reference image; S570. Using the shutter closure reference image as the original response image, the pixel response of the reflected radiation mapping image is separated to obtain the initial bias image. S580. Perform a residual reflection separation step based on the initial bias image to obtain a corrected bias image.

[0058] Centered on the target pixel, the reflection non-uniformity components of neighboring pixels within the neighborhood are obtained. Based on these components, the spatial response variation trend of the target pixel is calculated. The differences in reflection non-uniformity values ​​between the target pixel and its neighboring pixels can be calculated separately in the row and column directions. These differences are then combined into a comprehensive trend index, reflecting the severity and direction of spatial variation in reflection non-uniformity. Based on the spatial response variation trends of each pixel, clustering or connected component analysis is performed on the global pixel array to identify regions exhibiting continuously enhanced or highly concentrated spatial responses. These regions are then designated as reflected radiation clustering regions. Reflected radiation clustering regions characterize sets of pixels with significantly concentrated spatial distributions of reflected energy, reflecting potential sources of reflected radiation or areas of strong reflection.

[0059] Next, the reflected radiation clustering region, identified based on spatial response change trends, is obtained, and the non-uniform reflection components corresponding to all pixels within this region are extracted. Furthermore, statistical analysis is performed on the pixel responses within the reflected radiation clustering region to calculate the fluctuations in pixel responses with spatial location, forming regional response fluctuation characteristics. These regional response fluctuation characteristics characterize the dispersion or instability of the reflected response intensity within the region, and can be characterized by variance, standard deviation, or local extreme value rate of change. Large fluctuations in pixel responses within the region indicate the possible presence of random noise or unstructured interference; small fluctuations and a stable distribution indicate strong consistency in the reflected structure. Further, the spatial adjacency relationships of pixels within the reflected radiation clustering region are analyzed, calculating the continuous change relationship of the reflected response between adjacent pixels to form spatial response continuity characteristics. These spatial response continuity characteristics characterize the spatial extension and connectivity of the reflected response, and can be calculated using pixel connectivity length and gradient continuity.

[0060] The regional response fluctuation characteristic value and the spatial response continuity characteristic value of the reflected radiation accumulation region are read, and the two features are normalized, which can be achieved using Z-score standardization. Then, the two feature values ​​after normalization and orientation consistency processing are weighted and fused to obtain the spatial structure confidence score of each accumulation region. Weighted fusion involves assigning weight coefficients to each feature and then summing them. The weight coefficients can be preset according to the actual application scenario and detector characteristics. For example, if the detector noise is high, the internal fluctuation characteristic may be more easily affected by noise interference. In this case, the weight of the regional response fluctuation characteristic can be appropriately reduced, and the weight of the spatial response continuity characteristic can be increased. Conversely, if the shutter blade motion mechanism has high precision and stable boundaries, both can be given equal weights or optimized based on experimental data. The value obtained after weighted summation is the spatial structure confidence score of the reflected radiation accumulation region.

[0061] The propagation results of reflected radiation are determined by combining spatial structure confidence and reflected radiation accumulation regions. Reflected radiation accumulation regions identified based on spatial response change trends and their corresponding spatial structure confidence are obtained. Spatial structure confidence is used to characterize the reliability of reflected radiation accumulation regions in terms of spatial response stability and structural continuity. Further, a threshold screening is performed on the spatial structure confidence of each reflected radiation accumulation region. When the spatial structure confidence is greater than a preset confidence threshold, the reflected radiation accumulation region is determined as an effective propagation region and included in the reflected radiation propagation analysis set. The preset confidence threshold can be determined during the equipment calibration stage by repeatedly executing the reflection non-uniform component extraction process under various shutter operating conditions, statistically analyzing the confidence value distribution of all identified accumulation regions, and selecting the quantile that can effectively distinguish between true reflection accumulation and noise artifacts as the threshold. Subsequently, spatial connectivity analysis is performed on the pixels within the effective propagation region, and the propagation direction and propagation path structure within the region are determined based on the relationship between the reflection non-uniform component changes between pixels. The propagation path structure is used to characterize the spatial expansion order and energy transfer relationship of reflected radiation in the pixel array. The spatial propagation trajectory of reflected radiation can be constructed by tracing the reflection response enhancement path along the spatial gradient direction. Furthermore, the selected effective propagation area and its corresponding propagation path structure are used as the reflected radiation propagation results to characterize the actual spatial propagation distribution characteristics of reflected radiation in the detector pixel array.

[0062] The intensity and spatial influence range of reflected radiation are identified by analyzing the propagation results. For each concentrated area of ​​reflected radiation, the center is designated as the source point, and the maximum or average value of the non-uniformity of reflection within that area is taken as the peak intensity at that source point. Then, for each concentrated area, the change in the non-uniformity of reflection is tracked outward from the center. When the value decays to below a certain proportion of the peak intensity, the location is considered to be outside the substantial influence range of the concentrated area. This proportion can be preset to one-tenth or one-twentieth of the peak intensity. The spatial contour line corresponding to the peak intensity decaying to the above proportion is defined as the boundary of the spatial influence range of the concentrated area. All pixel locations within this boundary are considered to be significantly affected by the reflected radiation from the concentrated area. Next, based on the physical characteristics of reflected radiation propagating inside the detector, a spatial propagation function is established that decays from the center of each concentrated area outward. A distance attenuation function can be used, where the contribution value of reflected radiation received by a pixel location is equal to the peak intensity of the concentrated area multiplied by a coefficient that decreases with distance. This coefficient is inversely proportional to the distance from the pixel to the center of the concentrated area, or follows a Gaussian attenuation, exponential attenuation, or other laws. Next, the determined propagation intensity, spatial influence range, and propagation attenuation function of each clustering region are integrated to form a reflected radiation distribution model covering the entire detector array. For any pixel location in the detector array, the model's calculation logic is as follows: It is sequentially determined whether the pixel falls within the spatial influence range of each clustering region. For each clustering region within which the pixel falls, the reflected radiation contribution value of that clustering region to the pixel is calculated using its propagation attenuation function and peak intensity. If the pixel is simultaneously affected by multiple clustering regions, the contribution values ​​of each clustering region are superimposed. For pixels that do not fall within the influence range of any clustering region, their reflected radiation contribution value is set to zero.

