Fire hazard identification method and system based on image enhancement

Abstract

The present application relates to the technical field of fire monitoring, in particular to a fire hazard identification method and system based on image enhancement. The method comprises the following steps: synchronously collecting thermal images through an infrared thermal imager array arranged on the top of the inner side wall of the silo, forming a panoramic thermal image sequence through registration splicing and performing enhancement processing, and extracting a candidate thermal spot region; obtaining the spatio-temporal dynamic evolution characteristics of the thermal spot region through cross-frame tracking, and obtaining a comprehensive confidence score by weighted fusion of the spatio-temporal dynamic evolution characteristics and spatial morphological characteristics; and finally, combining a smoldering physical diffusion model verification to determine the smoldering hazard. The present application realizes early and accurate identification of smoldering hazards in the silo by fusing the dynamic evolution characteristics of the thermal spot region and the physical model verification, effectively overcoming the problems of high false alarm rate and early warning lag of the traditional static temperature threshold monitoring method.

Description

Technical Field

[0001] This invention relates to the field of fire monitoring technology, specifically to a method and system for identifying fire hazards based on image enhancement. Background Technology

[0002] Silos are commonly used as storage containers in the storage and processing of bulk materials such as coal, grain, and sawdust. During long-term storage, these materials are susceptible to slow oxidation and fermentation, which can lead to localized, sustained temperature increases and smoldering. Smoldering is highly concealed, and once outside air enters, it can easily trigger a deflagration.

[0003] Existing monitoring methods mainly rely on two types of technologies: the first is point sensor solutions, which suffer from large blind spots and delayed alarms, making it difficult to promptly detect the onset of smoldering. The second is thermal imaging technology, which achieves area monitoring, but most existing methods rely solely on static temperature thresholds from single-frame images for judgment. However, the environment inside silos is complex, with various thermal interferences such as residual heat from materials and mechanical heat dissipation, making false alarms highly likely based solely on static temperature characteristics. These methods cannot effectively distinguish developing smoldering hot spots from stable static heat sources in the early stages, resulting in insufficient early warning capabilities.

[0004] Therefore, there is an urgent need for a method that can effectively identify smoldering fires to meet the need for early and reliable identification of smoldering fire hazards in silos. Summary of the Invention

[0005] (1) Technical problems to be solved

[0006] The purpose of this invention is to provide a method and system for identifying fire hazards based on image enhancement, so as to achieve early, comprehensive and reliable identification of smoldering hazards inside silos.

[0007] (2) Technical solution

[0008] To achieve the above objectives, on the one hand, the present invention provides a fire hazard identification method based on image enhancement, the method comprising:

[0009] Step S1: Deploy an infrared thermal imager array on the top of the inner wall of the silo, the infrared thermal imager array including One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence.

[0010] Step S2: Perform cross-frame tracking and matching of the same candidate hot spot region in the candidate hot spot region image sequence; calculate the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hot spot region based on a single frame of the candidate hot spot region image; and perform weighted fusion of the spatial morphological features and spatiotemporal dynamic evolution features to obtain a comprehensive confidence score.

[0011] Step S3: When the overall confidence score exceeds the preset confidence threshold and the spatiotemporal dynamic evolution feature includes the velocity distribution feature derived from the smoldering physical diffusion model, the candidate hot spot region is determined to be a smoldering hazard.

[0012] Preferably, the method for deploying an infrared thermal imager array on the top of the inner wall of the silo includes:

[0013] Obtain the structural parameters of the target silo, including the silo's inner diameter and wall height; obtain the device parameters of the infrared thermal imager, including the horizontal and vertical field of view; and calculate the minimum number of infrared thermal imagers required to achieve 360-degree surround coverage based on the silo's inner diameter and horizontal field of view. The minimum number of infrared thermal imagers is calculated based on the silo's inner diameter, wall height, and vertical field of view. The required initial pitch angle.

[0014] Based on the structural parameters and device parameters, the minimum number of infrared thermal imagers is determined. The system checks whether a coverage blind spot exists at the initial elevation angle. If a coverage blind spot exists, the number of infrared thermal imagers is increased iteratively, and the elevation angle is recalculated until the coverage blind spot is eliminated, thus obtaining the actual required number of infrared thermal imagers. and optimize pitch angle .

[0015] Based on the number of infrared thermal imagers Calculations yielded A deployment azimuth sequence of infrared thermal imagers; based on the deployment azimuth sequence and optimized elevation angles... An array of infrared thermal imagers is deployed on the top of the inner wall of the silo.

[0016] Preferably, the method for extracting candidate hot spot regions from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence includes:

[0017] Based on the deployment azimuth sequence and optimized pitch angle, the silo material surface is divided into... A non-uniform grid region; for the For each grid region within the grid area, temperature data is statistically obtained based on a panoramic thermal image sequence within a preset historical time period; a reference temperature value is then calculated for each grid region based on the corresponding temperature data. and temperature fluctuation statistics ,in ; Obtain a frame of enhanced thermal image at the current moment; Calculate the real-time average temperature of each grid region in the frame of enhanced thermal image. According to the reference temperature value Temperature fluctuation statistics and real-time average temperature The adaptive segmentation thresholds for M grid regions were calculated. The adaptive segmentation threshold The calculation formula is:

[0018] .

[0019] in, and The weighting coefficients are preset; according to the above... Adaptive segmentation threshold for each grid region A bilinear interpolation algorithm is used to obtain a local adaptive threshold segmentation map with the same size as the enhanced thermal image frame. The enhanced thermal image frame and the local adaptive threshold segmentation map are compared pixel by pixel to obtain a binary image. If the temperature value of a pixel is higher than the threshold at the corresponding position in the threshold segmentation map, it is marked as a foreground pixel; otherwise, it is marked as a background pixel. Connected component analysis is performed on the binary image to obtain the connected regions of all foreground pixels. The bounding rectangle of each connected region is defined as a candidate hot spot region. The candidate hot spot regions are extracted from each enhanced thermal image frame in the enhanced thermal image sequence to obtain a candidate hot spot region image sequence corresponding to the enhanced thermal image sequence.

