Smoke Recognition Method Based on Smoke Remote Sensing Physical Model

Abstract

This invention belongs to the field of remote sensing image processing and forest fire monitoring technology, and relates to a method for smoke identification in multi-temporal remote sensing images based on a smoke remote sensing physical model. The method includes obtaining multispectral surface reflectance data of the image to be identified and smoke-free background reflectance data of the same area; determining the ideal smoke reflectance data corresponding to each band; calculating the smoke index corresponding to each band; constructing a multispectral smoke index feature matrix; performing factor analysis on the multispectral smoke index feature matrix to extract common components characterizing smoke spectral changes; dividing the first smoke index combination and the second smoke index combination according to factor loadings; constructing a multispectral fusion smoke identification index and a multispectral auxiliary discrimination index; setting a smoke discrimination threshold; determining smoke candidate pixels; analyzing and processing the smoke candidate pixels; eliminating non-smoke interference areas; and obtaining the remote sensing image smoke identification result. The calculation process of this method is clear, and the meaning of the parameters is explicit.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing and forest fire monitoring technology, specifically relating to a smoke recognition method for multi-temporal remote sensing images based on a smoke remote sensing physical model, and particularly relating to a remote sensing image smoke recognition method that uses smoke-free background reflectance and ideal smoke reflectance to construct physical indicators, eliminate complex background spectral interference, and indicate the spatial distribution of smoke concentration. Background Technology

[0002] Forest fires, grassland fires, and other similar fires are characterized by their suddenness, rapid spread, and wide impact, causing not only damage to ecological resources such as forests and grasslands but also threatening human life and property. Smoke, as a significant byproduct of fires, typically forms a spectral and spatial representation in remote sensing imagery even in the early stages of a fire. Therefore, the rapid and accurate identification of smoke using remote sensing imagery is crucial for early fire detection, assessment of fire extent, and disaster emergency response.

[0003] Remote sensing imagery, with its advantages of wide coverage, stable observation cycles, and low data acquisition costs, has become an important data source for fire smoke monitoring. Currently, smoke identification methods mainly include visual interpretation, single-band thresholding, spectral indexing, machine learning, and deep learning. Visual interpretation relies on human experience and has a low degree of automation; single-band thresholding and traditional spectral indexing are computationally simple but are sensitive to background type, smoke concentration, and imaging conditions; machine learning and deep learning methods can extract complex features, but typically require a large number of labeled samples, resulting in high model training costs, and are easily affected by differences in sample distribution when applied across regions and scenarios.

[0004] From the perspective of smoke spectral characteristics, smoke exhibits translucency, diffusivity, and concentration gradient variations. Its reflectance in remote sensing images is influenced not only by the smoke's own optical properties but also by the reflectance of the underlying background. Under complex background conditions such as vegetation, bare land, water bodies, clouds, and shadows, random changes in the background, such as clouds, cloud shadows, eutrophication of water bodies, and vegetation diseases and pests, can lead to spectral variations similar to those of smoke. This weakens the spectral difference between smoke pixels and non-smoke pixels, resulting in missed or false alarms in smoke identification. Especially under conditions of low-concentration smoke or complex background interference, directly using the original reflectance or a single index is insufficient to reliably characterize the degree of change of smoke pixels relative to a smoke-free background.

[0005] However, existing methods such as spectral indices and deep learning essentially rely on mathematical means to enhance the information differences between bands to achieve smoke discrimination, which is difficult to eliminate misjudgments of smoke caused by background changes. Therefore, there is an urgent need to construct an index with clear physical meaning based on the smoke radiative transfer mechanism, and to establish an identification model and method that can effectively eliminate spectral variation interference similar to smoke changes caused by clouds, cloud shadows, water eutrophication, and vegetation diseases and pests, so as to effectively solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a smoke identification method for multi-temporal remote sensing images based on a smoke remote sensing physical model, so as to solve the problem of insufficient physical interpretation of single bands or purely mathematical indicators.