[0063] The reflected radiation distribution model constructed in the preceding steps is obtained, and the pixels in the shutter-closed reference image are used as the calculation reference. The shutter-closed reference image characterizes the detector's reference response state under conditions of no external scene radiation input. The reflected radiation contribution value corresponding to each pixel position is calculated based on the reflected radiation distribution model, where the reflected radiation contribution value characterizes the degree of influence of reflected radiation on the overall response at that pixel position. Subsequently, the reflected radiation contribution values ​​of each pixel are arranged according to their spatial position to generate a reflected radiation mapping image with the same size as the shutter-closed reference image. This reflected radiation mapping image characterizes the distribution of reflection effects predicted by the reflected radiation distribution model under reference imaging conditions.

[0064] Next, a shutter closure reference image is acquired. This image characterizes the original response image output by the detector when the shutter is fully closed, including the detector's internal bias response and residual reflected radiation effects. Further, a reflected radiation mapping image spatially corresponding to the shutter closure reference image is acquired, characterizing the reflected radiation contribution distribution at each pixel location. Subsequently, pixel-level registration is performed between the shutter closure reference image and the reflected radiation mapping image, ensuring a pixel-by-pixel correspondence in spatial location. Further, a response separation operation is performed on each pixel, subtracting the reflection contribution value from the corresponding reflected radiation mapping image from the pixel response value in the shutter closure reference image to eliminate the influence of reflected radiation on the reference response. In one embodiment, pixel response separation can be expressed as: Initial bias value = Reference response value - Reflected radiation contribution value. The initial bias value characterizes the detector's inherent bias response after eliminating reflection effects. Further, the initial bias values ​​of all pixels are reconstructed according to their spatial locations to form an initial bias image.

[0065] A corrected bias image is obtained by performing a residual reflection separation step based on the initial bias image. First, an initial bias image is acquired by separating the shutter closure reference image and the reflected radiation mapping image. This initial bias image characterizes the detector bias response distribution after removing the influence of the first round of reflected radiation. Next, local region analysis is performed on the initial bias image to extract the local bias response features of each pixel. The bias change relationship is calculated based on the spatial relationship between adjacent pixels to characterize the spatial continuity and trend of the bias. Subsequently, bias anomaly regions in the initial bias image are identified based on the bias change relationship. These anomaly regions characterize local areas in the bias distribution where there are non-smooth changes or residual reflection interference. Further, local consistency correction processing is performed on the initial bias image for these anomaly regions to weaken their influence on the overall bias distribution, resulting in the corrected bias image. Based on this, the bias correction residual is calculated by combining the corrected bias image and the initial bias image. This residual correction residual characterizes the degree of residual reflection influence that has not yet been eliminated in the initial bias image, and the residual reflection influence area is determined based on the bias correction residual. Furthermore, the reflection radiation propagation intensity and spatial influence range in the reflection radiation distribution model are corrected based on the residual reflection influence area to update the reflection radiation propagation characteristics. A new reflection radiation mapping image is then generated based on the corrected reflection radiation distribution model. Finally, the pixel response separation process is performed again using the regenerated reflection radiation mapping image, and the separation result is updated to a corrected bias image, thereby achieving iterative elimination of residual reflection components.

[0066] By obtaining the corrected bias image, the accuracy of separation of reflected radiation components and the authenticity of the bias image are improved when the shutter of the thermal imaging device is closed. This avoids the problems of residual reflection aliasing, bias image distortion and non-uniformity correction reference offset caused by the insufficient characterization of the spatial non-uniform distribution characteristics of reflected radiation by the traditional global uniformity assumption or simple differential subtraction method.

[0067] In one embodiment of this invention, obtaining a corrected bias image by performing a residual reflection separation step based on an initial bias image includes: S610. Extract the local offset response features in the initial offset image, and use the local offset response features to calculate the offset change relationship between adjacent pixels. S620. Identify aberration regions in bias by combining the bias change relationship; S630. A corrected bias image is obtained by performing local consistency correction on the initial bias image through the bias anomaly region. S640. Calculate the corresponding offset correction residual by combining the corrected offset image and the initial offset image, and determine the residual reflection influence area through the offset correction residual; S650. Based on the residual reflection influence area, the reflection radiation propagation intensity and spatial influence range in the reflection radiation distribution model are corrected by feedback, and the reflection radiation mapping image is regenerated based on the corrected reflection radiation distribution model. S660. Re-perform pixel response separation using the regenerated reflective radiation map image, and update the separation result with the corrected bias image.

[0068] First, an initial bias image is acquired and spatially mapped based on the detector pixel array structure, ensuring each pixel corresponds to a unique spatial coordinate, thus forming a two-dimensional bias response distribution. Next, a local neighborhood window is constructed centered on the target pixel. This window can include eight neighboring pixels or multi-scale extended neighborhoods surrounding the target pixel. The bias values ​​of all pixels within this preset neighborhood range are extracted, constituting the local bias response features of that pixel. These local bias response features may include: the pixel's own bias value, the bias values ​​of all pixels within the neighborhood, the mean of the bias values ​​within the neighborhood, and the median of the bias values ​​within the neighborhood. For each pair of directly adjacent pixels in the detector array, the change in their bias values ​​is calculated. This is done by subtracting the bias value of one pixel from the bias value of the other, taking the absolute value, to obtain the bias change between the two adjacent pixels. After performing the above operation on all adjacent pixel pairs in the entire array, a set of bias change data is obtained. The extracted local bias response features and the calculated bias change relationship are then correlated and analyzed. For adjacent pixel pairs with large offset changes, trace back the local offset response characteristics of their location. Check if the change exceeds several times the normal fluctuation range of its neighborhood. Simultaneously, check if this location is on the edge of the influence range of a certain clustered region in the reflected radiation distribution model to determine if the abrupt change is related to residual reflection. The relationship between the local offset response characteristics and the offset changes between adjacent pixels is used as the output of this step.