[0020] Preferably, the silo material surface is divided into sections based on the deployment azimuth sequence and optimized pitch angle. Methods for handling non-uniform grid regions include:

[0021] The internal parameters of the infrared thermal imager, including focal length and pixel size, are acquired. Based on the deployment azimuth angle, optimized elevation angle, and internal parameters, the actual projected size of each pixel on the image plane of the infrared thermal imager on the surface of the silo material is calculated through inverse projection transformation. A spatial resolution distribution map of the silo material surface is determined based on the actual projected size of each pixel. According to the spatial resolution distribution map, areas with spatial resolution greater than a preset resolution threshold are divided using a fine-grained grid, while areas with spatial resolution less than or equal to the preset resolution threshold are divided using a coarse-grained grid. A non-uniform grid region.

[0022] Preferably, the method for calculating the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hotspot region based on a single frame of the candidate hotspot region image includes:

[0023] Based on the single-frame candidate hotspot region image, the gray-level gradient distribution is obtained through Gaussian filtering and second-order derivative calculation; based on the gray-level gradient distribution, the feature elements of the degree of drastic local gray-level changes are calculated. Based on the single-frame candidate hotspot region image, a contour point set of the candidate hotspot region is obtained through a contour extraction algorithm; based on the contour point set, the contour fractal dimension feature elements are calculated using the box counting method. The feature element representing the degree of drastic change in local grayscale and contour fractal dimension eigenvalues The spatial feature vectors are obtained by combining them.

[0024] Based on the continuous multi-frame candidate hot spot region image sequence, an area change sequence is obtained by calculating the area of ​​the candidate hot spot region in each frame; the equivalent radius change rate feature element is then calculated based on the area change sequence. Based on the sequence of candidate hot spot region images in multiple consecutive frames, a boundary movement velocity sequence is obtained through cross-frame contour point matching and average displacement calculation; based on the boundary movement velocity sequence, median filtering is applied to obtain boundary outward expansion velocity feature elements. The equivalent radius change rate feature element and boundary expansion velocity characteristic element The temporal evolution feature vector is obtained by combining the features.

[0025] Preferably, the method for calculating the fractal dimension eigenvalues ​​of the contour based on the contour point set using box counting includes:

[0026] The contour point set is fitted with a straight line according to a preset length window to obtain the direction variance and fitting residual of each window; when the direction variance is not greater than a preset variance threshold and the fitting residual is not greater than a preset residual threshold, the corresponding window is marked as a regular straight line segment; the regular straight line segments are removed from the contour point set to obtain a cleaned contour point set; the cleaned contour point set is covered with a series of boxes of preset size, and the coverage number sequence is obtained; based on the coverage number sequence, the fractal dimension of the cleaned contour point set is calculated by linear fitting, and used as the contour fractal dimension feature element.

[0027] Preferably, the method for obtaining the boundary movement velocity sequence by cross-frame contour point matching and average displacement calculation based on the continuous multi-frame candidate hotspot region image sequence includes:

[0028] Obtain the hot spot region contour point set of the previous frame from the consecutive multi-frame candidate hot spot region image sequence. With the hot spot region contour point set of the current frame Based on the hot spot region contour point set of the current frame Calculate the set of outward normal unit vectors for each contour point in the current frame; for the set of contour points of the hot spot region in the current frame... Each point in the hot spot region contour point set of the previous frame A local nearest neighbor search is performed to obtain a set of candidate corresponding points. The set of candidate corresponding points is then filtered according to preset maximum allowable displacement constraints and maximum direction deviation constraints to obtain a set of valid corresponding points. A set of normal displacements is calculated based on the set of valid corresponding points and the set of outward normal unit vectors. The median is calculated based on the set of normal displacements. Finally, the instantaneous boundary spread velocity is calculated based on the median and the inter-frame time interval. ; Expand the instantaneous boundary outward velocity Write the velocity time series; perform median filtering on the velocity time series to obtain the boundary movement velocity series.

[0029] Preferably, the method for weighted fusion of the spatial feature vector and the temporal evolution feature vector to obtain the comprehensive confidence score includes:

[0030] The spatial feature vector and the temporal evolution feature vector are concatenated sequentially to obtain a fused feature vector; according to preset weight coefficients... , , , The integrated confidence score is obtained by linearly weighting the fused feature vectors. .

[0031] Preferably, the spatiotemporal dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model, including:

[0032] Obtain the boundary motion velocity sequence of consecutive frames; perform smoothing filtering on the boundary motion velocity sequence to obtain a stable velocity sequence; when the stable velocities in the stable velocity sequence are all within a preset theoretical velocity range... If the value is within a certain range, it is determined that the speed distribution characteristics are met.

[0033] Based on the same inventive concept, this invention also provides an image enhancement-based fire hazard identification system, the system comprising:

[0034] An imaging module is used to deploy an infrared thermal imager array on the top of the inner wall of the silo, the infrared thermal imager array comprising: One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence.

[0035] The characterization module is used to perform cross-frame tracking and matching of the same candidate hot spot region in the candidate hot spot region image sequence; calculate the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hot spot region based on a single frame of the candidate hot spot region image; and perform weighted fusion of the spatial morphological features and spatiotemporal dynamic evolution features to obtain a comprehensive confidence score.

[0036] The determination module is used to determine the candidate hot spot region as a smoldering hazard when the comprehensive confidence score exceeds a preset confidence threshold and the spatiotemporal dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model.

