Thermal imaging image correction method and system based on feature correlation analysis

Abstract

The present application relates to the technical field of image correction, in particular to a thermal imaging image correction method and system based on feature correlation analysis, comprising the following steps: constructing a thermal imaging time sequence, calculating the temperature difference of adjacent frame pixels by pixel difference for the time sequence to generate a change gradient distribution, judging the continuity of the change gradient distribution and counting the number of continuous consistency to divide a trend interval, calculating and generating a gradient distribution result according to the trend interval division result, constructing a mapping based on the gradient distribution result and generating a correction mapping diagram, performing correction on the surface temperature field and redistributing the pixel position and temperature. In the present application, the number of continuous consistency of temperature change gradient is counted to divide a trend interval and perform frame-by-frame weighted update calculation, which completely gets rid of the dependence on external calibration board and feature point extraction, effectively overcomes the matching error caused by image edge blur, deeply adapts to the dynamic temperature fluctuation and thermal response difference of equipment, and practically enhances the remapping precision and gray correction consistency.

Description

Technical Field

[0001] This invention relates to the field of image correction technology, and in particular to a thermal imaging image correction method and system based on feature correlation analysis. Background Technology

[0002] Image correction technology involves error correction and consistency adjustment of acquired image data, mainly including geometric distortion correction, radiation non-uniformity correction, sensor response compensation, and multi-source image registration. This field typically revolves around the characteristics of imaging equipment, pixel grayscale distribution patterns, and spatial relationships. By establishing pixel coordinate mapping relationships, constructing grayscale response models, and combining calibration parameters, it systematically processes deformation, offset, and brightness inconsistencies in the original image to improve the consistency and usability of the image in terms of spatial structure and grayscale representation. Traditional thermal imaging image correction methods based on feature correlation analysis address pixel response differences and spatial distortion in infrared images acquired by thermal imaging equipment. First, corner or edge points in the image are extracted as feature points. Matching pairs are established by calculating the distance and positional correspondences between feature points. Then, a mapping matrix from pixel coordinates to actual spatial coordinates is calculated based on reference images acquired by a calibration board. A correction parameter table is constructed by combining the offsets of corresponding feature points between multiple images. Finally, the original thermal imaging image is remapped pixel by pixel and grayscale values ​​are corrected to complete the image correction process.

[0003] Traditional thermal imaging image correction methods based on feature correlation analysis rely on extracting image edge feature points for distance matching. Due to the low contrast and blurred edges of thermal imaging images, feature extraction is difficult and matching is prone to errors. The method of calculating the mapping matrix based on an external static calibration board reference image and constructing a parameter table by combining feature point offsets is overly dependent on external references and cannot adapt to the dynamic and minute temperature fluctuations and complex thermal response differences generated by equipment operation. This results in poor spatial remapping accuracy and limited pixel grayscale correction, directly affecting the consistency of the overall image structure and grayscale expression. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a thermal imaging image correction method based on feature correlation analysis, comprising the following steps: S1: Acquire thermal imaging images at multiple time points and sort them by time to construct a thermal imaging time series. Perform pixel-by-pixel difference calculation on the thermal imaging time series and calculate the temperature difference between adjacent frame pixels. Establish the correspondence between temperature difference and time order and generate temperature change gradient distribution. S2: Perform a continuity judgment on the temperature change gradient distribution and obtain a continuous time change direction sequence for each pixel. Perform a sign consistency comparison on the change direction sequence and count the number of consecutive consistency. Perform an interval judgment based on the relationship between the number of consecutive consistency and the threshold to generate a trend interval division result. S3: Perform segmented calculation on the temperature change gradient distribution according to the trend interval division result, perform point-by-point accumulation on continuous and consistent intervals and stop updating within the interval, and perform frame-by-frame weighting on non-continuous and consistent intervals and continuously update to generate gradient distribution results; S4: Construct a time correction mapping based on the gradient distribution results and perform a unified mapping calculation on the gradient distribution results obtained by the thermal imager. Compare the differences between adjacent pixels on the mapping relationship and generate a time correction mapping map. S5: Correct the surface temperature field of the monitored device according to the time correction map and perform a redistribution calculation on the pixel position and temperature to generate a thermal imaging image correction result.

[0005] As a further aspect of the present invention, the temperature change gradient distribution includes pixel temperature change amplitude, temperature change direction identifier, time series gradient value, and spatial distribution gradient features; the trend interval division result includes temperature continuously rising interval, temperature continuously falling interval, temperature basically stable interval, and temperature change fluctuation interval; the gradient distribution result includes interval cumulative gradient value, dynamically weighted gradient value, local region gradient representation, and overall gradient distribution state; the time correction mapping map includes pixel time offset relationship, spatial position correspondence relationship, multi-device mapping association relationship, and adjacent pixel difference relationship; and the thermal imaging image correction result includes temperature redistribution result, pixel position correction result, temperature field spatial distribution result, and feature association expression result.

[0006] As a further aspect of the present invention, the interval judgment is performed based on the relationship between the number of consecutive consistent occurrences and a threshold: the time series is divided into a stable trend interval and an unstable interval based on whether the number of consecutive consistent occurrences reaches a set threshold.

[0007] As a further aspect of the present invention, the step of performing frame-by-frame weighted and continuous updates on discontinuous and consistent intervals is as follows: within the trend discontinuity interval, the gradient value is cumulatively updated for each frame according to the weight to reflect dynamic changes.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire time-point thermal imaging images, sort them by acquisition time and label the frame sequence number, construct a time index and arrange the frame set to generate a thermal sequence frame set; S102: Based on the thermal sequence frame set, read the temperature values ​​of adjacent frames pixel by pixel and perform difference calculation, bind the difference with the time index, establish the correspondence between pixels and time, and generate a temperature difference time mapping matrix. S103: Based on the temperature difference time mapping matrix, calculate the pixel difference change rate along the time axis, reconstruct the gradient in spatial coordinates, and generate a temperature change gradient distribution.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Obtain the temperature change gradient distribution, extract the gradient sign value along the time axis pixel by pixel and record the positive and negative change states, arrange them in time order to form a pixel direction sequence, and generate a direction sequence set; S202: Based on the set of direction sequences, perform a sequential consistency comparison on symbols at consecutive time points, count the number of consecutive occurrences of the same symbol and mark the boundaries of sequence segments, establish a symbol continuous segment index relationship, and generate a continuous counting sequence. S203: Based on the continuous counting sequence, call the preset continuous count threshold to perform interval determination, identify the segments that meet the threshold conditions and divide the time range, form a pixel interval mapping relationship, and generate trend interval division results.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Obtain the trend interval division result and the temperature change gradient distribution, extract the time interval gradient value according to the pixel position and establish the correspondence between the interval index and the gradient value to generate the interval gradient sequence; S302: Based on the interval gradient sequence, perform gradient value accumulation operation point by point for continuous and consistent intervals, stop updating and record the accumulated value when the interval termination mark is reached, form an interval cumulative value set, and generate an interval accumulation matrix. S303: Based on the interval gradient sequence, extract gradient values ​​frame by frame for non-continuous and consistent intervals, call the weight parameters to perform multiplication operations, accumulate and update the calculation results in time order, merge the interval accumulation matrix, form a pixel distribution mapping relationship, and generate gradient distribution results.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the gradient distribution result, perform alignment processing according to device identifier and timestamp, unify pixel coordinates and time reference and establish correspondence between devices, and generate a multi-source gradient alignment matrix; S402: Based on the multi-source gradient alignment matrix, perform mapping calculations on gradient values ​​at the same pixel position and form a unified set of values. Establish pixel mapping index relationships based on spatial coordinates and generate a gradient mapping matrix. S403: Based on the gradient mapping matrix, extract the values ​​of adjacent positions pixel by pixel and perform difference comparison, calibrate the boundary of the mapping relationship change according to the difference distribution and construct the spatial mapping structure to generate a time correction mapping map.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Obtain the time correction mapping map and the temperature field distribution area on the device surface, extract the mapping index and temperature value according to pixel coordinates and establish the position correspondence, form a pixel relocation sequence list, and generate a position mapping matrix; S502: According to the position mapping matrix, perform a redistribution calculation on the pixel temperature values, replace and fill the temperature values ​​corresponding to the mapped positions, record the redistribution path, form a temperature rearrangement set, and generate a temperature redistribution matrix. S503: Based on the temperature redistribution matrix, perform numerical integration on all pixels and restore the spatial distribution structure, construct the arrangement relationship of the corrected image data, and generate the thermal imaging image correction result.

