An infrared image denoising method and system for gas leakage detection

Abstract

The application relates to the technical field of gas leakage detection, and discloses an infrared image denoising method and system for gas leakage detection, which comprises the following steps: acquiring an original infrared image containing noise; dividing the image into overlapping image blocks, searching for similar blocks and combining the similar blocks into a three-dimensional image block group; performing three-dimensional transformation on each three-dimensional group, inversely transforming back to the spatial domain after filtering and denoising in the transformation domain, generating a preliminary denoised image through weighted averaging, marking the image blocks in the same group as the same similar region, and forming initial region division; based on the division, independently performing contrast stretching on each similar region to enhance the contrast between the gas leakage region and the background, so as to determine the leakage position. The application organically fuses denoising and region division, efficiently suppresses noise, significantly improves the contrast of target features through region adaptive enhancement, and provides a high-quality image basis for subsequent accurate positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gas leakage detection, in particular to an infrared image denoising method and system for gas leakage detection. BACKGROUND

[0002] In the field of power equipment safety monitoring, sulfur hexafluoride (SF6) is widely used in gas insulated switchgear, circuit breakers, disconnectors and other high-voltage power equipment due to its excellent electrical insulation and good arc extinguishing performance, but its leakage will pose a serious threat to equipment safety and the environment. In the detection of gas leakage, infrared imaging technology plays a crucial role because it can visualize gas leakage non-contact. By capturing the thermal radiation difference between gas and background environment, infrared cameras can present colorless and odorless leakage gas in the form of "smoke" in the monitoring screen, greatly improving the efficiency and safety of inspection. However, the signal-to-noise ratio of infrared images in practical applications is generally low, and the inherent noise, non-uniform response and complex environmental interference seriously blur the boundary between the leakage gas and the background, making it difficult to accurately identify the weak leakage signal, directly affecting the reliability of early warning.

[0003] To solve the noise problem, the prior art usually uses classical image filtering algorithms such as Gaussian filtering or median filtering. Although these methods can smooth noise, they will inevitably blur the texture and edge details of the image while filtering out noise. For gas leakage detection, the edge features of the leakage area are exactly the key to distinguish it from the background. The loss of such details will cause the shape of the leakage gas cloud to become incomplete and the boundary to be blurred, and even cause missed detection or false detection in the subsequent image segmentation step. More importantly, the traditional denoising process and the subsequent target segmentation process are usually two independent steps; the denoising module does not provide any auxiliary information to the segmentation module that helps to distinguish "target" from "background". This fragmentation in the processing flow limits the further improvement of the overall system performance. SUMMARY

[0004] To this end, the technical problem to be solved by the present application is to overcome the defect that the denoising process and the target region segmentation process in the prior art are mutually fragmented, resulting in difficulty in effectively preserving and enhancing the feature details of the gas leakage area while suppressing noise. The present application provides an infrared image denoising method for gas leakage detection, which can organically integrate denoising and preliminary region division. On the basis of efficiently suppressing image noise, the region-adaptive enhancement strategy significantly improves the feature contrast between the gas leakage area and the background, thereby providing a high-quality, high-contrast image basis for subsequent accurate positioning of the leakage location.

[0005] To solve the above technical problems, the present application provides an infrared image denoising method for gas leakage detection, comprising the following steps:

[0006] Obtaining an original infrared image for capturing potential thermal radiation differences of gas leakage, to obtain initial image data containing noise;

[0007] Dividing the original infrared image into a plurality of overlapping image blocks, searching for similar area blocks, combining all similar image blocks into a three-dimensional image block group, to obtain a plurality of three-dimensional image block groups;

[0008] Performing three-dimensional transformation on each three-dimensional image block group to obtain transformation domain coefficients, and performing filtering processing on the transformation domain coefficients to suppress noise in the initial image data;

[0009] Performing three-dimensional inverse transformation on the filtered transformation domain coefficients to obtain a denoised three-dimensional image block group, and returning each image block in the denoised three-dimensional image block group to its original image position, and generating a preliminary denoised image by weighted average aggregation of all image blocks; wherein the image blocks belonging to the same three-dimensional image block group are marked as the same similar area, and the collection of all similar areas constitutes the initial area division of the preliminary denoised image;

[0010] Based on the initial area division, for each similar area, according to the gray scale distribution of the internal pixels, independently performing contrast stretching transformation to enhance the feature contrast of the gas leakage area and the background area, and determining the gas leakage position.

[0011] In an embodiment of the present application, the searching for similar area blocks comprises the following steps:

[0012] Taking one of the current image blocks as the core, calculating the feature vector distance between all candidate image blocks and the current image block within its preset search neighborhood;

[0013] According to the feature vector distance, the candidate image blocks are divided into a plurality of temporary clusters by dynamic clustering;

[0014] Selecting all candidate image blocks belonging to the same temporary cluster as the current image block to determine as similar area blocks;

[0015] Wherein, the feature vector is composed of the gray scale statistical feature and the spatial gradient feature of the image block.