[0037] (3) Beneficial effects

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] By integrating the spatiotemporal dynamic evolution characteristics of hot spot regions with physical diffusion models, early and accurate identification of smoldering hazards is achieved, and the false alarm rate is significantly reduced. Through inverse projection non-uniform grid partitioning and local adaptive threshold segmentation technology, high-sensitivity and accurate extraction of weak thermal anomalies in complex thermal environments under panoramic coverage is achieved. Attached Figure Description

[0040] Figure 1 This is a flowchart of the fire hazard identification method based on image enhancement according to Embodiment 1 of the present invention;

[0041] Figure 2 This is a block diagram of the image enhancement-based fire hazard identification system according to Embodiment 2 of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Before providing examples, it is necessary to describe the application scenarios of this invention. This invention is particularly suitable for early fire monitoring during the storage and transportation of organic materials such as coal, grain, and biomass pellets in large silos. These materials are prone to slow oxidation, fermentation, or self-heating when piled up, gradually forming a smoldering phenomenon without a visible flame. The initial temperature rise of such smoldering fire sources is slow, their spatial distribution is concealed, and their heat release characteristics are easily confused with steady-state thermal interferences such as mechanical heat sources, localized residual heat, or solar radiation within the silo. Traditional monitoring methods struggle to achieve reliable early identification. This invention, by integrating multi-view infrared imaging, adaptive image enhancement, and spatiotemporal dynamic feature analysis, achieves sensitive detection and intelligent identification of weak and dynamic thermal anomalies across the entire surface of the silo.

[0044] Example 1: As Figure 1 As shown in the figure, this embodiment provides a fire hazard identification method based on image enhancement, the method including:

[0045] Step S1: Deploy an infrared thermal imager array on the top of the inner wall of the silo, the infrared thermal imager array including One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence.

[0046] Step S2: Perform cross-frame tracking and matching of the same candidate hot spot region in the candidate hot spot region image sequence; calculate the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hot spot region based on a single frame of the candidate hot spot region image; and perform weighted fusion of the spatial morphological features and spatiotemporal dynamic evolution features to obtain a comprehensive confidence score.

[0047] Step S3: When the overall confidence score exceeds the preset confidence threshold and the spatiotemporal dynamic evolution feature includes the velocity distribution feature derived from the smoldering physical diffusion model, the candidate hot spot region is determined to be a smoldering hazard.

[0048] For example, the implementation scenario is a circular silo storing bituminous coal, with a diameter of 30 meters and a wall height of 25 meters. Based on the silo's geometry and the selected infrared thermal imager parameters, 16 infrared thermal imagers were deployed on the top ring beam of the silo wall at an azimuth interval of 22.5 degrees and a pitch angle of 15 degrees to ensure no blind spots on the material surface. The 16 infrared thermal imagers were triggered by a hardware synchronization signal, synchronously acquiring thermal images at a frequency of one frame per minute. The image resolution was 640×512 pixels, and the temperature measurement range was -20 degrees Celsius to 550 degrees Celsius. The acquired 16 image sequences were transmitted to an industrial control computer via gigabit Ethernet. A stitching algorithm based on SIFT feature matching and bilinear interpolation was used to synthesize a panoramic thermal image sequence in real time. Subsequently, a multi-scale Retinex enhancement algorithm was used for processing; for example, three Gaussian surround scales were used (the scale parameters were set to...). , , (Pixels) to enhance local contrast and detail visibility in the image. From the enhanced image, based on pre-divided grid regions and their respective calculated adaptive thresholds, temperature anomaly regions are extracted as candidate hotspot regions through pixel-by-pixel thresholding and connected component analysis. Cross-frame matching and tracking are performed on the same hotspot region in a continuous image sequence. For continuously tracked candidate hotspot regions, their spatial morphological features and spatiotemporal dynamic evolution features are calculated, and a comprehensive confidence score is obtained by linear weighted fusion. For example, the feature element representing the degree of drastic change in local grayscale is calculated for a certain hotspot region. The eigenvalue is 0.85, representing the fractal dimension of the contour. The eigenvalue is 1.28, representing the rate of change of the equivalent radius. The boundary expansion velocity eigenvalue is 0.02 per hour. for The weighted fusion yields the overall confidence score. It is 1.1665. The value exceeds the preset confidence threshold (in this example, the threshold is set to 0.75 based on historical data validation), and the average boundary expansion velocity of the hot spot region is stable within the empirical range of smoldering diffusion velocity for bituminous coal, determined based on extensive experimental observations. Inside. This candidate hot spot area is determined to be a smoldering hazard, and an alarm is triggered.

[0049] Preferably, the method for deploying an infrared thermal imager array on the top of the inner wall of the silo includes:

[0050] Obtain the structural parameters of the target silo, including the silo's inner diameter and wall height; obtain the device parameters of the infrared thermal imager, including the horizontal and vertical field of view; and calculate the minimum number of infrared thermal imagers required to achieve 360-degree surround coverage based on the silo's inner diameter and horizontal field of view. The minimum number of infrared thermal imagers is calculated based on the silo's inner diameter, wall height, and vertical field of view. The required initial pitch angle.

[0051] Based on the structural parameters and device parameters, the minimum number of infrared thermal imagers is determined. The system checks whether a coverage blind spot exists at the initial elevation angle. If a coverage blind spot exists, the number of infrared thermal imagers is increased iteratively, and the elevation angle is recalculated until the coverage blind spot is eliminated, thus obtaining the actual required number of infrared thermal imagers. and optimize pitch angle .

[0052] Based on the number of infrared thermal imagers Calculations yielded A deployment azimuth sequence of infrared thermal imagers; based on the deployment azimuth sequence and optimized elevation angles... An array of infrared thermal imagers is deployed on the top of the inner wall of the silo.