[0069] To identify anomalous bias regions by combining bias change relationships, the bias change data between adjacent pixels in the entire array is first read. Next, a judgment threshold is set, which can be based on the statistical distribution of bias changes across the entire array. The mean and standard deviation of the bias changes for all adjacent pixel pairs are calculated, and the threshold is set as the mean plus several times the standard deviation; bias changes exceeding this threshold are considered anomalous. All adjacent pixel pairs in the detector array are traversed, and the bias change of each pair is compared with the judgment threshold. If the bias change is less than or equal to the threshold, the pixel pair is marked as normal. If the bias change is greater than the threshold, the pixel pair is marked as anomalous. When a pair of pixels is marked as anomalous, both pixels in the pair are marked as anomalous candidate pixels. Spatially, connectivity analysis is performed on all pixels marked as anomalous candidate pixels. A region growing operation is performed on the anomalous candidate pixels, merging spatially adjacent anomalous candidate pixels into a single connected region. An unprocessed anomalous candidate cell is selected as a seed point. The presence of other anomalous candidate cells in its four- or eight-neighborhood is checked. If any exist, they are included in the current region. The region is then expanded outwards from the newly included cell until no more anomalous candidate cells are included. After region growing, each connected set of anomalous candidate cells constitutes an initial biased anomalous region, thus yielding the biased anomalous region.

[0070] A corrected bias image is obtained by performing local consistency correction on the initial bias image through bias anomaly regions. First, the initial bias image and its corresponding bias anomaly regions are acquired. The bias anomaly regions characterize the set of pixels in the initial bias image where the bias response abruptly changes or does not conform to spatial continuity. A local correction window is constructed centered on the bias anomaly region, and the bias response distribution of normal pixels within the correction window is extracted as a reference. Using the normal bias value provided by the reference, the bias value of each pixel within the bias anomaly region is re-estimated using a spatial interpolation method. Distance-weighted interpolation can be selected. For a pixel to be corrected within the anomaly region, several sample points from the reference sample set that are close to it are taken, and weights are assigned according to the inverse of the distance, with closer sample points having higher weights. The bias values ​​of the sample points are then weighted and averaged to obtain the corrected bias value of the pixel to be corrected. The calculated corrected bias value is written one by one into the position of the corresponding pixel within the anomaly region, replacing the original bias values ​​of these pixels in the initial bias image. Iterate through all identified bias anomaly regions and perform the above operation on each region until all anomaly regions have been corrected.

[0071] The difference between the corrected bias image and the initial bias image is calculated pixel-by-pixel at each pixel location in the detector array. For any pixel, the value in the corrected bias image is subtracted from the value in the initial bias image; the difference is the bias correction residual for that pixel. A positive difference indicates that the bias value of that pixel has been increased during the correction operation. A negative difference indicates that the bias value of that pixel has been decreased during the correction operation. A zero difference indicates that the pixel has not been corrected, and its bias value remains consistent before and after the correction. The absolute value of the residual reflects the extent of the correction to that pixel. Next, a threshold is set to distinguish between the normal correction amount and the correction amount affected by residual reflection. This threshold can be set based on historical operating data. During actual use of the equipment, the bias correction residuals in each shutter correction event are continuously recorded. The distribution of these residuals is statistically analyzed using a sliding window, and the upper quantile of the absolute value distribution of the residuals within the window is taken as the dynamic threshold, allowing the threshold to adaptively adjust as the equipment ages or the environment changes. For each pixel in the detector array, the absolute value of its offset correction residual is compared with a decision threshold. If the absolute value of the residual is greater than the threshold, the pixel is marked as a candidate pixel for residual reflection impact. If the absolute value of the residual is less than or equal to the threshold, the correction amount of the pixel is considered to be within the normal fluctuation range and is not marked. A region growing operation is performed on the candidate pixels, merging spatially adjacent candidate pixels into the same connected region. After region growing, each set of connected candidate pixels constitutes a residual reflection impact region. The residual reflection impact region is used to characterize the concentrated spatial distribution range of reflection interference that has not been eliminated after local consistency correction.