[0013] A thermal imaging image correction system based on feature correlation analysis includes: The time series construction and gradient acquisition module is used to implement S1: acquire thermal imaging images at multiple time points and construct thermal imaging time series by sorting them by time, perform pixel-by-pixel difference calculation on thermal imaging time series and calculate the temperature difference between adjacent frame pixels, establish the correspondence between temperature difference and time order and generate temperature change gradient distribution; The trend analysis and interval division module is used to implement S2: perform a continuity judgment on the change direction of the temperature change gradient distribution and obtain the continuous time change direction sequence of each pixel, perform a sign consistency comparison on the change direction sequence and count the number of consecutive consistency, perform interval judgment based on the relationship between the number of consecutive consistency and the threshold, and generate trend interval division results; The gradient segmentation calculation module is used to implement S3: perform segmentation calculation on the temperature change gradient distribution according to the trend interval division result, perform point-by-point accumulation on continuous and consistent intervals and stop updating within the interval, and perform frame-by-frame weighting on non-continuous and consistent intervals and continuously update to generate gradient distribution results; The time correction mapping module is used to implement S4: constructing a time correction mapping based on the gradient distribution results and performing a unified mapping calculation on the gradient distribution results obtained by the thermal imager, performing adjacent pixel difference comparison on the mapping relationship and generating a time correction mapping map; The temperature field correction module is used to implement S5: perform correction on the surface temperature field of the monitored device according to the time correction mapping and perform redistribution calculation on pixel position and temperature to generate thermal imaging image correction results.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a thermal imaging time series is constructed and pixel-by-pixel differential is performed to obtain the temperature difference between adjacent pixels. The number of consecutive consistent changes in the temperature change gradient distribution is used to divide the trend interval. The gradient distribution result is generated by segmented calculation and point-by-point accumulation or frame-by-frame weighted update based on the trend interval. Based on this result, a time correction mapping is constructed and a unified mapping is performed to form a time correction mapping map. The surface temperature field of the monitored device is corrected and the pixel position and temperature are redistributed. This completely eliminates the dependence on external calibration plates and edge feature extraction, overcomes the matching error caused by image edge blurring, deeply adapts to the dynamic temperature fluctuation of the device, and effectively enhances the spatial remapping accuracy and grayscale correction consistency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a thermal imaging image correction method based on feature correlation analysis, comprising the following steps: S1: By acquiring thermal imaging images at multiple time points, a thermal imaging time series is obtained. Then, pixel-by-pixel differential calculation is performed on the thermal imaging time series to extract the temperature change of corresponding pixels in adjacent frames and establish a correspondence with the time sequence to generate a temperature change gradient distribution. S2: Perform continuity judgment on the temperature change gradient distribution, obtain the change direction sequence of each pixel at continuous time points, perform sign consistency comparison on the change direction sequence, and perform interval judgment based on the relationship between the number of consecutive consistency and the threshold to generate trend interval division results. S3: Based on the trend interval division results, the temperature change gradient distribution is calculated in segments. For continuous and consistent intervals, point-by-point cumulative calculation is performed and updates are stopped within the interval. For non-continuous and consistent intervals, frame-by-frame weighted calculation is performed and updates are continuously performed to generate gradient distribution results. S4: Based on the gradient distribution results, perform time correction mapping construction and processing. Acquire gradient distribution results from multiple thermal imagers in the industrial equipment operation monitoring site and perform unified mapping calculation. At the same time, perform adjacent pixel difference comparison processing on the mapping relationship to generate a time correction mapping map. S5: Perform correction and adjustment processing based on the time correction map. Obtain the time correction map for the temperature field distribution area on the surface of the monitored device and perform pixel position and temperature redistribution calculation to generate thermal imaging image correction results.

[0020] The temperature change gradient distribution includes pixel temperature change amplitude, temperature change direction identifier, time series gradient value, and spatial distribution gradient features. The trend interval division results include temperature continuously rising intervals, temperature continuously falling intervals, temperature basically stable intervals, and temperature change fluctuation intervals. The gradient distribution results include interval cumulative gradient values, dynamically weighted gradient values, local region gradient representation, and overall gradient distribution status. The time correction mapping map includes pixel time offset relationships, spatial position correspondence relationships, multi-device mapping association relationships, and adjacent pixel difference relationships. The thermal imaging image correction results include temperature redistribution results, pixel position correction results, temperature field spatial distribution results, and feature association expression results.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire time-point thermal imaging images, sort them by acquisition time and label the frame sequence number, construct a time index and arrange the frame set to generate a thermal sequence frame set; An industrial-grade dual-band infrared thermal imaging sensor, configured at the monitoring site, continuously captures images of the target device surface at a fixed sampling frequency of 60 Hz. The raw infrared radiation data stream is directly read from the hardware interface, and the accompanying two-dimensional temperature distribution matrix and underlying absolute timestamp information are extracted. The raw image data is then imported into a denoising process, where a two-dimensional Gaussian filtering algorithm is used to perform spatial smoothing calculations on the two-dimensional temperature distribution matrix of each frame. Specifically, the Gaussian filter kernel size is set to a 5x5 two-dimensional matrix space, and the spatial standard deviation parameter is set to 1.5. Temperature values ​​of the center and surrounding neighborhood of each pixel are extracted. A weighted sum is then performed on these temperature values ​​according to a Gaussian weight distribution. The resulting weighted sum replaces the temperature value of the original center pixel, eliminating high-frequency hotspot noise caused by environmental interference, resulting in a cleaned net thermal imaging image. The absolute timestamp values ​​of all net thermal imaging images are obtained, and a time value comparison operation is performed. The image sequence is then rearranged according to an ascending order, ensuring strict sorting by acquisition time. Each sorted image sequence is assigned a unique identifier, starting from 1 and incrementing sequentially, to mark its frame number. The frame number, absolute timestamp value, and corresponding denoised two-dimensional temperature distribution matrix of each image are extracted and used to establish a many-to-one mapping network in the underlying storage space, constructing a time index. Based on this time index, all two-dimensional temperature distribution matrices are continuously stitched and arranged according to their frame numbers to generate a thermal sequence frame set. Continuous monitoring data with a time span of 10 seconds is acquired, generating a total of 600 net thermal imaging images. The timestamp of the first frame is read as 14:30:00:00 and the timestamp of the second frame is read as 14:30:00:16. The absolute time difference between the two is compared to determine their order. This process is repeated until the 600th frame with a timestamp of 14:30:09:984. Frame 1 is assigned sequence number 1, frame 2 is assigned sequence number 2, and so on up to sequence number 600, constructing a thermal sequence frame set containing 600 consecutive time nodes. Simultaneously, the device's horizontal resolution of 640 pixels and vertical resolution of 480 pixels are extracted, establishing that each frame contains 307,200 independent temperature monitoring nodes. The data from all 600 frames, comprising 307,200 nodes, are stacked in three-dimensional space according to their corresponding frame numbers, completing the final data architecture construction and encapsulation operation for the thermal sequence frame set.