[0053] For example, with an inner diameter 30 meters, warehouse wall height Taking a 25-meter silo as an example, the horizontal field of view is selected. for Vertical field of view for The infrared thermal imager. First, calculate the minimum number of infrared thermal imagers required to achieve 360-degree surround coverage. The calculation formula is: Substituting the data yields Platform. Then calculate the initial pitch angle. Through geometric relationships We obtain, among which Let the installation height of the infrared thermal imager be 0.5 meters. Substituting this into the equation, we get... Next, the coverage blind zone is determined. The initial parameters are calculated using a 3D geometric projection model. tower, The coverage of the material surface under this configuration is a specific implementation method. Calculation results show that there are blind spots within a certain range (e.g., an area ratio of approximately 5.7%) under this configuration. The core of this invention lies in eliminating all blind spots through iterative optimization (such as increasing the number of devices and adjusting the pitch angle), rather than being specific to a particular geometric projection algorithm. Any well-known method in the art capable of calculating and verifying blind spots can be used for this step. Iterative optimization is performed to this end: the number of infrared thermal imagers is increased to 16. First, the calculation of the new number... The initial pitch angle is [value]. The new azimuth interval is [value]. Based on the geometric relationships, recalculate the initial pitch angle. Subsequently, to completely eliminate the blind spot, a compensation angle was introduced. Fine-tuning is performed. The compensation angle... The method is determined through algorithm optimization. For example, a one-dimensional golden section search method is used, with the goal of eliminating the blind area. Within the scope of A rapid optimization is performed, eventually converging to Substitute into the formula The optimized pitch angle is obtained. Re-verification showed that the blind spot had been eliminated. The final number of infrared thermal imagers was determined. Platform, optimize pitch angle Calculate the deployment azimuth sequence based on the quantity N: azimuth ,in ,get Azimuth sequence with intervals Finally, according to this azimuth sequence and The infrared thermal imager array is deployed at the top of the inner wall of the silo.

[0054] Preferably, the method for extracting candidate hot spot regions from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence includes:

[0055] Based on the deployment azimuth sequence and optimized pitch angle, the silo material surface is divided into... A non-uniform grid region; for the For each grid region within the grid area, temperature data is statistically obtained based on a panoramic thermal image sequence within a preset historical time period; a reference temperature value is then calculated for each grid region based on the corresponding temperature data. and temperature fluctuation statistics ,in ; Obtain a frame of enhanced thermal image at the current moment; Calculate the real-time average temperature of each grid region in the frame of enhanced thermal image. According to the reference temperature value Temperature fluctuation statistics and real-time average temperature The adaptive segmentation thresholds for M grid regions were calculated. The adaptive segmentation threshold The calculation formula is:

[0056] .

[0057] in, and The weighting coefficients are preset; according to the above... Adaptive segmentation threshold for each grid region A bilinear interpolation algorithm is used to obtain a local adaptive threshold segmentation map with the same size as the enhanced thermal image frame. The enhanced thermal image frame and the local adaptive threshold segmentation map are compared pixel by pixel to obtain a binary image. If the temperature value of a pixel is higher than the threshold at the corresponding position in the threshold segmentation map, it is marked as a foreground pixel; otherwise, it is marked as a background pixel. Connected component analysis is performed on the binary image to obtain the connected regions of all foreground pixels. The bounding rectangle of each connected region is defined as a candidate hot spot region. The candidate hot spot regions are extracted from each enhanced thermal image frame in the enhanced thermal image sequence to obtain a candidate hot spot region image sequence corresponding to the enhanced thermal image sequence.

[0058] For example, based on the deployment azimuth and 52-degree elevation angle of 16 infrared thermal imagers, and combined with the internal parameters of the infrared thermal imagers, the surface of the silo material was divided into 256 non-uniform grid regions through inverse projection transformation. Taking a grid located 15 meters from the center of the silo wall as an example, a sequence of panoramic thermal images collected over the past 24 hours was selected, and the historical temperature data of each grid region was statistically analyzed to calculate the baseline temperature value for each grid. The temperature was 42.3 degrees Celsius, with a standard deviation of 42.3 degrees Celsius. The temperature is 2.1 degrees Celsius. After acquiring the enhanced thermal image of the current frame, the real-time average temperature of the grid is calculated. The temperature is 49.8 degrees Celsius. The adaptive threshold for this mesh is calculated using the adaptive segmentation threshold calculation formula. Celsius, where the weighting coefficients have been experimentally verified and are taken as follows: and This combination achieves a good balance between sensitivity and anti-interference. After calculating the adaptive threshold for all 256 grid cells, a local adaptive threshold segmentation map with the exact same size as the enhanced thermal image is generated through bilinear interpolation. The enhanced thermal image is compared pixel by pixel with the threshold segmentation map; pixels with temperature values ​​higher than the local threshold are marked as 1, and those lower than the threshold are marked as 0, thus obtaining a binary image. Eight-connected component analysis is performed on this binary image to identify all connected regions, and the minimum bounding rectangle of each connected region is defined as the candidate hot spot region, thereby completing the extraction of the candidate hot spot region image.

[0059] Preferably, the silo material surface is divided into sections based on the deployment azimuth sequence and optimized pitch angle. Methods for handling non-uniform grid regions include:

[0060] The internal parameters of the infrared thermal imager, including focal length and pixel size, are obtained. Based on the deployment azimuth angle, optimized elevation angle, and internal parameters, and according to the pinhole camera model and known camera installation geometry, the actual projected size of each pixel on the image plane of the infrared thermal imager on the silo material surface is calculated through inverse projection transformation. A spatial resolution distribution map of the silo material surface is determined based on the actual projected size of each pixel. Based on the spatial resolution distribution map, areas with spatial resolution greater than a preset resolution threshold are divided using a fine-grained grid, while areas with spatial resolution less than or equal to the preset resolution threshold are divided using a coarse-grained grid. A non-uniform grid region.

[0061] For example, to address the issue of uneven spatial resolution across different areas of the silo material surface caused by perspective projection in infrared thermal imagers, this method performs the following steps: using the focal length as... Pixel size is Taking infrared thermal imagers as an example, combined with Optimize pitch angle and The azimuth-interval deployment geometry is used to calculate the actual projected size of the pixels on the material surface through inverse projection transformation. Calculations show that each pixel in the area directly below the silo center corresponds to approximately... The actual size is square, while each cell in the edge area near the warehouse wall corresponds to approximately The actual size of the square image shows significant differences in spatial resolution. Based on this spatial resolution distribution map, a resolution threshold is set. This resolution threshold is determined according to the minimum hotspot size the system needs to detect and the imaging performance of the infrared thermal imager, requiring a trade-off between target recognition accuracy and computational efficiency. For example, it is set to meet the system's requirement for the smallest hotspot area to be identified (e.g., the size of the hotspot region to be identified). Based on the detection requirements, it is set to correspond to each pixel. Actual size. For the central region with a resolution higher than this threshold (i.e., the actual size per pixel). A fine-grained grid is used, with a grid size of 4×4 pixels; for lower resolution edge regions (i.e., each pixel corresponds to an actual size of...),... Coarse-grained mesh generation is used, with a mesh size of [value missing]. Pixel.