[0072] The reflection radiation propagation intensity and spatial influence range in the reflection radiation distribution model are corrected based on the residual reflection influence area. This involves first obtaining the offset correction residual for each pixel within the residual reflection influence area and the reflection radiation contribution value within that area. The offset correction residual represents the remaining error between the pixel and the target reference state after offset compensation, while the reflection radiation contribution value represents the degree of reflection influence predicted by the current reflection radiation distribution model for that pixel. Further calculations of the reflection intensity deviation and spatial extension deviation are then performed based on the offset correction residual. The reflection intensity deviation is used to determine if the current model has insufficient or excessive reflection contribution retention; the spatial extension deviation is used to determine if there is insufficient or excessive diffusion in the reflection propagation range. Next, the mean reflection radiation contribution value within the residual reflection influence area is calculated, and scaling is performed according to a preset step size based on the proportional relationship between the reflection intensity deviation and the mean reflection radiation contribution value to determine the propagation intensity coefficient corresponding to the reflection radiation distribution model. Finally, the propagation spread coefficient is determined based on the spatial extension deviation and the spatial coverage area corresponding to the residual reflection influence area. The propagation spread coefficient represents the adjustment range of the spatial influence range of reflected radiation in the pixel array. When the spatial extension deviation indicates that the actual reflection influence range is greater than the model's predicted range, the propagation spread coefficient is increased to expand the propagation coverage area; when the actual influence range is smaller than the predicted range, the propagation spread coefficient is decreased to shrink the propagation range. The reflected radiation propagation intensity is updated based on the determined propagation intensity coefficient, and the spatial influence range is updated based on the propagation spread coefficient. This allows the updated reflected radiation distribution model to simultaneously correct the propagation energy attenuation law and the spatial diffusion law, thereby gradually reducing the residual reflection influence and improving the accuracy of the true offset image restoration. Next, after completing the feedback correction of the reflected radiation propagation intensity and spatial influence range, the reflected radiation contribution value corresponding to each pixel is recalculated using the updated propagation parameters. The reflected radiation distribution result is then reconstructed based on the spatial distribution relationship between pixels, generating a new reflected radiation mapping image. This reflected radiation mapping image characterizes the degree to which each pixel is affected by reflected radiation in the current scene and can serve as the basis for subsequent offset compensation and true radiation restoration, thereby reducing residual reflection errors and improving the accuracy of thermal imaging image correction.

[0073] After correcting the reflected radiation distribution model and regenerating the reflected radiation mapping image, the updated reflected radiation contribution results are used as the basis for new reflection component estimation. The pixel response signals in the thermal imaging image are then re-separated into components. Since the pixel received signal typically contains the true target radiation response, system bias response, and environmental reflected radiation response simultaneously, the reflected radiation contribution value corresponding to each pixel is recalculated using the regenerated reflected radiation mapping image. The corresponding reflection response components are then extracted from the original pixel response signals to obtain updated pixel residual response results. Subsequently, bias components are extracted from the updated pixel residual responses, and a new bias distribution result is constructed based on the pixel spatial distribution relationship. The extracted bias components are then mapped to generate a corrected bias image. This corrected bias image characterizes the true bias distribution state after removing the influence of residual reflections. It can serve as input for subsequent thermal imaging offset compensation, non-uniformity correction, or true radiation recovery, thereby reducing reflection error accumulation and improving the accuracy of thermal imaging image correction.

[0074] By obtaining the corrected bias image, the problems of residual reflection accumulation, local anomalies in the bias image, and deviations in the spatial distribution estimation of reflected radiation caused by the inability of traditional single-separation or fixed reflective radiation distribution models to adapt to local reflective characteristics differences are avoided. This improves the adaptive correction capability of the reflective radiation model, the accuracy of bias image reconstruction, and the stability of thermal imaging offset compensation.

[0075] In one embodiment of this invention, feedback correction of the reflected radiation propagation intensity and spatial influence range in the reflected radiation distribution model based on the residual reflection influence area includes: S710. Obtain the offset correction residual of each pixel in the residual reflection influence area and the reflected radiation contribution value in the residual reflection influence area. S720. Calculate the reflection intensity deviation and spatial extension deviation based on the offset correction residual; S730. Calculate the average value of reflected radiation contribution in the residual reflection influence area, and determine the propagation intensity coefficient of the reflected radiation distribution model by scaling it according to the ratio of the reflection intensity deviation to the average value of reflected radiation contribution according to the preset step size. The propagation intensity coefficient is used to represent the degree of retention of the reflection contribution of each pixel during the propagation of reflected radiation. S740. Determine the propagation spread coefficient of the reflected radiation distribution model based on the spatial extension deviation and the spatial coverage area corresponding to the residual reflection influence area. The propagation spread coefficient is used to represent the adjustment range of the spatial influence range of reflected radiation in the pixel array. S750. Update the reflected radiation propagation intensity based on the propagation intensity coefficient, and update the spatial influence range based on the propagation spread coefficient.

[0076] For each pixel within the residual reflection influence area, the corresponding offset correction residual is obtained. Specifically, the output value of the current pixel after offset compensation is calculated by difference between the corresponding reference response value, resulting in the offset correction residual. This residual represents the degree of response error remaining after correction for the current pixel. A larger offset correction residual indicates that the reflection influence at the current location has not been sufficiently eliminated or that overcompensation exists. Simultaneously, based on the current reflected radiation distribution model, the reflection propagation results for each pixel are calculated retrospectively, extracting the reflected radiation contribution value for each pixel within the residual reflection influence area. This reflected radiation contribution value represents the contribution of the reflected radiation predicted by the model to the total response, and can be calculated using reflection propagation weights, propagation attenuation results, and spatial diffusion relationships.

[0077] After obtaining the offset correction residuals for each pixel within the residual reflection influence area, an error source analysis is performed on the residuals. The residual error is split into two parts: propagation intensity error and spatial propagation error. The reflection intensity deviation is used to measure the estimation deviation of the current reflection radiation distribution model for the reflected radiation intensity within the residual reflection influence area. The offset correction residuals of all pixels within the residual reflection influence area are statistically analyzed, and the average of these residuals is taken as the reflection intensity deviation. If the average residual is positive, it indicates that the local consistency correction has increased the offset value; if the average residual is negative, it indicates that the reflected radiation intensity was overestimated during initial reflection separation, and the propagation intensity of the current model is too high. The spatial extension deviation is used to measure the estimation deviation of the current reflection radiation distribution model for the spatial influence range. The positional difference between the boundary of the residual reflection influence area and the boundary of the spatial influence range of the clustered area is calculated. If the boundary of the residual reflection influence area exceeds the boundary of the spatial influence range of the clustered area in a certain direction, the excess distance in that direction is the component of the spatial extension deviation in that direction. The maximum value of the excess distance in all directions, or the root mean square value of the excess distance in each direction, is taken as the comprehensive spatial extension deviation.