[0022] S102: Based on the thermal sequence frame set, read the temperature values ​​of adjacent frames pixel by pixel and perform difference calculation, bind the difference with the time index, establish the correspondence between pixels and time, and generate a temperature difference time mapping matrix. Based on the thermal sequence frame set, the two-dimensional temperature distribution matrices of the current frame and the previous frame, adjacent to each other in the time index, are retrieved. Following the same horizontal and vertical coordinate mapping relationship, the actual temperature values ​​at the same spatial location in adjacent frames are read pixel by pixel. The pixel temperature value of the current frame and the pixel temperature value at the same location in the previous frame are extracted, and an arithmetic subtraction operation is performed, i.e., the temperature value of the current frame is subtracted from the temperature value of the previous frame, to obtain the difference representing the absolute temperature change of that specific pixel within this time interval. This difference result is jointly stored with the timestamp value and frame number corresponding to the current frame to establish a pixel-time correspondence. The temperature change of the coordinate position at different time nodes is arranged according to the time progression. After traversing all pixel coordinates in the image and completing the above subtraction and difference operations and relationship binding, a time change dimension is added to the two-dimensional space to generate a temperature difference time mapping matrix. The data of a specific pixel with a horizontal coordinate of 150 and a vertical coordinate of 200 is extracted, and its temperature value in the first frame is extracted as 45.2℃, and its temperature value in the second frame is extracted as 46.5℃. An arithmetic subtraction operation is performed, subtracting 45.2℃ from 46.5℃, resulting in a difference of 1.3℃. This 1.3℃ positive temperature difference value is then bound to the timestamp 14:30:00:16 of the second frame. The temperature value of this pixel in the third frame is extracted as 47.8℃, and subtracted from 46.5℃, resulting in a difference of 1.3℃. This second 1.3℃ temperature difference value is then bound to the timestamp 14:30:00:32 of the third frame. The same temporal difference operation is performed on all 307,200 pixels at a 640x480 resolution. During this process, a time interval baseline of 16 milliseconds is set. This baseline value is directly derived from the reciprocal of the 60Hz sampling frequency of the thermal imaging sensor and serves as the underlying reference baseline. After all pixels have completed the calculation of the difference between adjacent frames 599 times in the whole sequence, all temperature difference values ​​are recombined into a three-dimensional matrix according to spatial coordinates and time sequence number, and finally a temperature difference time mapping matrix with dimensions of 640 x 480 x 599 is generated.

[0023] S103: Based on the temperature difference time mapping matrix, calculate the pixel difference change rate along the time axis, reconstruct the gradient in spatial coordinates, and generate the temperature change gradient distribution; Based on the temperature difference time mapping matrix, the temperature difference values ​​of a single pixel at three consecutive time nodes are extracted. The temperature difference value at the next time node is subtracted from the temperature difference value at the previous time node to obtain the absolute change in temperature difference, representing the acceleration property. The sampling time interval value between adjacent frames is obtained, and the calculated absolute change in temperature difference is divided by this sampling time interval value to calculate the pixel difference change rate along the time axis. The horizontal and vertical coordinate values ​​of the pixel in the two-dimensional image are obtained, and the horizontal and vertical coordinates and the corresponding calculated pixel difference change rate are aligned in a three-dimensional data structure to reconstruct gradient features in spatial coordinates. After completing the change rate calculation and coordinate alignment for all pixels, a complete two-dimensional spatial temperature transient change rate layer is formed. All layers are superimposed in time order to generate a temperature change gradient distribution. For the specific pixel with horizontal coordinate 150 and vertical coordinate 200, its initial temperature difference between the second frame and the first frame is 1.3℃, and its subsequent temperature difference between the third frame and the second frame is 1.7℃. Subtraction is performed, subtracting 1.3℃ from 1.7℃, resulting in an absolute change of 0.4℃. The underlying time interval value of 16 milliseconds is retrieved and converted to 0.016 seconds. Division is performed, dividing 0.4℃ by 0.016 seconds, calculating the current pixel difference change rate as 25.0℃ per second. The adjacent pixels with x-coordinate 150 and y-coordinate 201 are extracted, with an initial temperature difference of 1.1℃ and a subsequent temperature difference of 0.8℃. Subtracting 1.1℃ from 0.8℃ yields an absolute change of -0.3℃. Dividing this by 0.016 seconds, the pixel difference change rate is calculated as -18.75℃ per second. The x-coordinates 150 and 200 are associated with the corresponding change rate of 25.0℃ per second, and the x-coordinates 201 are associated with the corresponding change rate of -18.75℃ per second, forming discrete spatial gradient nodes. The data from all spatial nodes in the entire map are aggregated, integrated, and standardized to generate the overall temperature change gradient distribution data.