[0062] Preferably, the method for calculating the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hotspot region based on a single frame of the candidate hotspot region image includes:

[0063] Based on the single-frame candidate hotspot region image, the gray-level gradient distribution is obtained through Gaussian filtering and second-order derivative calculation; based on the gray-level gradient distribution, the feature elements of the degree of drastic local gray-level changes are calculated. Based on the single-frame candidate hotspot region image, a contour point set of the candidate hotspot region is obtained through a contour extraction algorithm; based on the contour point set, the contour fractal dimension feature elements are calculated using the box counting method. The feature element representing the degree of drastic change in local grayscale and contour fractal dimension eigenvalues The spatial feature vectors are obtained by combining them.

[0064] Based on the continuous multi-frame candidate hot spot region image sequence, an area change sequence is obtained by calculating the area of ​​the candidate hot spot region in each frame; the equivalent radius change rate feature element is then calculated based on the area change sequence. Based on the sequence of candidate hot spot region images in multiple consecutive frames, a boundary movement velocity sequence is obtained through cross-frame contour point matching and average displacement calculation; based on the boundary movement velocity sequence, median filtering is applied to obtain boundary outward expansion velocity feature elements. The equivalent radius change rate feature element and boundary expansion velocity characteristic element The temporal evolution feature vector is obtained by combining the features.

[0065] For example, Gaussian filtering is applied to a frame of candidate hotspot region image (using... Pixel window, standard deviation The gray-level gradient distribution is obtained by performing second-order derivative operations (using the Laplacian operator), and then the standard deviation of the gradient amplitude within the hotspot region is calculated. Based on this, the characteristic element of the drastic local gray-level changes is obtained. Its value is 0.85. The contour point set of the hot spot region is obtained using a contour extraction algorithm (such as the Suzuki85 algorithm), and box counting (using a box-scale sequence) is employed. The fractal dimension feature element of the contour is calculated from the pixel. Its value is 1.28. and Combining to form spatial feature vectors Based on a sequence of ten consecutive images of the hot spot region, the area of ​​each frame is calculated and converted into an equivalent radius. The equivalent radius change sequence is obtained, and then the equivalent radius change rate characteristic element is calculated by linearly fitting the change of this sequence over time. The value is 0.02 per hour. The boundary movement velocity sequence is obtained through cross-frame contour point matching and displacement calculation. After median filtering with a window size of 5, the boundary expansion velocity feature elements are obtained. for .Will and Combining to form temporal evolution feature vectors .

[0066] Preferably, the method for calculating the fractal dimension eigenvalues ​​of the contour based on the contour point set using box counting includes:

[0067] The contour point set is fitted with a straight line according to a preset length window to obtain the direction variance and fitting residual of each window; when the direction variance is not greater than a preset variance threshold and the fitting residual is not greater than a preset residual threshold, the corresponding window is marked as a regular straight line segment; the regular straight line segments are removed from the contour point set to obtain a cleaned contour point set; the cleaned contour point set is covered with a series of boxes of preset size, and the coverage number sequence is obtained; based on the coverage number sequence, the fractal dimension of the cleaned contour point set is calculated by linear fitting, and used as the contour fractal dimension feature element.

[0068] For example, hot spot region contours often contain regular straight line segments generated by material accumulation or silo structure edges. Directly using box counting to calculate the fractal dimension will lead to estimation bias due to these regular segments. A linear fit is performed on the contour point set with a preset window length of 10 pixels. The variance and fitting residual of each window direction are calculated, and based on typical contour noise levels, the variance threshold is set to 0.15 and the residual threshold is set to 1.2 pixels. Windows that simultaneously satisfy a variance no greater than 0.15 and a residual no greater than 1.2 pixels are marked as regular straight line segments and removed from the original contour point set to obtain a cleaned contour point set. The preset variance threshold and preset residual threshold are empirical values ​​set based on statistical analysis of a large number of hot spot contours and structural edge contours on silo material surfaces. They are used to effectively distinguish between irregular contours characterizing smoldering diffusion and regular geometric contours caused by material accumulation or silo wall structures. Subsequently, box counting is used with a series of preset sizes (e.g., The pixels are used to cover the cleaned contour point set to obtain a cover number sequence. The relationship between the coverage number and the box size was analyzed, and a linear fit was performed using the least squares method. The final calculated fractal dimension value was approximately 1.28, which was used as the feature element of the contour fractal dimension.

[0069] Preferably, the method for obtaining the boundary movement velocity sequence by cross-frame contour point matching and average displacement calculation based on the continuous multi-frame candidate hotspot region image sequence includes:

[0070] Obtain the hot spot region contour point set of the previous frame from the consecutive multi-frame candidate hot spot region image sequence. With the hot spot region contour point set of the current frame Based on the hot spot region contour point set of the current frame Calculate the set of outward normal unit vectors for each contour point in the current frame; for the set of contour points of the hot spot region in the current frame... Each point in the hot spot region contour point set of the previous frame A local nearest neighbor search is performed to obtain a set of candidate corresponding points. The set of candidate corresponding points is then filtered according to preset maximum allowable displacement constraints and maximum direction deviation constraints to obtain a set of valid corresponding points. A set of normal displacements is calculated based on the set of valid corresponding points and the set of outward normal unit vectors. The median is calculated based on the set of normal displacements. Finally, the instantaneous boundary spread velocity is calculated based on the median and the inter-frame time interval. ; Expand the instantaneous boundary outward velocity Write the velocity time series; perform median filtering on the velocity time series to obtain the boundary movement velocity series.