[0007] This invention is achieved through the following technical solution:

[0008] A method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model includes the following steps:

[0009] A. Acquire remote sensing images of the area to be identified containing smoke, and obtain multispectral surface reflectance data of the image to be identified after preprocessing.

[0010] B. Obtain the smoke-free background image corresponding to the area to be identified, and obtain the smoke-free background reflectance data of the same area after preprocessing.

[0011] C. Based on the smoke sample or the limiting spectral characteristics of smoke, determine the ideal smoke reflectance data corresponding to each band;

[0012] D. Select bands B1 to B7 from the multispectral remote sensing image and calculate the smoke index corresponding to each band.

[0013] E. Construct a multispectral smoke index feature matrix based on the smoke indices corresponding to bands B1 to B7.

[0014] F. Perform factor analysis on the multispectral smoke index feature matrix, extract common components that characterize the changes in smoke spectrum, and divide the smoke index corresponding to the B1 band to the B7 band into a first smoke index combination and a second smoke index combination according to the factor loading.

[0015] G. The smoke indices corresponding to the B1 band to the B4 band are determined as the first smoke index combination, and a multispectral fusion smoke recognition index based on the B1 band to the B4 band is constructed using the first smoke index combination.

[0016] H. Determine the smoke index corresponding to the B5 band to the B7 band as the second smoke index combination, and construct a multispectral auxiliary discrimination index based on the B5 band to the B7 band using the second smoke index combination.

[0017] I. Based on the multispectral fusion smoke recognition index obtained in step G and the multispectral auxiliary discrimination index obtained in step H, set the smoke discrimination threshold and determine the smoke candidate pixels;

[0018] J. Analyze and process candidate smoke pixels, remove non-smoke interference areas, and obtain smoke recognition results from remote sensing images.

[0019] Furthermore, in steps A and B, the preprocessing includes one or more of the following: radiometric calibration, atmospheric correction, geometric correction, band registration, resampling, and study area cropping.

[0020] Furthermore, in step B, the smoke-free background image is a smoke-free image of the area to be identified acquired under the same sensor, the same spatial range, and the same or similar seasonal conditions, and is used to characterize the background reflectivity state of the area to be identified.

[0021] Furthermore, in step C, the ideal smoke reflectance data is determined by the reflectance statistics of high-concentration smoke samples in the smoke sample set, or by the statistical analysis of the limiting concentration spectral characteristics of smoke samples from different regions.

[0022] The smoke index for each band is calculated using the following formula:

[0023]

[0024] Among them, NDSI b R is the smoke index corresponding to the b-th band, where b∈{B1,B2,B3,B4,B5,B6,B7}; b R represents the surface reflectance of the b-th band in the image to be identified. 0,b R represents the background reflectance of the b-th band in a smoke-free background image of the same area. s,b Let be the ideal smoke reflectance corresponding to the b-th band.

[0025] Furthermore, in step E, the multispectral smoke index feature matrix is ​​constructed using pixels as sample units and the smoke indices corresponding to bands B1 to B7 as variables.

[0026] Further, in step F, factor analysis includes correlation testing, common component extraction, and factor loading calculation of the multispectral smoke index feature matrix; the first smoke index combination is the smoke index corresponding to bands B1, B2, B3, and B4, and the second smoke index combination is the smoke index corresponding to bands B5, B6, and B7.

[0027] Furthermore, in steps G and H, the fusion coefficient of the smoke index for each band involved in the fusion is 1. The multispectral fusion smoke recognition index is calculated using the following formula:

[0028]

[0029] Among them, INDSI B1_B4 NDSI is a multispectral fusion smoke recognition index based on bands B1 to B4. B1 NDSI B2 NDSI B3 NDSI B4 These are the smoke indices for the B1, B2, B3, and B4 bands, respectively.