[0016] In an embodiment of the present application, combining all similar image blocks into a three-dimensional image block group comprises:

[0017] The current image block to be processed is established as a reference block, and in all the searched similar blocks, spatial offset compensation is performed on each similar block with the spatial coordinates of the reference block as a reference, so as to ensure that all the blocks are geometrically aligned in the same local coordinate system;

[0018] Linear normalization processing of the gray scale values in each image block, including the reference block, is independently performed, so as to eliminate overall brightness deviation caused by local light or temperature difference, and to make all the blocks at the same intensity level;

[0019] Each two-dimensional image block with the size of M*N pixels after the alignment and normalization processing is regarded as an independent two-dimensional plane, and the two-dimensional planes are stacked along a third spatial dimension perpendicular to the planes, and the stacking order of each stacked plane is determined according to the similarity degree of the plane to the reference block; finally, a three-dimensional data cube with the size of M*N*K is formed, wherein K is the total number of the stacked image blocks, and the cube is the three-dimensional image block group.

[0020] In an embodiment of the present application, three-dimensional transformation is performed on each three-dimensional image block group to obtain transformation domain coefficients, including:

[0021] A transformation sequence composed of three one-dimensional orthogonal transformation bases in sequence is preset, and the three transformation bases correspond to the height, width and depth dimensions of the three-dimensional data cube respectively;

[0022] A first one-dimensional orthogonal transformation is independently applied to each row of pixel sequences along the height direction of the three-dimensional data cube, so as to convert the pixel values into frequency spectrum coefficients in the dimension;

[0023] A second one-dimensional orthogonal transformation is independently applied to each column of pixel sequences along the width direction of the three-dimensional data cube, so as to convert the pixel values into frequency spectrum coefficients in the dimension;

[0024] A third one-dimensional orthogonal transformation is independently applied to each frequency spectrum coefficient sequence obtained after the first two-dimensional transformation along the depth direction of the three-dimensional data cube, i.e. the stacking direction of the image blocks, so as to convert each element in the three-dimensional data cube from a spatial pixel value into a three-dimensional transformation domain coefficient.

[0025] In an embodiment of the present application, when the transformation domain coefficients are filtered, the similarity degrees of the image blocks in the three-dimensional image block group are sorted, and the filtering strength is dynamically adjusted according to the similarity degrees of the image blocks in the three-dimensional image block group; for the three-dimensional image block group with a high similarity degree, stronger filtering strength is adopted; and for the three-dimensional image block group with a low similarity degree, weaker filtering strength is adopted.

[0026] In one embodiment of the present application, the similarity degree of each image block in a three-dimensional image block group is sorted, comprising:

[0027] For each three-dimensional image block group, a reference image block is generated by calculating the median value of the gray scale of all image blocks in the group at each pixel position;

[0028] The pixel-level absolute difference between each image block in the group and the reference image block is calculated respectively, and the sum of all pixel difference values of each image block is summed to obtain the cumulative difference degree of each image block relative to the reference block, and the cumulative difference degrees of all image blocks constitute a difference degree matrix;

[0029] The cumulative difference degrees in the difference degree matrix are arranged in ascending order of numerical value, and the lower the numerical value represents the more similar to the reference block, and according to the arrangement order, a similarity degree sorting sequence of each image block in the three-dimensional image block group is generated.

[0030] In one embodiment of the present application, a weighted average aggregation is performed on all image blocks to generate a preliminary denoising image, comprising:

[0031] For each image block processed by the three-dimensional inverse transform, a weight value determined based on its reconstruction quality is assigned;

[0032] According to the weight value, all pixel values in each image block are weighted to obtain a series of weighted image blocks;

[0033] All weighted image blocks are superimposed and fused according to their original image positions, and the values of each pixel position in the fused image are normalized based on the total weight, thereby generating the preliminary denoising image.

[0034] In one embodiment of the present application, the reconstruction quality is evaluated by calculating the core energy proportion retained in the transform domain before and after filtering, and the higher the core energy proportion, the greater the weight assigned.

[0035] In one embodiment of the present application, a contrast stretching transform is independently performed according to the gray scale distribution of the internal pixels, comprising:

[0036] Identifying the main range of the gray scale distribution of the similar region;

[0037] Linearly mapping the main range to all or a major part of the preset output gray scale interval, while suppressing the discrete noise pixel points distributed outside the main range to the dark end of the output gray scale interval.