[0071] For example, the hot spot region contour is susceptible to noise interference and local deformation during cross-frame matching, leading to outliers in the displacement vector calculation. This method first obtains the hot spot region contour point sets of two adjacent frames, where the hot spot region contour point set of the current frame is... Contains 215 points, the hotspot region outline point set from the previous frame. It contains 209 points. Calculation The set of outward normal unit vectors for each point in the image. The search range is determined based on the maximum possible velocity of smoldering diffusion, the ground sampling distance of the image, and the inter-frame time interval. The maximum possible diffusion velocity of bituminous coal obtained from the experiment is... The ground sampling distance of the image in this example is approximately... The maximum possible pixel displacement, calculated per pixel and over a 60-second inter-frame time interval, is approximately 0.00692 pixels. To accommodate instantaneous fluctuations, matching uncertainties, and noise, the radius of the local search is conservatively set to 1 pixel. The hotspot region contour point set for the current frame is then analyzed. Each point in the hot spot region outline point set in the previous frame A local nearest neighbor search is performed to initially obtain candidate corresponding points. Subsequently, filtering is performed based on preset maximum allowable displacement constraints and maximum directional deviation constraints. The preset maximum allowable displacement constraint is calculated and set based on the upper limit of the empirical range of smoldering diffusion rate described in this invention and the image ground sampling distance (GSD), and is set to 5 pixels in this example. The maximum directional deviation constraint is set based on typical noise and deformation levels. After screening, the proportion of valid corresponding points is approximately 89%. Based on the set of valid corresponding points and the outward normal vector, the normal displacement of each point is calculated to obtain the set of normal displacements. Calculate the median of this set. The value is 2.3 pixels. The inter-frame time interval is known. The instantaneous boundary expansion velocity was calculated over a period of 60 seconds. Pixels / second. The current velocity value is written into the velocity time series, and median filtering is applied to the velocity time series of multiple consecutive frames (e.g., 10 frames). The window size is used to strike a balance between smoothing noise and maintaining response speed, ultimately resulting in a stable boundary motion velocity sequence, the latest value of which is... .

[0072] Preferably, the method for weighted fusion of the spatial feature vector and the temporal evolution feature vector to obtain the comprehensive confidence score includes:

[0073] The spatial feature vector and the temporal evolution feature vector are concatenated sequentially to obtain a fused feature vector; according to preset weight coefficients... , , , The integrated confidence score is obtained by linearly weighting the fused feature vectors. .

[0074] For example, the weighting coefficient , , , The values ​​were determined through a large amount of labeled smoldering and non-smoldering hot spot sample data, trained using a classification algorithm (such as Support Vector Machine, SVM) or optimized using a grid search. A set of validated values ​​were... , , , The principle for determining this weight combination is: to assign dynamic evolutionary characteristics ( ) and spatial morphological characteristics ( Equally important, to align with the physical nature of smoldering hazards, which possess both abnormal spatial morphology and continuous diffusion over time; simultaneously, within the dynamic characteristics, a boundary expansion velocity is assigned ( ) Rate of change of equivalent radius ( Higher weights are assigned to more sensitively respond to boundary movements. This example can be fine-tuned based on validation results for specific application scenarios. For the example hotspot region, the feature element values ​​are: the degree of drastic local grayscale change. fractal dimension of the outline Equivalent radius change rate Boundary expansion speed Before substituting into the formula, each characteristic element needs to be normalized to eliminate differences in dimensions and orders of magnitude. Then, substituting into the formula for linear weighted calculation: .

[0075] Preferably, the spatiotemporal dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model, including:

[0076] Obtain the boundary motion velocity sequence of consecutive frames; perform smoothing filtering on the boundary motion velocity sequence to obtain a stable velocity sequence; when the stable velocities in the stable velocity sequence are all within a preset theoretical velocity range... If the value is within a certain range, it is determined that the speed distribution characteristics are met.

[0077] For example, the preset theoretical speed range The value is determined based on statistical analysis of historical smoldering case data, observation of smoldering experiments under controlled laboratory conditions, and data from publicly available industry standards or academic literature. For common materials, empirical values ​​can be used. For example, for bituminous coal in the examples, the value is taken as [value missing]. This empirical range is derived from statistical analysis of historical smoldering case data, effectively covering the typical diffusion rate of bituminous coal smoldering and distinguishing it from non-smoldering heat changes such as mechanical heat sources (usually faster or zero) and solar thermal interference (usually without sustained outward expansion velocity). For other materials, such as grains or biomass pellets, different empirical ranges are used. These ranges constitute a preset material-velocity range mapping table, which can be loaded from the built-in database based on the material type selected by the user during system initialization. The system obtains a sequence of boundary outward expansion velocities for multiple consecutive frames (e.g., 10 frames) of the hotspot region. Its original value is To eliminate the influence of outliers (such as 10.2 in the sequence), the sequence is smoothed using a median filter (exemplarily, a median filter with a window length of 3) to obtain a stable velocity sequence. for The stable velocity sequence has been determined. All velocity values ​​are within the theoretical velocity range preset for bituminous coal. Therefore, it is determined that the spatiotemporal dynamic evolution characteristics of the hot spot region conform to the velocity distribution characteristics.

[0078] Example 2: Based on the same inventive concept, such as Figure 2 As shown, this embodiment also provides a fire hazard identification system based on image enhancement, the system comprising:

[0079] An imaging module is used to deploy an infrared thermal imager array on the top of the inner wall of the silo, the infrared thermal imager array comprising: One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence.

[0080] The characterization module is used to perform cross-frame tracking and matching of the same candidate hot spot region in the candidate hot spot region image sequence; calculate the spatial morphological features and spatiotemporal dynamic evolution features of the tracked candidate hot spot region based on a single frame of the candidate hot spot region image; and perform weighted fusion of the spatial morphological features and spatiotemporal dynamic evolution features to obtain a comprehensive confidence score.