[0024] Please see Figure 3 The specific steps of S2 are as follows: S201: Obtain the temperature change gradient distribution, extract the gradient sign value along the time axis pixel by pixel and record the positive and negative change states, arrange them in time order to form a pixel direction sequence, and generate a direction sequence set; The temperature change gradient distribution is obtained, and the pixel difference change rate is read pixel by pixel at each consecutive node on the time axis. The extracted pixel difference change rate values ​​are compared with zero. If the pixel difference change rate value is greater than zero, the gradient sign value is extracted as a positive indicator; if the pixel difference change rate value is less than zero, the gradient sign value is extracted as a negative indicator; if the pixel difference change rate value is exactly equal to zero, it is recorded as a stationary indicator. The positive and negative change states of the current pixel at each time node are recorded. The sign indicators of each pixel under consecutive timestamps are extracted, and the positive, negative, and stationary indicators are concatenated and arranged in chronological order from earliest to latest to form the pixel direction sequence corresponding to a single pixel. All spatial pixels are traversed and their corresponding state sequence chains are independently constructed. The sequence chains of 307,200 pixels are summarized, combined, and uniformly encoded to generate a direction sequence set. A specific pixel with an x-coordinate of 180 and a y-coordinate of 220 is extracted. The pixel difference change rates at 10 consecutive time points are 12.5℃ / s, 15.0℃ / s, 3.2℃ / s, -2.1℃ / s, -8.5℃ / s, 0.0℃ / s, 5.4℃ / s, 7.1℃ / s, 9.8℃ / s, and 11.2℃ / s, respectively. The 12.5℃ / s is compared to zero and is assigned a positive sign. Similarly, the 2nd and 3rd nodes are also assigned positive signs. The 4th node, -2.1℃ / s, is assigned a negative sign. The 5th node is also assigned a negative sign. The 6th node, 0.0℃ / s, is assigned a stable sign. The last four nodes are all greater than zero and are assigned positive signs. Sorted by time, this forms a precise pixel direction sequence consisting of the characters: positive, positive, positive, negative, negative, stable, positive, positive, positive, positive, positive. The above comparison and sign assignment logic is performed iteratively on all pixels in the entire domain, and finally all one-dimensional sequence vectors are integrated and assembled to generate a huge set of direction sequences.

[0025] S202: Based on the set of directional sequences, perform item-by-item consistency comparisons on symbols at consecutive time points, count the number of consecutive occurrences of the same symbol and mark the boundaries of sequence segments, establish the index relationship of consecutive symbol segments, and generate a continuous counting sequence. Based on the set of direction sequences, the complete pixel direction sequence corresponding to any single pixel is retrieved. Starting from the first position of the sequence, the symbol identifier of the current time node and the symbol identifier of the immediately following time node are read item by item. The two are compared to see if they belong to the same identifier type, and an item-by-item consistency comparison operation is performed. If the current symbol identifier and the next symbol identifier are both positive or both negative, an internal counter is triggered to increment by 1, counting the exact number of consecutive occurrences of the same symbol. If the current symbol identifier and the next symbol identifier are not of the same type, the counting process of the current stage is terminated, and the current frame number corresponding to the changed time node is recorded, marking it as the sequence segment boundary. The starting frame number and ending frame number that generated this boundary are extracted, and they are associated and structured with the accumulated consecutive occurrence count and the specific symbol type to establish a symbol continuous segment index relationship. The entire direction sequence chain is traversed to complete all segment marking and continuous counting operations, and all segmented data are merged to generate a continuous counting sequence. Obtain the sequence of positive, positive, positive, negative, negative, neutral, positive, positive, positive, positive, positive, starting from the first position and comparing each position sequentially. If the first positive position matches the second positive position, the counter records 2 times; if the second positive position matches the third positive position, the counter increments 3 times. When the third positive position matches the fourth negative position, a truncation boundary calibration is triggered, and frames 1 to 3 are recorded as the positive continuous segment boundary, accumulating 3 consecutive counts. Then, the counter is cleared and restarted. If the fourth negative position matches the fifth negative position, accumulating 2 counts; if the fifth negative position matches the sixth neutral position, frames 4 to 5 are truncated again as the negative continuous segment, accumulating 2 counts. After skipping the single neutral marker, the positive markers of the last 4 positions match, accumulating 4 counts. Based on this comparison and judgment logic, the start and end frame numbers, symbol direction and accumulated value of each segment are written into the corresponding matrix, and finally a continuous counting sequence containing the boundaries of each segment and the accumulated count value is generated.

[0026] S203: Based on the continuous counting sequence, call the preset continuous count threshold to perform interval determination, identify the segments that meet the threshold conditions and divide the time range, form a pixel interval mapping relationship, and generate trend interval division results; Based on the continuous counting sequence, a preset consecutive occurrence threshold is obtained. The consecutive occurrence count of each segment recorded in the continuous counting sequence is compared with this preset consecutive occurrence threshold to perform interval determination. The preset consecutive occurrence threshold is set by extracting 500 sets of historical, interference-free thermal fluctuation raw sequence data of the same equipment under normal and stable operating conditions, and counting the maximum consecutive number of random symbol flips caused by minor environmental airflows, finding that the maximum consecutive occurrence of random flips is 4. A redundancy margin of 2 is added to this, ultimately setting the preset consecutive occurrence threshold to an absolute value of 6. During interval determination, if the consecutive occurrence count of a segment is greater than or equal to 6, a monotonic trend marker is added to that segment; if the consecutive occurrence count is strictly less than 6, it is marked as a random fluctuation marker. The start and end absolute timestamps of all segments with monotonic trend markers are extracted to define precise time ranges. The corresponding pixel two-dimensional spatial coordinate coefficient values ​​are extracted, and the two-dimensional horizontal and vertical coordinates are correlated with the defined absolute time ranges to form pixel interval mapping relationships, generating trend interval division results. The pixel with x-coordinate 300 and y-coordinate 150 is read to have 11 consecutive positive occurrences between frames 15 and 25. These 11 occurrences are compared to a preset threshold of 6 consecutive occurrences; 11 is greater than 6, thus meeting the threshold condition. The absolute timestamp of frame 15 (14:30:00:240) is extracted as the starting range of the time span, and the absolute timestamp of frame 25 (14:30:00:400) is extracted as the ending range. This 160-millisecond time range is defined as the positive warming monotonic trend interval for this specific pixel. The same pixel is read to have 3 consecutive negative occurrences between frames 26 and 28; 3 is less than 6, so a random fluctuation label is assigned. After the traversal processing is complete, the assembly generates a trend interval division result containing the start and end times of the valid monotonic intervals.

[0027] Please see Figure 4 The specific steps of S3 are as follows: S301: Obtain the trend interval division results and temperature change gradient distribution, extract the time interval gradient values ​​according to pixel position and establish the correspondence between interval index and gradient value to generate interval gradient sequence; The trend interval division results and temperature change gradient distribution data are obtained, and each pixel node in the space is located one by one according to the underlying horizontal and vertical coordinate mapping table. The start and end frame numbers of all the target pixels with monotonic trend indicators recorded in the trend interval division results are extracted to form time interval data pairs. Based on the start and end frame numbers of these time interval data pairs, the pixel difference change rate values ​​of each independent time node within the exact time range are retrieved from the temperature change gradient distribution matrix by address. The extracted start and end numbers of the time intervals, the corresponding horizontal and vertical coordinates of the pixel space, and a series of pixel difference change rate values ​​sorted by time are subjected to a one-to-many database binding operation, establishing a correspondence between interval indexes and specific gradient values ​​in the cache. All bound data sets are structured and rearranged according to the spatial scanning order from left to right and top to bottom to generate interval gradient sequences. Table 1 is introduced to show in detail the specific mapping of gradient value extraction corresponding to some pixel time intervals.