[0038] To solve the above technical problems, the present application also provides an infrared image denoising system for gas leakage detection for implementing the above method, comprising:

[0039] an image acquisition module configured to acquire a raw infrared image capturing a potential thermal radiation difference of a gas leakage, to obtain initial image data containing noise;

[0040] a three-dimensional block group construction module configured to divide the raw infrared image into a plurality of overlapping image blocks, search for similar region blocks, and combine all the similar image blocks into a three-dimensional image block group, to obtain a plurality of three-dimensional image block groups;

[0041] a collaborative filtering module configured to perform a three-dimensional transform on each three-dimensional image block group, to obtain transform domain coefficients, and filter the transform domain coefficients to suppress the noise in the initial image data;

[0042] an image reconstruction module configured to perform a three-dimensional inverse transform on the filtered transform domain coefficients, to obtain a three-dimensional image block group after noise reduction, and return each image block in the three-dimensional image block group after noise reduction to its original image position, to generate a preliminary denoised image by performing a weighted average aggregation on all the image blocks; wherein the image blocks belonging to the same three-dimensional image block group are marked as the same similar region, and a collection of all the similar regions constitutes an initial region division of the preliminary denoised image;

[0043] a gas positioning module configured to, based on the initial region division, for each similar region, independently perform a contrast stretching transform according to the grayscale distribution of the pixels inside the similar region, to enhance the feature contrast between the gas leakage region and the background region, and determine the gas leakage position.

[0044] The above technical solutions of the present application have the following advantages over the prior art:

[0045] The infrared image denoising method for gas leakage detection according to the present application firstly, through three-dimensional collaborative filtering, removes noise while better preserving the detail structure and edge information of the gas leakage region. Secondly, the intermediate result of the denoising process, i.e., the similar region division feature, is used to guide the subsequent image enhancement, so that the entire processing flow forms an organic whole, solving the problem of the separation of denoising and segmentation in the traditional flow. Finally, the region-adaptive contrast stretching can accurately enhance the target region, effectively improving the contrast between the gas leakage region and the complex background, thereby providing reliable technical support for the fast and accurate positioning of the final leakage point. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the accompanying drawings, in which:

[0047] Figure 1is a step flow chart of the infrared image denoising method for gas leakage detection of the present application;

[0048] Figure 2 is a step flow chart of searching similar area blocks of the present application;

[0049] Figure 3 is a step flow chart of constructing multiple three-dimensional image block groups of the present application;

[0050] Figure 4 is a step flow chart of performing three-dimensional transformation on the three-dimensional image block groups to obtain transform domain coefficients of the present application;

[0051] Figure 5 is a step flow chart of sorting the similarity of each image block in the three-dimensional image block groups of the present application;

[0052] Figure 6 is a step flow chart of weighted average aggregation of image blocks of the present application;

[0053] Figure 7 is a structural framework diagram of the infrared image denoising system for gas leakage detection of the present application. DETAILED DESCRIPTION

[0054] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not limiting to the present application.

[0055] Referring to Figure 1 The present application provides an infrared image denoising method for gas leakage detection, which combines the denoising process with preliminary area division, thereby laying a solid foundation for subsequent accurate detection, including the following steps:

[0056] Obtaining an original infrared image to capture the potential thermal radiation difference of gas leakage, to obtain initial image data containing noise;

[0057] Dividing the original infrared image into multiple overlapping image blocks, searching for similar area blocks, combining all similar image blocks into a three-dimensional image block group, to obtain multiple three-dimensional image block groups;

[0058] The method starts with the acquisition of the original infrared image, and then enters its core processing stage: dividing the image into numerous overlapping small blocks, and aggregating these small blocks into several three-dimensional image block groups according to their similarity; this "three-dimensional grouping" operation itself is an adaptive clustering based on image content, which pre-classifies pixels with similar texture and grayscale.

[0059] performing a three-dimensional transformation on each of the three-dimensional image block groups to obtain transform domain coefficients, filtering the transform domain coefficients to suppress noise in the initial image data;

[0060] performing a three-dimensional inverse transformation on the filtered transform domain coefficients to obtain a denoised three-dimensional image block group, and returning each image block in the denoised three-dimensional image block group to its original image position, and generating a preliminary denoised image by performing weighted average aggregation on all the image blocks; wherein the image blocks belonging to the same three-dimensional image block group are marked as the same similar region, and a collection of all the similar regions constitutes an initial region division of the preliminary denoised image.

[0061] Next, the three-dimensional groups are collaboratively filtered in the transform domain, which can extremely efficiently separate noise from real signals, because noise has no correlation within the group, while real image structures (including potential leakage regions) exhibit high consistency in three-dimensional space; after inverse transformation and weighted average, the filtered data not only generates a high-quality preliminary denoised image, but more importantly, the method marks the image blocks belonging to the same three-dimensional image block group as the same similar region, and a collection of all these regions constitutes an initial region division map.

[0062] Based on the initial region division, for each similar region, a contrast stretching transformation is independently performed according to the gray scale distribution of the pixels inside it, to enhance the feature contrast of the gas leakage region and the background region, and determine the gas leakage position.