[0078] After obtaining the reflected radiation contribution value of each pixel within the residual reflection influence area, the reflected radiation contribution values ​​of all pixels within the area are statistically calculated to obtain the average reflected radiation contribution value. The calculated reflection intensity deviation is then compared with the average reflected radiation contribution value to obtain the normalized propagation deviation level. The average reflected radiation contribution value is used as a benchmark. Further, the propagation intensity coefficient is determined by scaling the ratio according to a preset step size. The preset step size limits the adjustment range of a single propagation intensity, and can be set based on the area proportion of the residual reflection influence area. When the residual area coverage is large, the step size is appropriately increased; when the residual area is concentrated in a local location, the step size is decreased to improve the spatial stability of the propagation intensity update process. The propagation intensity coefficient represents the degree of retention of the reflection contribution of each pixel during the propagation of reflected radiation. When the reflection intensity deviation is large relative to the average reflected radiation contribution value, the propagation intensity coefficient is increased to enhance the retention of reflection contribution during propagation; when the reflection intensity deviation is small, the propagation intensity coefficient is decreased to weaken the impact of reflection propagation. The updated propagation intensity coefficients are written into the reflection radiation distribution model to recalculate the reflection propagation results for each pixel.

[0079] The spatial extension deviation corresponding to the residual reflection area is read. This deviation represents the distance by which the boundary of the residual reflection area exceeds the boundary of the spatial influence range of the current model's clustered area. Simultaneously, the spatial coverage information of the residual reflection area is obtained, including the boundary contour of the area within the detector array and the extension dimensions of the area in various directions. From the current reflection radiation distribution model, the spatial influence range parameters of the clustered area associated with the residual reflection area are read, including the boundary contour of the clustered area's spatial influence range and its coverage dimensions in the row and column directions. Next, the propagation spread coefficient is calculated by dividing the spatial extension deviation by the original size of the current clustered area's spatial influence range in that direction, resulting in a spatial extension ratio. This ratio represents the relative extent by which the spatial influence range needs to expand outward. The spatial extension ratio is added to a constant, and then multiplied by a preset spread step size coefficient to obtain the propagation spread coefficient. The spread step size coefficient controls the magnitude of a single correction, and its value is typically greater than zero and does not exceed one. If the spatial extension deviation is zero, the propagation spread coefficient is one, indicating that the spatial influence range does not require adjustment. The propagation spread coefficient is applied to the boundary of the spatial influence range. This involves taking the center of the clustered region as the origin and multiplying the extension dimensions of the current spatial influence range boundary in each direction by the propagation spread coefficient to obtain the corrected boundary dimensions. The corrected spatial influence range boundary is the updated boundary.

[0080] The propagation intensity of reflected radiation is updated based on the propagation intensity coefficient, and the spatial influence range is updated based on the propagation spread coefficient. This involves multiplying the original propagation intensity by the propagation intensity coefficient to obtain the updated propagation intensity. If the propagation intensity coefficient is greater than one, the updated propagation intensity is greater than the original value, and the model's estimate of reflected radiation in that area will be enhanced. If the propagation intensity coefficient is less than one, the updated propagation intensity is less than the original value, and the model's estimate of reflected radiation in that area will be weakened. If the propagation intensity coefficient is equal to one, the propagation intensity remains unchanged. Next, the propagation spread coefficient is written into the spatial propagation parameters of the reflected radiation distribution model to update the spatial diffusion range of reflected radiation in the pixel array. Specifically, the propagation neighborhood, propagation radius, or spatial weight coverage area is adjusted according to the propagation spread coefficient, allowing the influence range of the reflected radiation propagation process to dynamically change according to the actual distribution of residual reflection. Specifically, when the propagation spread coefficient increases, the reflection propagation coverage area expands accordingly, allowing the reflection influence to propagate to more neighboring pixels; when the propagation spread coefficient decreases, the propagation range shrinks, reducing the reflection propagation influence on peripheral pixels.

[0081] The above steps improve the accuracy and iterative convergence efficiency of the adaptive correction of the parameters of the thermal imaging equipment's reflectance radiation distribution model, avoiding the problems of accumulated model estimation bias, incomplete separation of residual reflection, and slow iterative convergence caused by the inability of the traditional fixed reflectance radiation distribution model to dynamically adjust according to the actual residual reflection distribution.

[0082] In one embodiment of this example, determining the shutterless compensation parameters by combining the initial bias correction parameters, the reflected radiation distribution model, and the current thermal imaging image includes: S810: Acquire the real-time shutter position parameters when acquiring thermal imaging images; S820: Generates the current reflection prediction map based on the reflection radiation distribution model and real-time shutter position parameters; S830: Map the initial offset correction parameters to the pixel positions in the thermal imaging image to obtain the current offset reference map; S840. Combine the current bias reference map and the current reflection prediction map to determine the non-uniform background response in the current thermal imaging image; S850 generates shutter-free compensation parameters based on the non-uniform background response.

[0083] Real-time shutter position parameters are obtained when acquiring thermal imaging images. These parameters can be acquired through a position detection unit within the shutter drive mechanism, such as an angle encoder or displacement sensor. Real-time shutter position parameters include shutter position coordinates, shutter occlusion status, and shutter occlusion ratio.