[0028] Table 1. Pixel Interval Gradient Numerical Extraction Mapping Table

[0029] As shown in Table 1, the pixel difference change rate data set of three adjacent target pixels within a specific time range was extracted. Specifically, for a specific pixel with an x-coordinate of 120 and a y-coordinate of 250, within the continuous time interval from frame 30 to frame 35, six specific continuous gradient values ​​of 2.1, 2.4, 2.7, 3.1, 3.5, and 3.8 were extracted frame by frame in sequence according to the frame number. Similarly, the negative gradient values ​​of the pixel with an x-coordinate of 121 and a y-coordinate of 250 were extracted within the interval from frame 15 to frame 22. These value sequences, along with their spatial coordinates and frame ranges, were sealed together to form the complete interval gradient sequence content of the entire image.

[0030] S302: Based on the interval gradient sequence, perform gradient value accumulation operation point by point for continuous and consistent intervals. Stop updating and record the accumulated value when the interval termination mark is reached to form an interval cumulative value set and generate an interval accumulation matrix. Based on the interval gradient sequence, a series of pixel difference change rate values ​​corresponding to continuous and consistent intervals with monotonic trend indicators are read. Starting from the beginning of the time interval, the first gradient value is extracted and arithmetically added to the accumulator register initialized to zero. Then, stepping to the next time node according to the frame number, the second gradient value is extracted, added to the currently stored value in the accumulator register, and the accumulator register is overwritten and updated. This monotonic superposition operation is repeatedly performed, accumulating gradient values ​​point by point, until the internal pointer detects that the current reading time node equals the end frame number of the interval. At this point, a stop update operation instruction is triggered, and accumulation stops. The final result stored in the accumulator register at this time is extracted and recorded as the accumulated value. This final accumulated value is forcibly bound and stored with specific pixel 2D coordinates and the corresponding interval number, forming an interval cumulative value set. After traversing all pixels with monotonic trend intervals in the entire image and completing the above addition and summation operation, all independently generated cumulative values ​​are spatially arranged according to the pixel physical 2D coordinate grid position to generate an interval accumulation matrix. For the specific pixel with horizontal coordinate 120 and vertical coordinate 250, its gradient values ​​of 2.1, 2.4, 2.7, 3.1, 3.5, and 3.8 are retrieved sequentially within a consistent interval from frame 30 to frame 35. Addition is performed: the first value 2.1 is added to the second value 2.4 to obtain 4.5. This 4.5 is added to the third value 2.7 to obtain 7.2. This 7.2 is added to the fourth value 3.1 to obtain 10.3. This 10.3 is added to the fifth value 3.5 to obtain 13.8. Finally, 13.8 is added to the last value 3.8 to calculate the final accumulated absolute value of 17.6. When the data cursor reaches frame 35, the update stop condition is met. 17.6 is used as the representative value of this pixel node within this time period and written into the grid space corresponding to horizontal coordinate 120 and vertical coordinate 250 in the interval accumulation matrix.

[0031] S303: Based on the interval gradient sequence, the gradient values ​​are extracted frame by frame for non-continuous and consistent intervals, and the weight parameters are called to perform the product operation. The calculation results are accumulated and updated in time order, and the interval accumulation matrix is ​​merged to form a pixel distribution mapping relationship and generate the gradient distribution result. Based on interval gradient sequences, non-continuous, consistent, and valid intervals with random fluctuation indicators are identified and accurately located within the sequence. The rate of change of pixel differences within each unstable interval is extracted frame by frame. A preset fluctuation weight parameter is obtained. This parameter is set by extracting the absolute amplitude of random thermal noise samples from 50 historical test batches, calculating their variance, and then performing an inversion and normalization operation on the variance to obtain a fixed constant. Here, the preset fluctuation weight parameter is set to 0.3. The rate of change of pixel differences read in the current frame is extracted and arithmetically multiplied with the preset fluctuation weight parameter of 0.3. The weight parameter is then called to perform the product operation, yielding the proportionally weakened corrected gradient value. The corrected gradient value obtained in each calculation is added to the cumulative result remaining from the previous frame, and the current calculation result is updated chronologically. After completing the correction and accumulation operation for all frames within the non-continuous consistent interval, the final corrected cumulative value is added and merged with the existing accumulated values ​​at the same pixel coordinates in the interval accumulation matrix generated in the previous steps to form a pixel distribution mapping relationship, ultimately generating the gradient distribution result. A specific pixel is determined to belong to the non-continuous consistent interval between frames 40 and 44, and its gradient values ​​for this interval are extracted as 1.5, -2.0, 1.8, -1.2, and 1.0, respectively. First, multiplication operations are performed frame by frame: 1.5 multiplied by the parameter 0.3 yields 0.45; -2.0 multiplied by 0.3 yields -0.60; 1.8 multiplied by 0.3 yields 0.54; -1.2 multiplied by 0.3 yields -0.36; and 1.0 multiplied by 0.3 yields 0.30. The addition is performed sequentially over time. Adding 0.45 to -0.60 yields -0.15, adding 0.54 gives 0.39, adding -0.36 gives 0.03, and finally adding 0.30 gives a final corrected cumulative value of 0.33 for the discontinuous interval. The monotonic trend interval value of 17.6 for this pixel is retrieved from the interval accumulation matrix. This value is then combined with the 0.33 obtained here, yielding 17.93. This 17.93 is established as the final gradient distribution result for this pixel node.

[0032] Please see Figure 5 The specific steps of S4 are as follows: S401: Obtain gradient distribution results, perform alignment processing according to device identifier and timestamp, unify pixel coordinates and time base and establish correspondence between devices, and generate multi-source gradient alignment matrix; The process involves acquiring gradient distribution data independently generated and uploaded by multiple infrared thermal imagers from different spatial shooting angles. The hardware media access control address information encapsulated in each gradient distribution result data packet is extracted as a unique device identifier, along with the initial synchronization absolute timestamp in the packet header. The time reference attached to the primary thermal imager (device identifier 1) is forcibly selected as the zero-point reference, and the initial synchronization absolute timestamp attached to the packet header of the auxiliary thermal imager (device identifier 2) is extracted. The absolute timestamp of thermal imager 2 is subtracted from the absolute timestamp of thermal imager 1, and the specific time offset is calculated using the difference. This time offset value is then uniformly subtracted from the timestamps of all subsequent sequence data from thermal imager 2 to complete the timestamp alignment between devices. Subsequently, the physical feature parameters of the spatial calibration target board deployed on-site are acquired, and the horizontal and vertical coordinates of pixels obtained by different devices from shooting specific crosshair feature points on the same target board are extracted. Based on the offset distance of the same feature points, the affine transformation matrix parameters for transforming the coordinate system of thermal imager 2 to that of thermal imager 1 are calculated. This affine transformation matrix is ​​then used to perform matrix multiplication and translation addition transformations on the coordinates of all pixel grids output by thermal imager 2, unifying the coordinate and time references of all pixels. Multi-source fused data with identical absolute timestamps and unified physical coordinates after transformation are associated, bound, and recorded to establish a clear mapping correspondence between devices, generating a multi-source gradient alignment matrix. The initial timestamp of the data packet from device 1 is recorded as 10:00:00:100 milliseconds, and the initial timestamp of the data packet from device 2 is recorded as 10:00:00:120 milliseconds. The offset between the two is calculated to be 20 milliseconds. Alignment is achieved by subtracting 20 milliseconds from all timestamp data on the timeline of device 2. The affine transformation formula is applied to precisely map the pixel node in device 2 with coordinates of x-coordinate 200 and y-coordinate 300 to x-coordinate 180 and y-coordinate 280 in the viewpoint of device 1. The adjusted time axis and coordinate grid information are solidified to generate a multi-source gradient alignment matrix.