[0063] Through the initial region division, the image is intelligently preliminarily decomposed into multiple sub-regions with homogeneous characteristics while being denoised, and on this basis, a contrast stretching transformation is independently performed for each similar region based on this division. Since the gas leakage region usually forms a continuous region that is different from the surrounding background in thermal radiation characteristics, the gray scale distribution inside this region is relatively concentrated. Through this region-adaptive enhancement, the gray scale dynamic range of this specific region can be maximally stretched, making the difference between it and the surrounding background be significantly amplified, without excessively stretching other irrelevant regions to introduce new background noise. Ultimately, this targeted enhancement enables the position of the gas leakage to be clearly determined in the image.

[0064] Reference Figure 2As shown, when performing the step of searching similar region blocks, the embodiment aims to more accurately and adaptively identify truly similar image blocks through a structured feature matching and dynamic clustering mechanism, including the following steps: taking the current image block to be processed as the core, traversing within its preset search neighborhood, for each candidate image block in the neighborhood, instead of performing a simple pixel value comparison, a vector distance in a feature space composed of a gray scale statistical feature and a spatial gradient feature is calculated for the current block; this design overcomes the limitations of single gray scale comparison, the gray scale statistical feature reflects the brightness distribution characteristics of the block, while the spatial gradient feature describes its internal edges and texture structure, the combination of the two makes the judgment of image block similarity more comprehensive and robust, which can effectively distinguish regions with similar gray scales but different texture structures.

[0065] After obtaining the feature vector distances of all candidate blocks and the current block, instead of using a fixed threshold to arbitrarily divide similarity or not, a dynamic clustering process is introduced, which automatically divides the candidate image blocks in the search neighborhood into several temporary clusters according to the calculated feature vector distances. The advantage of this dynamic clustering is its adaptability, which can dynamically determine the division standard of "similarity" according to the actual distribution of image blocks in the current local region, thereby avoiding the recognition bias that may be caused by a global fixed threshold under different image regions or different imaging conditions.

[0066] After dynamic clustering, instead of considering all candidate blocks as similar blocks, a precise attribution judgment is performed: only those candidate blocks that are divided into the same temporary cluster as the current core image block are selected as the final truly similar region blocks; this mechanism means that similarity is relative to the current local context, only blocks that belong to the same class as the current core block in the local feature distribution pattern will be included in the similar set, which not only improves the internal consistency of the similar block set, but also provides a reliable guarantee for subsequent construction of high-quality three-dimensional image block groups.

[0067] Referring to Figure 3As shown, the embodiment combines all similar image blocks into a three-dimensional image block group for precise construction, the core of which is to construct a three-dimensional data unit highly consistent in geometric space and signal intensity from a group of two-dimensional image blocks preliminarily considered similar through a series of normalization preprocessing, thereby maximizing the efficiency of subsequent collaborative filtering, including the following steps: establishing the current image block to be processed as a geometric reference, and taking the spatial coordinates of the reference block as the unified reference system, fine spatial offset compensation is performed on all searched similar blocks; this geometric alignment step is crucial because it ensures that in the subsequent stacking, the pixels at the same depth position in the three-dimensional data body strictly correspond to the same physical point in the scene, and if this step is missing, even similar blocks will produce signal blur after stacking due to slight pixel-level misalignment, thereby severely weakening the potential of three-dimensional transformation in mining intra-group correlation.

[0068] After completing the unification of geometric space, linear normalization processing of the gray value within each aligned image block, including the reference block itself, is independently performed; this operation aims to eliminate overall brightness deviations introduced by non-uniform illumination, detector response differences or local scene temperature fluctuations, after normalization, all image blocks are placed on the same intensity level, and their gray distribution characteristics are determined only by their internal structure and texture, and are decoupled from absolute brightness values; this allows subsequent processing to focus on mining common structural information between blocks, rather than being misled by intensity differences, thereby more accurately separating real structural signals from random noise.

[0069] Finally, after completing the above-mentioned geometric and intensity normalization, these size-unified two-dimensional image blocks are sequentially stacked along the third dimension perpendicular to their plane, thereby constructing the final three-dimensional data cube. It is particularly worth noting that the stacking order is not arbitrarily arranged, but is sorted according to the similarity of each block to the reference block, this ordered stacking allows the highest quality signal and the most representative blocks to be concentrated at one end of the three-dimensional data cube, forming a signal quality gradient from the core to the edge; this regularity inherent in the data structure itself allows real and consistent signal energy to be more concentrated in a specific frequency band after subsequent three-dimensional transformation, while noise and inconsistent interference are isolated to other areas, which creates extremely favorable conditions for the next precise filtering operation.

[0070] Reference Figure 4As shown, the embodiment further discloses the specific process of performing three-dimensional transformation on each three-dimensional image block group to obtain transformed domain coefficients, including the following steps: presetting a transformation sequence composed of three one-dimensional orthogonal transformation bases in a specific order, the setting of this sequence is the logical blueprint of the whole transformation, which specifies the dimension order of processing three-dimensional data cubes, i.e. height, width, and finally depth; selecting separable orthogonal transformation bases, the core advantage of which is to decompose a complex three-dimensional transformation problem into three consecutive one-dimensional transformations, thereby reducing the computational complexity to an engineering implementation level while ensuring mathematical rigor and signal processing effect.