[0084] The real-time shutter position parameters are input into the reflected radiation distribution model as propagation constraints. Based on the reflection propagation intensity, spatial propagation range, and inter-pixel propagation relationships in the model, the reflected radiation propagation process under the current acquisition conditions is calculated. Specifically, for each target pixel, the reflected radiation contribution value is calculated by combining the occlusion ratio corresponding to the current shutter position, the propagation path, and the neighborhood propagation relationship. The propagation results are then spatially accumulated according to the reflection propagation attenuation law to obtain the reflection influence intensity experienced by each pixel at the current moment. Further, the reflected radiation contribution values ​​corresponding to all pixels are mapped and reconstructed according to their spatial positional relationships to generate the current reflection prediction map. Each pixel in the current reflection prediction map is used to characterize the predicted reflected radiation influence degree experienced by its corresponding position under the current shutter state, serving as the input basis for subsequent pixel response separation, offset extraction, and offset compensation processing.

[0085] After obtaining the initial offset correction parameters, a mapping relationship is established between each offset correction parameter and its corresponding pixel position based on the pixel array structure of the thermal imaging detector and the spatial coordinate relationship of the image. The initial offset correction parameters describe the offset response characteristics of each pixel under the reference state and may include pixel offset values, region offset weights, etc. According to the position index of each pixel in the thermal imaging image, the corresponding initial offset correction parameters are assigned pixel by pixel to the target position. For regions that do not directly correspond, neighborhood interpolation, spatial smoothing, or region propagation are used for completion processing to construct a complete offset distribution result. Furthermore, the offset results obtained from mapping all pixels are reconstructed according to the spatial arrangement relationship of the thermal imaging image to generate the current offset reference map. Each pixel in the current offset reference map represents the expected offset response at the corresponding position at the current time.

[0086] After obtaining the current bias reference map and the current reflection prediction map, they are jointly analyzed as components of the background response affecting the thermal imaging output. The current bias reference map characterizes the basic bias distribution caused by the detector's internal bias offset, pixel response differences, and device operating state; the current reflection prediction map characterizes the reflection effect distribution formed at each pixel location after the environmental reflected radiation propagates. The current bias reference map and the current reflection prediction map are added pixel-by-pixel at each pixel location. For any pixel in the detector array, the non-uniform background response value of that pixel at the current moment is obtained by adding the bias reference value of that pixel in the current bias reference map to the predicted reflected radiation value of that pixel in the current reflection prediction map. In the current thermal image, the non-scene radiation interference experienced by each pixel is equal to the pixel's intrinsic bias plus the contribution of blade reflected radiation at the current shutter position; the sum of these two constitutes the total background response that needs to be subtracted from the original thermal image. Next, the non-uniform background response values ​​calculated pixel by pixel across the entire array are integrated into a complete image with the same resolution as the thermal imaging image, which is the non-uniform background response image. The non-uniform background response is used to characterize the background interference components in the current thermal imaging image other than the radiation from the real target, including the spatial non-uniform distribution formed by the influence of detector bias and environmental reflection.

[0087] Generating shutterless correction compensation parameters based on the non-uniform background response involves converting the determined non-uniform background response into a set of real-time compensation parameters that can be directly used to correct the current frame's thermal imaging image, serving as the final execution parameters for offset correction in shutterless mode. In this embodiment, generating shutterless correction compensation parameters based on the non-uniform background response includes: Acquire the current thermal imaging image and calculate the temporal frame difference between the current thermal imaging image and the previous thermal imaging image; Calculate the local spatial gradient value of each pixel in the current thermal imaging image, and combine the temporal frame difference and the local spatial gradient value to determine the scene activity factor of the corresponding pixel. The scene activity factor is positively correlated with both the temporal frame difference and the local spatial gradient value. The background stability weight corresponding to each pixel is determined based on the scene activity factor, and a spatiotemporally adaptive update weight matrix is ​​constructed through the background stability weight. The value of each element in the update weight matrix is ​​negatively correlated with the scene activity factor. Spatial mask filtering of the non-uniform background response is performed using the updated weight matrix to obtain the filtered temperature drift response component. The filtered temperature drift response component is combined with the shutterless compensation parameter from the previous moment for time-series smoothing accumulation to generate the current shutterless compensation parameter.

[0088] The thermal imaging acquisition module acquires the thermal image corresponding to the current moment and reads the thermal image of the previous frame adjacent to the current moment. The current thermal image and the previous frame have the same spatial resolution and pixel arrangement to ensure temporal comparability between corresponding pixels. The difference between the current and previous frames is calculated pixel-by-pixel. For any pixel in the detector array, the original response value of that pixel in the current frame is subtracted from the original response value in the previous frame; the difference is the temporal frame difference for that pixel. A positive difference indicates that the response value of that pixel in the current frame has increased compared to the previous frame, possibly corresponding to the entry or movement of an object with increased temperature at that location in the scene. A negative difference indicates that the response value of that pixel has decreased compared to the previous frame, possibly corresponding to the entry of an object with decreased temperature at that location in the scene or the removal of the original object.

[0089] The local spatial gradient value of each pixel in the current thermal image is calculated. Specifically, in the row direction, the difference in response values ​​between the left and right adjacent pixels is calculated as the row gradient. In the column direction, the difference in response values ​​between the top and bottom adjacent pixels is calculated as the column gradient. The local spatial gradient value can be the square root of the sum of the squares of the row and column gradients, or the maximum of their absolute values. For each pixel, the absolute value of its temporal frame difference and the local spatial gradient value are combined to obtain the scene activity factor for that pixel. The scene activity factor is positively correlated with both the temporal frame difference and the local spatial gradient value; that is, an increase in either input value increases the scene activity factor. The combined calculation method is to multiply the absolute value of the temporal frame difference by the local spatial gradient value; the product is the scene activity factor. This calculation method can amplify the activity of pixels with both high temporal variation and high spatial gradient, while giving a relatively low activity rating to pixels with high values ​​only in a single dimension. Scene activity factor is used to characterize the degree to which a target pixel possesses both temporal and spatial variation characteristics.