[0033] S402: Based on the multi-source gradient alignment matrix, perform mapping calculations on gradient values ​​at the same pixel position and form a unified set of values. Establish pixel mapping index relationships based on spatial coordinates and generate a gradient mapping matrix. Based on the multi-source gradient alignment matrix, retrieve all multi-source gradient value entries associated with the same physical grid pixel position at the same absolute calibration time point. Extract the gradient value recorded by device 1 at the unified coordinate position, and simultaneously extract the aligned gradient value recorded by device 2 after affine transformation to the same coordinate position. Calculate the arithmetic mean of these two independent gradient values. Specifically, perform an addition operation on the gradient value extracted by device 1 and the aligned gradient value extracted by device 2, then divide the sum by the total number of devices involved in the calculation (2), and perform spatial data fusion mapping calculation. Use the finally calculated average gradient value as the unique representation value of the actual spatial physical position, eliminating the original device attribute differences to form a unified value set. Read the two-dimensional spatial horizontal and vertical coordinates under this unified coordinate system, and establish a one-to-one correspondence retrieval pointer between the horizontal and vertical coordinate combinations and the unique average gradient value calculated above, establishing a pixel mapping index relationship based on the unified spatial grid coordinates. Arrange all average gradient values ​​with unified index relationships in coordinate order across the entire image to generate a standardized gradient mapping matrix. To clearly demonstrate the logic of scene data integration, Table 2 is introduced to show the mapping calculation execution process of some multi-source gradient data.

[0034] Table 2 Multi-source gradient data mapping calculation table

[0035] As shown in Table 2, for the specific pixel physical location with an x-coordinate of 210 and a y-coordinate of 320 under a unified spatial coordinate system, the original gradient value of 15.4 detected by device 1 is extracted, and the gradient value of 16.2 detected by device 2 after alignment processing is also extracted. Adding 15.4 and 16.2 together yields 31.6. Dividing 31.6 by 2 yields the average gradient value of 15.8. This value of 15.8 is then used as the unique feature value for this point and filled into the corresponding coordinates of the final output gradient mapping matrix to eliminate numerical discrepancies caused by observation parallax.

[0036] S403: Based on the gradient mapping matrix, extract the values ​​of adjacent positions pixel by pixel and perform difference comparison. According to the difference distribution, calibrate the boundary of the mapping relationship change and construct the spatial mapping structure to generate a time correction mapping map. Based on the gradient mapping matrix, the average gradient value of the pre-defined target center pixel position is extracted. Following the standard eight-neighbor search rule in image processing, the average gradient values ​​of the eight surrounding nodes (top, bottom, left, right, and four diagonal adjacent nodes) are extracted. The average gradient value of the target center pixel is then subtracted from the average gradient values ​​of these eight adjacent positions to calculate the gradient differences in eight spatial dimensions. The absolute values ​​of these eight spatial differences are calculated and compared. A fixed spatial boundary gradient difference threshold of 5.0 is pre-set. The absolute values ​​of the eight extracted differences are compared one-way with the constant 5.0. If any of these eight comparisons shows an absolute difference greater than 5.0, a severe data mapping discontinuity is identified in that adjacent direction, and the target center pixel position is officially designated as a boundary feature point of the mapping relationship change. All pixel grids designated as boundary feature points in the entire image are interconnected to construct closed or open geometric contour lines, thus building an abnormal spatial mapping structure. The line information of the spatial mapping structure is overlaid on the original unified coordinate layer of the pixels to generate a time-corrected mapping map. The average gradient value of the target center pixel at coordinates x250 x250 is extracted as 20.5. The average gradient value of its right neighboring pixel at coordinates x251 x250 is extracted as 12.3. Subtraction is performed to subtract 12.3 from 20.5, and the absolute value of the difference is 8.2. The difference 8.2 is directly compared with the boundary judgment constant threshold of 5.0. The result shows that 8.2 is greater than 5.0, thus establishing a spatial physical state discontinuity between x250 and x251 at this absolute time point, confirming the mapping discontinuity. The median line of this adjacent point is marked as the boundary line. All boundary line segment data are traversed and summarized to generate a time-corrected mapping map with spatial obstruction indicators.

[0037] Please see Figure 6 The specific steps of S5 are as follows: S501: Obtain the time correction mapping map and the temperature field distribution area on the device surface, extract the mapping index and temperature value according to pixel coordinates and establish the position correspondence, form a pixel relocation sequence list, and generate a position mapping matrix; The process involves acquiring the boundary attribute identifiers of each pixel grid recorded in the time-correction mapping map generated by the preceding operations, as well as the original infrared data of the surface temperature field distribution area detected by the corresponding device. Based on the unified horizontal and vertical coordinates carried in the time-correction mapping map, the corresponding physical location is accurately retrieved and located in the unmodified original temperature field distribution data matrix. The current unified coordinate sequence is extracted as the primary location retrieval keyword, and the previously generated device affine mapping relationship index is retrieved. Simultaneously, the corresponding actual physical temperature value attached to the underlying original infrared radiation flow data is extracted. Mapping index parameters and specific temperature values ​​are precisely extracted according to the pixel coordinate system grid nodes, and a strict positional correspondence is established in the underlying database memory table. The unified spatial coordinates, the original device-specific viewpoint coordinates, and the absolute temperature value measured at that location are packaged into a standardized record entry. Following the rule of increasing horizontal and vertical coordinates, the record entry is sequentially inserted into the blank data list, forming a pixel relocation sequence table that can trace the original data. The entire surface temperature field distribution area within the calibration range is traversed to complete the retrieval and table insertion of all target pixel information, generating a complete location mapping matrix. A specific spatial node in the time-corrected mapping map is extracted, with its unified horizontal coordinate set to 300 and unified vertical coordinate set to 400. Through reverse lookup of the relational index, it is found that the node's unique physical horizontal coordinate in the initial temperature field of device 1 is 310 and its unique physical vertical coordinate is 405. The original temperature value recorded under this unique bottom-level coordinate node is extracted as 55.6℃. The unified coordinate value (horizontal 300, vertical 400) and the unique restored coordinate value (horizontal 310, vertical 405) are bound to this absolute temperature value of 55.6℃. This binding combination is written to the 10,000th row of the relocation sequence list. A full-screen, line-by-line scan is performed to scan and summarize the hundreds of thousands of record entries generated, ultimately producing a location mapping matrix with complete traceability capabilities.