[0071] The transformation operation first expands along the height direction of the three-dimensional data cube, and independently applies the first one-dimensional orthogonal transformation to each row of pixel sequences. This step can be understood as frequency domain analysis of the vertical structural features within each image block, which converts the spatial brightness variation pattern of each row in the image into a set of spectral coefficients representing the strength of different frequency components in the row. Next, the process enters the second dimension, i.e. along the width direction, and independently applies the second one-dimensional orthogonal transformation to each column sequence in the data cube after the row transformation. After this row-column transformation twice, the original spatial image block has been completely converted to the two-dimensional frequency domain, and the coefficients in the data body at this time can accurately represent the spatial frequency characteristics within each image block. After the first two-dimensional transformation lays the foundation for the spatial frequency domain, the process finally applies the third one-dimensional orthogonal transformation to each spectral coefficient sequence obtained after the first two-dimensional transformation along the depth direction of the three-dimensional data cube, i.e. the stacking direction of the image block. This is the key to achieving coordination: it no longer analyzes the characteristics within a single image block, but analyzes the correlation at the same spatial frequency point between different image blocks in the group. Specifically, if all blocks are highly similar at a certain spatial frequency point (i.e. high consistency within the group), then in the frequency dimension within the group, the energy will be highly concentrated in the direct current or low frequency coefficients representing the consistent signal; conversely, noise or block difference parts will be represented as high frequency components within the group.

[0072] Finally, through the recursive and separable transformation of the three dimensions, each element in the three-dimensional data cube is converted from a spatial pixel value to a three-dimensional transformed domain coefficient that encodes both spatial frequency and group consistency frequency information. A new three-dimensional "signal-noise" separation space is created, in which real and stable image structures within the group (including potential gas leakage features) are energy concentrated due to their high correlation, facilitating identification and retention; while randomly distributed noise is energy dispersed due to its irrelevance, making it easy to be located and filtered out, which ensures that the subsequent filtering operation can be targeted, thereby achieving the best balance between noise suppression and detail preservation, providing an essential high-quality image prerequisite for the final accurate positioning of gas leakage.

[0073] Specifically, when filtering the above-mentioned transform domain coefficients, the filtering strength should not be fixed, but must be dynamically associated with the inherent quality of the three-dimensional image block group, which is specifically manifested as the similarity degree of each image block in the group. The logic is that a group composed of highly similar image blocks means that the internal signal structure is stable and consistent, and random noise is the main interference factor. Therefore, stronger filtering means can be used to boldly suppress the coefficients with smaller amplitudes in the transform domain, so as to more thoroughly eliminate noise without worrying too much about damaging the real signal. On the contrary, a group with a lower similarity degree may contain complex structures such as edges and textures. Therefore, a more conservative "weaker" filtering strength is needed to protect these structural information. The establishment of this principle fundamentally solves the drawbacks brought by the "one-size-fits-all" of traditional filtering methods.

[0074] In order to practice this principle, the embodiment also provides a set of specific, robust and non-manual quality evaluation and sorting mechanism, as shown in Figure 5 The steps include: a virtual reference block is synthesized by calculating the gray value median of all blocks in the group at each pixel position. The median is used instead of the mean because it can naturally resist the influence of individual abnormal blocks (such as noise points or atypical interference) in the group, so as to obtain a more reliable reference that can represent the mainstream consensus characteristics of the group.

[0075] After establishing this robust reference, the difference matrix construction step is used to accurately score each member in the group. The absolute difference of all pixels between each image block and the reference block is calculated and summed to obtain the cumulative difference. This value quantitatively describes the deviation of each block from the consensus in the group. The lower the value, the closer the block is to the consensus, i.e. the higher the similarity degree.

[0076] Finally, in the sorting sequence generation step, only the cumulative difference values are arranged in ascending order of numerical value, and a clear quality sorting sequence can be directly generated. This sequence objectively reflects the gradient distribution of the core high-quality signal to the edge similar members in the group.

[0077] In this embodiment, the filtering algorithm no longer operates blindly, but intelligently selects or even dynamically adjusts filtering parameters based on the intra-group quality distribution revealed by the sorting sequence. For example, a more aggressive threshold can be used for highly similar groups. During reconstruction, different weights can be assigned to different blocks based on the sorting. This mechanism ensures that the denoising process can adapt to the image content: strong denoising is performed in flat background areas, while fine-grained maintenance is applied in areas with rich details or edges. This ultimately allows the method to achieve extremely clean denoising results while minimizing the introduction of blur or artifacts. In particular, it can achieve unprecedented protection and enhancement for weak signals at the boundary between gas leaks and the background, greatly improving the accuracy of subsequent gas localization.