[0090] After calculating the scene activity factor for each pixel, a negative correlation is established between it and the background stability weight. That is, pixels with higher scene activity factors correspond to lower background stability weights, and vice versa. The background stability weight can be determined using the following formula: Ws = (1 - Activity Factor (x, y)) r ; Ws represents the background stability weight, and r represents the nonlinear control coefficient, which can be set according to the intensity of scene changes and the requirements for background update stability. It is used to control the degree of nonlinear adjustment in the process of mapping scene activity factors to background stability weights. For scenes with frequent changes or obvious target movement, the value of γ can be appropriately increased to enhance the weight suppression ability of highly active regions; for scenes with relatively stable changes, the value of γ can be appropriately decreased to improve the background model update speed.

[0091] Based on the established mapping relationship, for each pixel in the detector array, its scene activity factor is input into the mapping function to calculate the background stability weight corresponding to that pixel. The background stability weights of each pixel in the entire array are arranged according to the pixel's row and column position in the detector array, forming an updated weight matrix with the same resolution as the thermal imaging image. This matrix is ​​the spatiotemporally adaptive updated weight matrix. The temporal adaptability of the matrix is ​​reflected in the fact that as the device continuously acquires images and the scene content constantly changes, the scene activity factor and background stability weight are recalculated for each frame, and the updated weight matrix is ​​dynamically refreshed accordingly, always adapting to the actual state of the current scene.

[0092] Subsequently, the generated non-uniform background response image is read. Simultaneously, the spatiotemporally adaptive update weight matrix constructed in the previous step is read. The non-uniform background response image and the update weight matrix are multiplied pixel-by-pixel at each pixel location. For any pixel in the detector array, the non-uniform background response value of that pixel is multiplied by the background stability weight corresponding to that pixel in the update weight matrix; the resulting product is the weighted response value of that pixel. For static region pixels with high weights, their non-uniform background response values ​​are completely preserved or only slightly attenuated. For dynamic region pixels with low weights, their non-uniform background response values ​​are significantly attenuated or even completely reduced to zero. Spatial smoothing filtering is applied to the weighted result. Spatial smoothing filtering can be achieved using the neighborhood averaging method, that is, for each pixel, the weighted response values ​​of all pixels within a preset neighborhood are averaged, and the average value is used as the smoothed response value of that pixel. The spatially smoothed array response values ​​are then integrated into an image with the same resolution as the thermal imaging image; this is the filtered temperature drift response component image.

[0093] After obtaining the temperature drift response component of the current thermal imaging image, this component is first filtered to suppress noise and transient abnormal responses, resulting in a smoothed temperature drift response. Then, the filtered temperature drift response component is weighted and accumulated with the shutterless compensation parameters from the previous moment according to a preset smoothing coefficient to obtain the shutterless compensation parameters for the current moment. The smoothing coefficient is used to adjust the weight ratio between historical compensation values ​​and the current response to achieve smooth tracking of temperature drift changes.

[0094] This application also provides a machine-readable storage medium storing instructions that cause a machine to perform the above-described composite compensation-based thermal imaging image drift correction method.

[0095] This application also provides an electronic device, including: The memory is configured to store instructions; and The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the aforementioned composite compensation-based thermal imaging image drift correction method.

[0096] 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.

[0097] 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, as well as 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.

[0098] 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.

[0099] 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 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0101] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0102] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0103] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0104] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A thermal imaging image drift correction method based on composite compensation, characterized in that, include: In response to the received shutter closing command, during the transition period of shutter blade movement, shutter position parameters are acquired and multiple transition frame images corresponding to the shutter closing process are captured. After the shutter is fully closed, a reference image of the shutter closure is captured; A shutter occlusion state model is constructed based on shutter position parameters, and non-uniform reflection components are extracted based on the shutter occlusion state model and transition frame images. A reflection radiation distribution model is constructed based on the non-uniform reflection components, and the corrected bias image is obtained by separating the shutter closure reference image and the reflection radiation distribution model. Initial bias correction parameters are determined based on the corrected bias image; The thermal imaging image is acquired while the shutter is open, and the shutter-free compensation parameters are determined by combining the initial offset correction parameters, the reflected radiation distribution model, and the current thermal imaging image. The thermal image is corrected based on the shutterless correction compensation parameters to obtain the target thermal image; The step of constructing a shutter occlusion state model based on shutter position parameters, and extracting the non-uniform reflection component based on the shutter occlusion state model and the transition frame image includes: Determine the spatial distribution of pixels in the detector pixel array of the thermal imaging device; The shutter exposure state is determined by combining the preset shutter control timing signal and shutter position parameters; The light transmission state of each pixel in the pixel array is determined by combining the pixel spatial distribution and shutter exposure state, and the mapping relationship between the pixel position and the shutter exposure time is determined based on the light transmission state. A shutter occlusion state model is constructed based on the mapping relationship between pixel position and shutter exposure time; Establish the temporal correspondence between multiple transition frame images and shutter occlusion state models by combining multiple transition frame images and shutter occlusion state models; The pixel received signal components are obtained by decomposing the light transmission state in multiple transition frame images based on the time correspondence. The non-uniform reflection component is extracted by combining the pixel received signal components and the shutter occlusion state model.

2. The thermal imaging image drift correction method based on composite compensation according to claim 1, characterized in that, The extraction of non-uniform reflection components by combining pixel received signal components and shutter occlusion state model includes: The occlusion and light transmission states of each pixel are determined by the shutter occlusion state model; Based on the occlusion state and the light transmission state, the received signal components of the pixel are spatiotemporally filtered to determine the target pixel that is in an occlusion state and has a reflection response. For any target pixel, the reflection response component is extracted from the received signal component of the pixel, and the consistency of the reflection response component is verified by combining the light transmission state change process. For target pixels that pass the consistency verification, determine the temporal relationship of the occlusion state, and use the temporal relationship of the occlusion state to perform time alignment on the reflection response component to obtain the time alignment result; Based on the time alignment results, the reflection response components of the target pixel are spatially interpolated and reconstructed to obtain a continuous spatial reflection field; Extracting the non-uniform reflection component from a continuous spatial reflection field.