[0038] S502: Based on the position mapping matrix, perform a redistribution calculation on the pixel temperature values, replace and fill the temperature values ​​corresponding to the mapped positions, record the redistribution path, form a temperature rearrangement set, and generate a temperature redistribution matrix. Based on the location mapping matrix, read the coordinates of all anomalous temperature pixels explicitly marked within the boundary region of the mapping relationship change. Retrieve the spatial mapping structure lines established in the time-corrected mapping map, and expand outwards along the direction indicated by the structure lines to find adjacent reference pixel nodes in a normal state outside the boundary. Extract the effective temperature value of this normal reference pixel node and other undisturbed effective temperature values ​​in the neighborhood of the anomalous pixel node. Perform completion calculation using the bilinear interpolation algorithm: extract the temperature values ​​of the four nearest normal pixels (top, bottom, left, right) to the anomalous pixel in terms of physical distance. Calculate the reciprocal of each of the Euclidean physical distances between these four normal pixel coordinates and the anomalous pixel coordinates as interpolation weights. Multiply each of these four effective temperature values ​​by its corresponding calculated reciprocal distance weight, add the products together, and finally divide the sum by the sum of the four weight reciprocals to obtain the corrected interpolated temperature result. Perform a redistribution calculation operation on the anomalous pixel temperature values ​​at the fault boundary, forcibly replacing their original anomalous outlier temperature values ​​with the newly calculated interpolated temperature result. The process of interpolating involves recording the transfer vectors from the reference pixel node's x and y coordinates to the modified abnormal pixel's x and y coordinates, recording the redistribution path of this interpolation replacement, and forming a temperature rearrangement set. After completing the interpolation replacement calculations for all abnormal nodes within the defined boundary area, a temperature redistribution matrix is ​​output. To clearly demonstrate the mathematical basis of the repair phase, Table 3 is introduced to show the execution details of the redistribution calculations for some pixel temperature data.

[0039] Table 3 Pixel Temperature Redistribution Calculation Table

[0040] As shown in Table 3, fault anomaly pixels with coordinates at x-coordinate 350 and y-coordinate 450 are extracted. The effective temperature of neighboring point 1 (58.2) and its corresponding weight (0.6, calculated using the reciprocal of distance) are extracted from the safe zone. Similarly, the effective temperature of neighboring point 2 (59.8) and its corresponding weight (0.4) are extracted. Multiplication of 58.2 by 0.6 yields 34.92, and multiplication of 59.8 by 0.4 yields 23.92. Addition of 34.92 and 23.92 results in a weighted redistributed temperature value of 58.84℃. This 58.84℃ value overwrites the original anomaly grid data. Using the aforementioned strict distance-proportion weighted calculation, the frequent temperature faults and data abrupt changes at multi-view stitching boundaries are effectively removed.

[0041] S503: Based on the temperature redistribution matrix, numerical integration is performed on all pixels to restore the spatial distribution structure, construct the arrangement relationship of the corrected image data, and generate thermal imaging image correction results; Based on the temperature redistribution matrix, the new set of temperature values ​​generated by bilinear interpolation and replacement calculations, as well as the original correct temperature value set retained outside the spatial change boundary region, are extracted. According to the horizontal and vertical coordinate index sequences of the unified two-dimensional coordinate system constructed after multi-source alignment, the final temperature values ​​from these two different processing sources are sequentially filled into a new and blank two-dimensional data matrix network according to the physical spatial arrangement order of progressively increasing horizontal and vertical coordinates, performing physical integration of values ​​for all spatial pixel nodes. This thoroughly restores the physical adjacency relationships between all independent pixels, recovering the spatial temperature distribution structure of the target device surface. The latest temperature distribution value matrix of each coordinate point is finally packetized and bound with the absolute timestamp data generated and uniformly aligned in the previous steps, constructing a rigorous temporal and spatial arrangement relationship for the corrected image data. This complete two-dimensional spatial temperature distribution data matrix and its accompanying timestamps are formatted and output as a standard infrared image file stream format, generating the thermal imaging image correction result. Following the specific extraction and verification instructions, the boundary point grid value of 58.84℃, after interpolation modification, and the surrounding original correct point grid values ​​of 58.2℃ and 59.8℃ were extracted from the temperature redistribution matrix. These values ​​were then filled into the new matrix strictly according to the positional order rules of horizontal coordinates 350, 349, and 351. Subsequently, this was combined with a unified timestamp of 10:00:00:120 milliseconds to encapsulate a complete calibration image. The output image was applied to experimental testing. Experimental data shows that, compared to in-situ directly stitched images without this multi-source spatial and temporal unified calibration algorithm, this implementation successfully reduced the absolute value of the temperature jump error range in the sensitive area of ​​the device stitching boundary from the original 4.5℃ to within 0.6℃, and significantly improved the smoothness of the spatial temperature gradient at the image seams.

[0042] A thermal imaging image correction system based on feature correlation analysis includes: The time series construction and gradient acquisition module is used to implement S1: acquire thermal imaging images at multiple time points and construct thermal imaging time series by sorting them by time, perform pixel-by-pixel difference calculation on thermal imaging time series and calculate the temperature difference between adjacent frame pixels, establish the correspondence between temperature difference and time order and generate temperature change gradient distribution; The trend analysis and interval division module is used to implement S2: perform a continuity judgment on the change direction of the temperature change gradient distribution and obtain the continuous time change direction sequence of each pixel, perform a sign consistency comparison on the change direction sequence and count the number of consecutive consistency, perform interval judgment based on the relationship between the number of consecutive consistency and the threshold, and generate trend interval division results; The gradient segmentation calculation module is used to implement S3: perform segmented calculation on the temperature change gradient distribution based on the trend interval division results, perform point-by-point accumulation for continuous and consistent intervals and stop updating within the interval, and perform frame-by-frame weighting for non-continuous and consistent intervals and continuously update to generate gradient distribution results; The time correction mapping module is used to implement S4: constructing a time correction mapping based on the gradient distribution results and performing unified mapping calculation on the gradient distribution results obtained by the thermal imager, performing adjacent pixel difference comparison on the mapping relationship and generating a time correction mapping map; The temperature field correction module is used to implement S5: perform correction on the surface temperature field of the monitored device according to the time correction map and perform redistribution calculation on pixel position and temperature to generate thermal imaging image correction results.