[0078] Specifically, after performing a 3D inverse transformation to obtain a denoised 3D image patch group, when performing the step of returning each image patch in the denoised 3D image patch group to its original image position, because the reconstruction quality of each image patch after the aforementioned complex processing is not uniform, simply performing an arithmetic average will dilute the processing effect of high-quality patches by low-quality patches. Therefore, referring to... Figure 6 As shown, this embodiment introduces a quality-based weighted fusion mechanism to ensure that the final generated denoised image can gather the processing essence of all image blocks, thereby maximizing the overall denoising effect. The mechanism includes the following steps: assigning a weight value positively correlated with the reconstruction quality to each image block after the three-dimensional inverse transform processing. The reconstruction quality measures the degree to which the true signal of the image block is preserved after undergoing a series of processing steps such as transformation, filtering, and inverse transform, rather than the amount of residual noise. This design is a prerequisite for quality optimization, which means that it can intelligently identify which image blocks have been processed better and cleaner.

[0079] After obtaining the weight of each block, the process enters the substantive weighted synthesis stage. First, according to their respective weight values, each pixel value in the block is weighted. In this stage, the pixel values ​​of high-quality image blocks will account for a larger proportion in the subsequent fusion; conversely, the influence of blocks with slightly lower quality will be suppressed accordingly.

[0080] Finally, all these weighted image blocks are sent back to the global image coordinate system according to their original acquisition positions for superimposed fusion. Since the image blocks are overlapped and divided, each pixel position will receive weighted contribution values from multiple different blocks. Simply summing these contribution values will cause brightness distortion, so the value of each pixel position in the fused image is normalized based on the total weight received by the position, which ensures that the final gray value of each pixel is the weighted average of the weighted contribution values provided by all the image blocks covering the position, thereby ensuring the accuracy and consistency of the output image in brightness.

[0081] From a mechanism analysis, this embodiment constructs a quality-driven image reconstruction channel. Through this set of weighted average aggregation mechanism, the processing advantages generated in the previous steps (such as block matching and three-dimensional filtering) can be transmitted and converged into the final denoised image with higher weights by high-quality image blocks. At the same time, individual defects or residual noise that may be generated in the processing process are effectively suppressed due to their low block weight, which has little effect on the final image. This is equivalent to a precise signal purification at the global image level, which not only makes the overall denoising effect better, but also effectively avoids the block-like artifacts or local blur caused by the quality difference between blocks, thereby generating a more uniform, clear and reliable preliminary denoised image, which lays an impeccable data foundation for the subsequent accurate positioning of the gas leakage area.

[0082] Specifically, the reconstruction quality is evaluated by calculating the proportion of core energy retained in the transform domain before and after filtering. The higher the proportion of core energy, the greater the weight. After transforming the three-dimensional image block group, the part with large amplitude in the transform domain coefficient usually corresponds to the real and significant structural information (such as edges, contours, etc.) in the image block, which can be regarded as "core energy". The small amplitude coefficient often contains detailed texture and a large amount of random noise. The essence of the subsequent filtering process is a threshold operation, which retains (or slightly attenuates) these core large coefficients, while significantly suppressing or even zeroing the small coefficients.

[0083] Based on this, the evaluation process is as follows: for a given image block, the total energy of its transform domain coefficients before and after filtering (usually represented as the sum of the squares of the coefficients) is calculated, and then the ratio of the energy of the coefficients retained after filtering to the original total energy before filtering is calculated. This ratio is the "proportion of core energy". This ratio has a clear physical meaning: it quantifies the proportion of the most core and most likely real information in the original signal of the image block that is retained after denoising.

[0084] If the core energy proportion of an image block is high after filtering, it indicates that the filtering process accurately removes noise (manifested as a large number of small coefficients being removed) while perfectly protecting the structural signals in the core, and the reconstruction quality of such an image block is naturally optimal, so it should be given a greater weight to play a leading role in the final image synthesis; on the contrary, if the core energy proportion of an image block is low, it means that important signals are damaged or residual noise energy is still considerable in the filtering process, and the reconstruction quality is poor, so a smaller weight is allocated to limit the possible negative impact on the final result.

[0085] From a mechanism analysis, using the reconstruction quality as a feature for weight allocation ensures the fairness and efficiency of weight allocation, which makes the weighted aggregation not a simple average, but a kind of survival of the fittest in signal fidelity. Through this mechanism, the preliminary denoising image generated finally is not only smoothed, but also contributed by image blocks that have survived the filtering test in each local area and have the highest signal fidelity. This ensures that the denoised image can retain the structural integrity of the real scene to the greatest extent while obtaining a low-noise background, especially for the weak edges and texture features representing gas leakage, which can be highlighted in the final image due to their successful preservation in the processing process (i.e., high core energy proportion), thereby providing an important guarantee for accurately determining the gas leakage position in the subsequent steps.