3. The thermal imaging image drift correction method based on composite compensation according to claim 1, characterized in that, The decomposition of the light transmission state in multiple transition frame images based on time correspondence to obtain pixel received signal components includes: Based on the time correspondence, determine the switching frame in which the light-transmitting state of each pixel changes in the transition frame image; For any pixel, extract the first set of transition frame images with a preset number of frames before the switching frame and the second set of transition frame images with a preset number of frames after the switching frame, and determine the signal value sequence in each image in the first set of transition frame images and the second set of transition frame images; Calculate the difference between the signal value sequences of each image in the first transition frame image set and the second transition frame image set, and use the difference as the scene light component of the pixel; The target occlusion frames are selected from the second transition frame image set by filtering the continuous frames in the second transition frame image set whose pixels are in a state of continuous occlusion and whose signal fluctuation is lower than a preset threshold. Calculate the mean of the signal values ​​corresponding to the pixels in all target occlusion frames, and use the mean of the signal values ​​as the reflection response component of the pixel; The reflection response component and the scene light component are used as the pixel received signal components.

4. The thermal imaging image drift correction method based on composite compensation according to claim 1, characterized in that, The step of constructing a reflection radiation distribution model based on the non-uniform reflection components and separating the shutter closure reference image and the reflection radiation distribution model to obtain the corrected bias image includes: The spatial response variation trend of each pixel is calculated based on the non-uniform reflection component, and the reflected radiation accumulation area is identified by the spatial response variation trend. The regional response fluctuation characteristics and spatial response continuity characteristics of the reflected radiation accumulation area are determined by the spatial response variation trend. The spatial structure confidence level of the reflected radiation accumulation region is determined by combining the regional response fluctuation characteristics and the spatial response continuity characteristics. The propagation results of reflected radiation are determined by combining the confidence level of spatial structure and the area of ​​concentrated reflected radiation. The intensity and spatial range of reflected radiation propagation are identified by the results of reflected radiation propagation, and a reflected radiation distribution model is constructed by combining the intensity and spatial range of reflected radiation propagation. The reflected radiation contribution value is calculated based on the reflected radiation distribution model, and the reflected radiation contribution value is used to generate a reflected radiation mapping image corresponding to the shutter closure reference image. The initial bias image is obtained by performing pixel response separation on the reflected radiation mapping image using the shutter closure reference image as the original response image; The corrected bias image is obtained by performing a residual reflection separation step based on the initial bias image.

5. The thermal imaging image drift correction method based on composite compensation according to claim 4, characterized in that, The step of performing residual reflection separation based on the initial bias image to obtain the corrected bias image includes: Local bias response features are extracted from the initial bias image, and the bias change relationship between adjacent pixels is calculated using the local bias response features; Identify regions of anomalous bias by combining bias change relationships; A corrected bias image is obtained by performing local consistency correction on the initial bias image through bias anomaly regions. The corresponding offset correction residual is calculated by combining the corrected offset image and the initial offset image, and the residual reflection influence area is determined by the offset correction residual; The reflected radiation propagation intensity and spatial influence range in the reflected radiation distribution model are corrected based on the residual reflection influence area, and the reflected radiation mapping image is regenerated based on the corrected reflected radiation distribution model. The pixel response separation is re-performed using the regenerated reflectance-radiation map image, and the separation results are updated to the corrected bias image.

6. The thermal imaging image drift correction method based on composite compensation according to claim 5, characterized in that, The feedback correction of the reflected radiation propagation intensity and spatial influence range in the reflected radiation distribution model based on the residual reflection influence area includes: Obtain the offset correction residual for each pixel within the residual reflection influence area and the reflected radiation contribution value within the residual reflection influence area; The reflection intensity deviation and spatial extension deviation are calculated based on the offset correction residual; The mean value of reflected radiation contribution in the residual reflection influence area is calculated, and the propagation intensity coefficient of the reflected radiation distribution model is determined by scaling the ratio of the reflection intensity deviation to the mean value of reflected radiation contribution according to a preset step size. The propagation intensity coefficient is used to represent the degree of retention of the reflection contribution of each pixel during the propagation of reflected radiation. The propagation spread coefficient of the reflected radiation distribution model is determined based on the spatial extension deviation and the spatial coverage area corresponding to the residual reflection influence area. The propagation spread coefficient is used to represent the adjustment range of the spatial influence range of reflected radiation in the pixel array. The intensity of reflected radiation propagation is updated based on the propagation intensity coefficient, and the spatial influence range is updated based on the propagation spread coefficient.

7. The thermal imaging image drift correction method based on composite compensation according to claim 1, characterized in that, The process of determining shutterless compensation parameters by combining initial bias correction parameters, reflected radiation distribution model, and current thermal imaging image includes: Acquire the real-time shutter position parameters when acquiring thermal imaging images; The current reflection prediction map is generated based on the reflection radiation distribution model and real-time shutter position parameters; The initial bias correction parameters are mapped to the pixel positions in the thermal imaging image to obtain the current bias reference map; The non-uniform background response in the current thermal image is determined by combining the current bias reference map and the current reflection prediction map; Generate shutter-free compensation parameters based on the non-uniform background response.

8. An electronic device, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the thermal imaging image drift correction method based on composite compensation according to any one of claims 1 to 7.

9. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to perform the thermal imaging image drift correction method based on composite compensation according to any one of claims 1 to 7.

Images