[0081] The determination module is used to determine the candidate hot spot region as a smoldering hazard when the comprehensive confidence score exceeds a preset confidence threshold and the spatiotemporal dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model.

[0082] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0083] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fire hazard identification method based on image enhancement, characterized in that, The method includes: An infrared thermal imager array is deployed on the top of the inner wall of the silo. The infrared thermal imager array includes... One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence. Cross-frame tracking and matching are performed on the same candidate hot spot region in the candidate hot spot region image sequence; based on the single frame of the candidate hot spot region image, the spatial morphological features and temporal dynamic evolution features of the tracked candidate hot spot region are calculated; the spatial morphological features and temporal dynamic evolution features are weighted and fused to obtain a comprehensive confidence score; When the overall confidence score exceeds the preset confidence threshold and the time-series dynamic evolution features include velocity distribution features derived from the smoldering physical diffusion model, the candidate hot spot region is determined to be a smoldering hazard. The method for calculating the spatial morphological features and temporal dynamic evolution features of the tracked candidate hotspot region based on a single frame of the candidate hotspot region image includes: Based on the candidate hotspot region image in a single frame, the gray-level gradient distribution is obtained through Gaussian filtering and second-order derivative calculation; based on the gray-level gradient distribution, the feature element representing the degree of drastic local gray-level changes is calculated. Based on a single frame of the candidate hot spot region image, a contour point set of the candidate hot spot region is obtained through a contour extraction algorithm; based on the contour point set, the contour fractal dimension feature elements are calculated using the box counting method. The feature element representing the degree of drastic change in local grayscale and contour fractal dimension eigenvalues Combining them yields spatial feature vectors; Based on the candidate hot spot region image sequence, an area change sequence is obtained by calculating the area of ​​the candidate hot spot region in each frame; the equivalent radius change rate feature element is then calculated based on the area change sequence. Based on the candidate hotspot region image sequence, a boundary movement velocity sequence is obtained through cross-frame contour point matching and average displacement calculation; based on the boundary movement velocity sequence, median filtering is applied to obtain boundary outward expansion velocity feature elements. The equivalent radius change rate feature element and boundary expansion velocity characteristic element The temporal evolution feature vector is obtained by combining the features. The method for obtaining the boundary movement velocity sequence based on the candidate hot spot region image sequence through cross-frame contour point matching and average displacement calculation includes: Obtain the hot spot region contour point set from the previous frame from the candidate hot spot region image sequence. With the hot spot region contour point set of the current frame Based on the hot spot region contour point set of the current frame Calculate the set of outward normal unit vectors for each contour point in the current frame; for the set of contour points of the hot spot region in the current frame... Each point in the hot spot region contour point set of the previous frame A local nearest neighbor search is performed to obtain a set of candidate corresponding points. The set of candidate corresponding points is then filtered according to preset maximum allowable displacement constraints and maximum direction deviation constraints to obtain a set of valid corresponding points. A set of normal displacements is calculated based on the set of valid corresponding points and the set of outward normal unit vectors. The median is calculated based on the set of normal displacements. Finally, the instantaneous boundary spread velocity is calculated based on the median and the inter-frame time interval. ; Expand the instantaneous boundary outward velocity Write the velocity time series; perform median filtering on the velocity time series to obtain the boundary movement velocity series.

2. The fire hazard identification method based on image enhancement according to claim 1, characterized in that, The method for deploying an infrared thermal imager array on the top of the inner wall of the silo includes: Obtain the structural parameters of the target silo, including the silo's inner diameter and wall height; obtain the device parameters of the infrared thermal imager, including the horizontal and vertical field of view; and calculate the minimum number of infrared thermal imagers required to achieve 360-degree surround coverage based on the silo's inner diameter and horizontal field of view. The minimum number of infrared thermal imagers is calculated based on the silo's inner diameter, wall height, and vertical field of view. The required initial pitch angle; Based on the structural parameters and device parameters, the minimum number of infrared thermal imagers is determined. The system checks whether a coverage blind spot exists at the initial elevation angle. If a coverage blind spot exists, the number of infrared thermal imagers is increased iteratively, and the elevation angle is recalculated until the coverage blind spot is eliminated, thus obtaining the actual required number of infrared thermal imagers. and optimize pitch angle ; Based on the number of infrared thermal imagers Calculations yielded A deployment azimuth sequence of infrared thermal imagers; based on the deployment azimuth sequence and optimized elevation angles... An array of infrared thermal imagers is deployed on the top of the inner wall of the silo.

3. The fire hazard identification method based on image enhancement according to claim 2, characterized in that, The method for extracting candidate hot spot regions from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence includes: Based on the deployment azimuth sequence and optimized pitch angle, the silo material surface is divided into... A non-uniform grid region; for the For each grid region within the grid area, temperature data is statistically obtained based on a panoramic thermal image sequence within a preset historical time period; a reference temperature value is then calculated for each grid region based on the corresponding temperature data. and temperature fluctuation statistics ,in ; Obtain a frame of enhanced thermal image at the current moment; Calculate the real-time average temperature of each grid region in the frame of enhanced thermal image. According to the reference temperature value Temperature fluctuation statistics and real-time average temperature The adaptive segmentation thresholds for M grid regions were calculated. The adaptive segmentation threshold The calculation formula is: ； in, and The weighting coefficients are preset; according to the above... Adaptive segmentation threshold for each grid region A bilinear interpolation algorithm is used to obtain a local adaptive threshold segmentation map with the same size as the enhanced thermal image frame. The enhanced thermal image frame and the local adaptive threshold segmentation map are compared pixel by pixel to obtain a binary image. If the temperature value of a pixel is higher than the threshold at the corresponding position in the threshold segmentation map, it is marked as a foreground pixel; otherwise, it is marked as a background pixel. Connected component analysis is performed on the binary image to obtain the connected regions of all foreground pixels. The bounding rectangle of each connected region is defined as a candidate hot spot region. The candidate hot spot regions are extracted from each enhanced thermal image frame in the enhanced thermal image sequence to obtain a candidate hot spot region image sequence corresponding to the enhanced thermal image sequence.