[0086] Specifically, when performing the step of independently performing a contrast stretching transformation according to the gray scale distribution of the internal pixels, a simple global contrast stretching may amplify the background noise while enhancing the target, resulting in the gas leakage area being submerged in the background clutter. To solve this problem, the embodiment further proposes a region-adaptive stretching strategy to achieve targeted enhancement, ensuring that the gas leakage features are effectively highlighted while irrelevant noise interference is actively suppressed, including the following steps: identifying the gray scale distribution main range of the similar region, rather than focusing on the gray scale distribution of the entire image, but focusing on the homogeneous region representing potential gas leakage or a specific background determined after the aforementioned denoising and region division process. The main range is usually determined by counting the main concentration interval of the pixel gray scale in the region (for example, by means of the main peak area of the histogram), which represents the most core and most common signal features in the region.

[0087] After the main body range is accurately defined, the main body range is linearly mapped to all or most of the preset output gray scale interval, and the core signal that may originally have insufficient contrast and appear flat within the similar area is maximally expanded in its gray scale dynamic range. This makes the subtle gray scale changes and texture features within the area more clearly distinguishable, thereby greatly enhancing the distinguishability of the gas leakage area from the surrounding background.

[0088] In sharp contrast to the enhancement processing of the core signal, the processing manner of the noise is simultaneously determined: the discrete noise pixels distributed outside the main body range are suppressed to the dark end (i.e., the low gray scale value region close to black) of the output gray scale interval. These discrete points are usually residual noise that cannot be completely eliminated by the previous noise reduction step or some abnormal values on the area boundary. By forcibly suppressing them to the dark end, the visual saliency of the background noise is actively reduced while the target signal is enhanced.

[0089] Referring to Figure 7 To implement the above method, the application further discloses an infrared image noise reduction system for gas leakage detection, which comprises:

[0090] An image acquisition module is configured to acquire an original infrared image for capturing potential thermal radiation differences of gas leakage, thereby obtaining initial image data containing noise.

[0091] A three-dimensional block group construction module is configured to divide the original infrared image into a plurality of overlapping image blocks, search for similar area blocks, and combine all similar image blocks into a three-dimensional image block group, thereby obtaining a plurality of three-dimensional image block groups.

[0092] A collaborative filtering module is configured to perform three-dimensional transformation on each three-dimensional image block group, obtain transformation domain coefficients, and perform filtering processing on the transformation domain coefficients to suppress noise in the initial image data.

[0093] An image reconstruction module is configured to perform three-dimensional inverse transformation on the filtered transformation domain coefficients, obtain a three-dimensional image block group after noise reduction, and return each image block in the three-dimensional image block group after noise reduction to its original image position. By performing weighted average aggregation on all image blocks, a preliminary noise reduction image is generated. Wherein, image blocks belonging to the same three-dimensional image block group are marked as the same similar area, and the collection of all similar areas constitutes an initial area division of the preliminary noise reduction image.

[0094] A gas positioning module is configured to, based on the initial area division, for each similar area, independently perform a contrast stretching transformation according to the gray scale distribution of the pixels within the similar area, to enhance the feature contrast of the gas leakage area and the background area, and determine the gas leakage position.

[0095] Obviously, the above-mentioned embodiments are only examples for clearly illustrating the present application, and are not intended to limit the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, it is not necessary and also impossible to enumerate all the embodiments. The obvious changes or variations derived from the above description are still within the protection scope of the present application.