[0043] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. A thermal imaging image correction method based on feature correlation analysis, characterized in that, Includes the following steps: S1: Acquire thermal imaging images at multiple time points and sort them by time to construct a thermal imaging time series. Perform pixel-by-pixel difference calculation on the thermal imaging time series and calculate the temperature difference between adjacent frame pixels. Establish the correspondence between temperature difference and time order and generate temperature change gradient distribution. S2: Perform a continuity judgment on the temperature change gradient distribution and obtain a continuous time change direction sequence for each pixel. Perform a sign consistency comparison on the change direction sequence and count the number of consecutive consistency. Perform an interval judgment based on the relationship between the number of consecutive consistency and the threshold to generate a trend interval division result. S3: Perform segmented calculation on the temperature change gradient distribution according to the trend interval division result, perform point-by-point accumulation on continuous and consistent intervals and stop updating within the interval, and perform frame-by-frame weighting on non-continuous and consistent intervals and continuously update to generate gradient distribution results; S4: Construct a time correction mapping based on the gradient distribution results and perform a unified mapping calculation on the gradient distribution results obtained by the thermal imager. Compare the differences between adjacent pixels on the mapping relationship and generate a time correction mapping map. S5: Correct the surface temperature field of the monitored device according to the time correction map and perform a redistribution calculation on the pixel position and temperature to generate a thermal imaging image correction result.

2. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The temperature change gradient distribution includes pixel temperature change amplitude, temperature change direction identifier, time series gradient value, and spatial distribution gradient features. The trend interval division results include temperature continuously rising intervals, temperature continuously falling intervals, temperature basically stable intervals, and temperature change fluctuation intervals. The gradient distribution results include interval cumulative gradient values, dynamically weighted gradient values, local region gradient representations, and overall gradient distribution status. The time correction mapping map includes pixel time offset relationships, spatial position correspondence relationships, multi-device mapping association relationships, and adjacent pixel difference relationships. The thermal imaging image correction results include temperature redistribution results, pixel position correction results, temperature field spatial distribution results, and feature association expression results.

3. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The interval judgment is performed based on the relationship between the number of consecutive consistent occurrences and the threshold: the time series is divided into stable trend intervals and unstable intervals based on whether the number of consecutive consistent occurrences reaches the set threshold.

4. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The step of performing frame-by-frame weighted and continuous updates on discontinuous and consistent intervals is as follows: within the intervals of discontinuous trends, the gradient value is cumulatively updated for each frame according to the weight to reflect dynamic changes.

5. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire time-point thermal imaging images, sort them by acquisition time and label the frame sequence number, construct a time index and arrange the frame set to generate a thermal sequence frame set; S102: Based on the thermal sequence frame set, read the temperature values ​​of adjacent frames pixel by pixel and perform difference calculation, bind the difference with the time index, establish the correspondence between pixels and time, and generate a temperature difference time mapping matrix. S103: Based on the temperature difference time mapping matrix, calculate the pixel difference change rate along the time axis, reconstruct the gradient in spatial coordinates, and generate a temperature change gradient distribution.

6. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Obtain the temperature change gradient distribution, extract the gradient sign value along the time axis pixel by pixel and record the positive and negative change states, arrange them in time order to form a pixel direction sequence, and generate a direction sequence set; S202: Based on the set of direction sequences, perform a sequential consistency comparison on symbols at consecutive time points, count the number of consecutive occurrences of the same symbol and mark the boundaries of sequence segments, establish a symbol continuous segment index relationship, and generate a continuous counting sequence. S203: Based on the continuous counting sequence, call the preset continuous count threshold to perform interval determination, identify the segments that meet the threshold conditions and divide the time range, form a pixel interval mapping relationship, and generate trend interval division results.

7. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Obtain the trend interval division result and the temperature change gradient distribution, extract the time interval gradient value according to the pixel position and establish the correspondence between the interval index and the gradient value to generate the interval gradient sequence; S302: Based on the interval gradient sequence, perform gradient value accumulation operation point by point for continuous and consistent intervals, stop updating and record the accumulated value when the interval termination mark is reached, form an interval cumulative value set, and generate an interval accumulation matrix. S303: Based on the interval gradient sequence, extract gradient values ​​frame by frame for non-continuous and consistent intervals, call the weight parameters to perform multiplication operations, accumulate and update the calculation results in time order, merge the interval accumulation matrix, form a pixel distribution mapping relationship, and generate gradient distribution results.

8. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Obtain the gradient distribution result, perform alignment processing according to device identifier and timestamp, unify pixel coordinates and time reference and establish correspondence between devices, and generate a multi-source gradient alignment matrix; S402: Based on the multi-source gradient alignment matrix, perform mapping calculations on gradient values ​​at the same pixel position and form a unified set of values. Establish pixel mapping index relationships based on spatial coordinates and generate a gradient mapping matrix. S403: Based on the gradient mapping matrix, extract the values ​​of adjacent positions pixel by pixel and perform difference comparison, calibrate the boundary of the mapping relationship change according to the difference distribution and construct the spatial mapping structure to generate a time correction mapping map.

9. The thermal imaging image correction method based on feature correlation analysis according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Obtain the time correction mapping map and the temperature field distribution area on the device surface, extract the mapping index and temperature value according to pixel coordinates and establish the position correspondence, form a pixel relocation sequence list, and generate a position mapping matrix; S502: According to the position mapping matrix, perform a redistribution calculation on the pixel temperature values, replace and fill the temperature values ​​corresponding to the mapped positions, record the redistribution path, form a temperature rearrangement set, and generate a temperature redistribution matrix. S503: Based on the temperature redistribution matrix, perform numerical integration on all pixels and restore the spatial distribution structure, construct the arrangement relationship of the corrected image data, and generate the thermal imaging image correction result.

10. A thermal imaging image correction system based on feature correlation analysis, characterized in that, The system is used to implement the thermal imaging image correction method based on feature correlation analysis as described in any one of claims 1-9, and the system comprises: The time series construction and gradient acquisition module is used to implement S1: acquire thermal imaging images at multiple time points and construct thermal imaging time series by sorting them by time, perform pixel-by-pixel difference calculation on thermal imaging time series and calculate the temperature difference between adjacent frame pixels, establish the correspondence between temperature difference and time order and generate temperature change gradient distribution; The trend analysis and interval division module is used to implement S2: perform a continuity judgment on the change direction of the temperature change gradient distribution and obtain the continuous time change direction sequence of each pixel, perform a sign consistency comparison on the change direction sequence and count the number of consecutive consistency, perform interval judgment based on the relationship between the number of consecutive consistency and the threshold, and generate trend interval division results; The gradient segmentation calculation module is used to implement S3: perform segmentation calculation on the temperature change gradient distribution according to the trend interval division result, perform point-by-point accumulation on continuous and consistent intervals and stop updating within the interval, and perform frame-by-frame weighting on non-continuous and consistent intervals and continuously update to generate gradient distribution results; The time correction mapping module is used to implement S4: constructing a time correction mapping based on the gradient distribution results and performing a unified mapping calculation on the gradient distribution results obtained by the thermal imager, performing adjacent pixel difference comparison on the mapping relationship and generating a time correction mapping map; The temperature field correction module is used to implement S5: perform correction on the surface temperature field of the monitored device according to the time correction mapping and perform redistribution calculation on pixel position and temperature to generate thermal imaging image correction results.

Images