4. The fire hazard identification method based on image enhancement according to claim 3, characterized in that, The material surface of the silo is divided into sections based on the deployment azimuth sequence and optimized pitch angle. Methods for handling non-uniform grid regions include: The internal parameters of the infrared thermal imager, including focal length and pixel size, are acquired. Based on the deployment azimuth angle, optimized elevation angle, and internal parameters, the actual projected size of each pixel on the image plane of the infrared thermal imager on the surface of the silo material is calculated through inverse projection transformation. A spatial resolution distribution map of the silo material surface is determined based on the actual projected size of each pixel. According to the spatial resolution distribution map, areas with spatial resolution greater than a preset resolution threshold are divided using a fine-grained grid, while areas with spatial resolution less than or equal to the preset resolution threshold are divided using a coarse-grained grid. A non-uniform grid region.

5. The fire hazard identification method based on image enhancement according to claim 1, characterized in that, The method for calculating the fractal dimension eigenvalues ​​of the contour based on the contour point set using box counting includes: The contour point set is fitted with a straight line according to a preset length window to obtain the direction variance and fitting residual of each window; when the direction variance is not greater than a preset variance threshold and the fitting residual is not greater than a preset residual threshold, the corresponding window is marked as a regular straight line segment; the regular straight line segments are removed from the contour point set to obtain a cleaned contour point set; the cleaned contour point set is covered with a series of boxes of preset size, and the coverage number sequence is obtained; based on the coverage number sequence, the fractal dimension of the cleaned contour point set is calculated by linear fitting, and used as the contour fractal dimension feature element.

6. The fire hazard identification method based on image enhancement according to claim 1, characterized in that, The method for weighted fusion of the spatial feature vector and the temporal evolution feature vector to obtain a comprehensive confidence score includes: The spatial feature vector and the temporal evolution feature vector are concatenated sequentially to obtain a fused feature vector; according to preset weight coefficients... , , , The integrated confidence score is obtained by linearly weighting the fused feature vectors. .

7. The fire hazard identification method based on image enhancement according to claim 1, characterized in that, The time-series dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model, including: Obtain the boundary motion velocity sequence of consecutive frames; perform smoothing filtering on the boundary motion velocity sequence to obtain a stable velocity sequence; when the stable velocities in the stable velocity sequence are all within a preset theoretical velocity range... If the value is within a certain range, it is determined that the speed distribution characteristics are met.

8. A fire hazard identification system based on image enhancement, characterized in that, The system includes: An imaging module is used to deploy an infrared thermal imager array on the top of the inner wall of the silo, the infrared thermal imager array comprising: One infrared thermal imager; the infrared thermal imager array synchronously acquires data about the interior of the silo. A thermal image subsequence; for the The thermal image subsequences are spatiotemporally registered and stitched together to obtain a panoramic thermal image sequence covering the entire surface of the silo material; the panoramic thermal image sequence is then subjected to image enhancement processing to obtain an enhanced thermal image sequence; candidate hot spot regions are extracted from the enhanced thermal image sequence to obtain a candidate hot spot region image sequence. The characterization module is used to perform cross-frame tracking and matching of the same candidate hot spot region in the candidate hot spot region image sequence; calculate the spatial morphological features and temporal dynamic evolution features of the tracked candidate hot spot region based on a single frame of the candidate hot spot region image; and perform weighted fusion of the spatial morphological features and temporal dynamic evolution features to obtain a comprehensive confidence score. The determination module is used to determine the candidate hot spot region as a smoldering hazard when the comprehensive confidence score exceeds a preset confidence threshold and the time-series dynamic evolution characteristics include velocity distribution characteristics derived from the smoldering physical diffusion model. The method for calculating the spatial morphological features and temporal dynamic evolution features of the tracked candidate hotspot region based on a single frame of the candidate hotspot region image includes: Based on the candidate hotspot region image in a single frame, the gray-level gradient distribution is obtained through Gaussian filtering and second-order derivative calculation; based on the gray-level gradient distribution, the feature element representing the degree of drastic local gray-level changes is calculated. Based on a single frame of the candidate hot spot region image, a contour point set of the candidate hot spot region is obtained through a contour extraction algorithm; based on the contour point set, the contour fractal dimension feature elements are calculated using the box counting method. The feature element representing the degree of drastic change in local grayscale and contour fractal dimension eigenvalues Combining them yields spatial feature vectors; Based on the candidate hot spot region image sequence, an area change sequence is obtained by calculating the area of ​​the candidate hot spot region in each frame; the equivalent radius change rate feature element is then calculated based on the area change sequence. Based on the candidate hotspot region image sequence, a boundary movement velocity sequence is obtained through cross-frame contour point matching and average displacement calculation; based on the boundary movement velocity sequence, median filtering is applied to obtain boundary outward expansion velocity feature elements. The equivalent radius change rate feature element and boundary expansion velocity characteristic element The temporal evolution feature vector is obtained by combining the features. The method for obtaining the boundary movement velocity sequence based on the candidate hot spot region image sequence through cross-frame contour point matching and average displacement calculation includes: Obtain the hot spot region contour point set from the previous frame from the candidate hot spot region image sequence. With the hot spot region contour point set of the current frame Based on the hot spot region contour point set of the current frame Calculate the set of outward normal unit vectors for each contour point in the current frame; for the set of contour points of the hot spot region in the current frame... Each point in the hot spot region contour point set of the previous frame A local nearest neighbor search is performed to obtain a set of candidate corresponding points. The set of candidate corresponding points is then filtered according to preset maximum allowable displacement constraints and maximum direction deviation constraints to obtain a set of valid corresponding points. A set of normal displacements is calculated based on the set of valid corresponding points and the set of outward normal unit vectors. The median is calculated based on the set of normal displacements. Finally, the instantaneous boundary spread velocity is calculated based on the median and the inter-frame time interval. ; Expand the instantaneous boundary outward velocity Write the velocity time series; perform median filtering on the velocity time series to obtain the boundary movement velocity series.

Images