Claims

1. An infrared image denoising method for gas leak detection, characterized in that: The method comprises the following steps: Obtaining original infrared images to capture potential thermal radiation differences of gas leakage, to obtain initial image data containing noise; Dividing the original infrared images into multiple overlapping image blocks, searching for similar area blocks, and combining all similar image blocks into a three-dimensional image block group to obtain multiple three-dimensional image block groups; searching for similar area blocks comprises the following steps: taking one of the current image blocks as the core, calculating the feature vector distance between all candidate image blocks and the current image block within its preset search neighborhood; according to the feature vector distance, the candidate image blocks are divided into multiple temporary clusters by dynamic clustering; all candidate image blocks belonging to the same temporary cluster as the current image block are selected and determined as similar area blocks; wherein the feature vector is composed of the gray statistical feature and the spatial gradient feature of the image block; combining all similar image blocks into a three-dimensional image block group comprises: establishing the current image block to be processed as a reference block, performing spatial offset compensation on each similar block with the spatial coordinates of the reference block as the reference among all searched similar blocks, to ensure geometric alignment of all blocks in the same local coordinate system; performing linear normalization processing of the gray value in each image block, including the reference block, to eliminate overall brightness deviation caused by local light or temperature differences, so that all blocks are at the same intensity level; regarding each two-dimensional image block with a size of MxN pixels after alignment and normalization processing as an independent two-dimensional plane; stacking these two-dimensional planes along the third spatial dimension perpendicular to the plane, and determining the stacking order of each stacked plane according to its similarity degree with the reference block; finally forming a three-dimensional data cube with a size of MxNxK, where K is the total number of stacked image blocks, and the cube is the three-dimensional image block group. The three-dimensional transformation is performed on each three-dimensional image block group to obtain transform domain coefficients, and filtering processing is performed on the transform domain coefficients to suppress noise in the initial image data; the three-dimensional transformation is performed on each three-dimensional image block group to obtain transform domain coefficients, including: presetting a transform sequence composed of three one-dimensional orthogonal transform bases in sequence, and the three transform bases correspond to the height, width and depth dimensions of the three-dimensional data cube respectively; along the height direction of the three-dimensional data cube, a first one-dimensional orthogonal transform is independently applied to each row of pixel sequences to convert the pixel values into frequency spectrum coefficients in the dimension; along the width direction of the three-dimensional data cube, a second one-dimensional orthogonal transform is independently applied to each column of pixel sequences to convert the pixel values into frequency spectrum coefficients in the dimension; along the depth direction of the three-dimensional data cube, that is, the stacking direction of the image blocks, a third one-dimensional orthogonal transform is independently applied to each frequency spectrum coefficient sequence obtained after the first two-dimensional transformation to convert each element in the three-dimensional data cube from a spatial pixel value into a three-dimensional transform domain coefficient; when performing filtering processing on the transform domain coefficients, the similarity of the image blocks in the three-dimensional image block group is sorted, and the strength of the filtering operation is dynamically adjusted according to the similarity of the image blocks in the three-dimensional image block group; for a three-dimensional image block group with high similarity, a stronger filtering strength is used; for a three-dimensional image block group with low similarity, a weaker filtering strength is used; The three-dimensional inverse transformation is performed on the filtered transform domain coefficients to obtain a three-dimensional image block group after noise reduction, and each image block in the three-dimensional image block group after noise reduction is returned to its original image position, and a preliminary denoising image is generated by weighted average aggregation of all image blocks; wherein the image blocks belonging to the same three-dimensional image block group are marked as the same similar region, and the collection of all similar regions constitutes the initial region division of the preliminary denoising image; including: assigning a weight value determined based on the reconstruction quality to each image block processed by the three-dimensional inverse transformation; performing weighted processing on all pixel values in each image block according to the weight value to obtain a series of weighted image blocks; superimposing and fusing all weighted image blocks according to their original image positions, and performing total weight-based normalization calculation on the values of each pixel position in the fused image to generate the preliminary denoising image; Based on the initial region division, for each similar region, a contrast stretching transformation is independently performed according to the gray scale distribution of the internal pixels, including: identifying the main body range of the gray scale distribution of the similar region; linearly mapping the main body range to all or a major part of the preset output gray scale interval, while suppressing the discrete noise pixel points distributed outside the main body range to the dark end of the output gray scale interval, to enhance the feature contrast of the gas leakage region and the background region and determine the gas leakage position.

2. The infrared image denoising method for gas leak detection according to claim 1, characterized in that: The similarity of the image blocks in the three-dimensional image block group is sorted, including: For each three-dimensional image block group, a reference image block is generated by calculating the median value of the gray scale of all image blocks in the group at each pixel position; Calculate the pixel-level absolute difference between each image block in the group and the reference image block respectively, and sum all the pixel difference values of each image block to obtain the cumulative difference degree of each image block relative to the reference block, and the cumulative difference degrees of all image blocks constitute a difference degree matrix; Arrange the cumulative difference degrees in the difference degree matrix in ascending order of numerical value, and the lower the numerical value represents the more similar to the reference block, and generate a similarity degree sorting sequence of each image block in the three-dimensional image block group according to the arrangement order.

3. The infrared image denoising method for gas leak detection according to claim 1, wherein: The reconstruction quality is evaluated by calculating the core energy proportion of the image block remaining in the transform domain before and after filtering, and the higher the core energy proportion, the greater the weight given.

4. An infrared image denoising system for gas leak detection for implementing the method according to any one of claims 1 to 3, characterized in that: Comprise: An image acquisition module for acquiring original infrared images that capture potential thermal radiation differences of gas leakage to obtain initial image data containing noise; A three-dimensional block group construction module for dividing the original infrared images into a plurality of overlapping image blocks, searching for similar area blocks, and combining all similar image blocks into a three-dimensional image block group to obtain a plurality of three-dimensional image block groups; A collaborative filtering module for performing three-dimensional transformation on each three-dimensional image block group to obtain transform domain coefficients, and filtering the transform domain coefficients to suppress noise in the initial image data; An image reconstruction module for performing three-dimensional inverse transformation on the filtered transform domain coefficients to obtain a denoised three-dimensional image block group, and returning each image block in the denoised three-dimensional image block group to its original image position, and generating a preliminary denoised image by weighted average aggregation of all image blocks; wherein the image blocks belonging to the same three-dimensional image block group are marked as the same similar area, and the collection of all similar areas constitutes the initial area division of the preliminary denoised image; A gas positioning module for performing contrast stretching transformation on each similar area independently based on the initial area division according to the gray scale distribution of the pixels inside the similar area to enhance the feature contrast of the gas leakage area and the background area, and determining the gas leakage position.

Images