[0030] The multispectral-assisted discrimination index is calculated using the following formula:

[0031]

[0032] Among them, INDSI B5_B7 NDSI is a multispectral auxiliary discriminant index based on the B5 to B7 bands. B5 NDSI B6 NDSI B7 These are the smoke indices for the B5, B6, and B7 bands, respectively.

[0033] Furthermore, in step I, the smoke discrimination threshold includes a first threshold and a second threshold, where the first threshold is INDSI. B1_B4 >0.21, the second threshold is INDSI B5_B7 <1.3;

[0034] When determining candidate pixels for smoke detection, the smoke index corresponding to the B7 band is further used for auxiliary discrimination. When a pixel simultaneously meets the INDSI... B1_B4 >0.21, INDSI B5_B7 <1.3 and NDSI B7 When the value is less than 0.5, the pixel is identified as a smoke pixel.

[0035] Furthermore, in step J, spatial post-processing includes one or more of the following: exclusion of non-smoke interference areas, outlier screening, and boundary smoothing.

[0036] The results of smoke recognition in remote sensing images include one or more of the following: smoke binary mask, smoke spatial distribution map, and smoke area statistics.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] 1. This invention constructs a smoke index by introducing the reflectance of a smoke-free background and the reflectance of an ideal smoke in the same area. This makes smoke recognition no longer dependent on the original reflectance or the traditional dual-band difference ratio, but can characterize the degree of change of the pixel to be identified from a smoke-free background to an ideal smoke state, and has clear physical meaning.

[0039] 2. This invention performs factor analysis on the smoke index of the B1 to B7 bands, and uses the results of factor analysis to reveal the spectral physical response and characteristic laws of smoke to different bands. In this way, the B1 to B4 bands are determined to be the main smoke identification combination, and the B5 to B7 bands are the auxiliary discrimination combination, which enhances the rationality of the use of multi-band smoke information.

[0040] 3. This invention fuses the smoke indices corresponding to the B1 to B4 bands to construct INDSIB1_B4, which can enhance the response characteristics of smoke targets; at the same time, it combines INDSIB5_B7 and NDSIB7 for auxiliary constraints, which can reduce misjudgments caused by clouds, water bodies, bare ground, shadows and other complex backgrounds.

[0041] 4. This invention uses a combination of index calculation and threshold discrimination to achieve smoke recognition. The calculation process is clear and the parameters have clear meanings, which facilitates automated processing in remote sensing image processing platforms, geographic information system platforms or fire monitoring business systems. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 The flowchart shows a method for smoke recognition in remote sensing images based on a physical model of smoke remote sensing.

[0044] Figure 2 This is a Landsat8 / OLI remote sensing image from an embodiment of the present invention;

[0045] Figure 3 The pixel NDSI calculation results are shown in the embodiments of the present invention.

[0046] Figure 4 The pixel INDSI calculation results are shown in the embodiments of the present invention.

[0047] Figure 5 This is an image showing the smoke candidate pixel recognition results in an embodiment of the present invention;

[0048] Figure 6 This is a diagram showing the smoke pixel recognition results in an embodiment of the present invention;

[0049] Figure 7 This is a diagram illustrating the advantages of smoke recognition in an embodiment of the present invention;

[0050] Figure 8 for Figure 7 A schematic diagram of a local location in the image, where (a1) is a satellite image of a bare land area, (b1) is a smoke candidate pixel recognition result for the area, (c1) is a smoke pixel recognition result for the area, (a2) is a satellite image of a lake area, (b2) is a smoke candidate pixel recognition result for the area, and (c2) is a smoke pixel recognition result for the area. Detailed Implementation

[0051] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0052] This invention constructs a smoke index (NDSI) with clear physical meaning by introducing smoke-free background reflectance and ideal smoke reflectance, overcoming the shortcomings of existing methods that lack sufficient physical interpretation of single-band or purely mathematical indicators. Based on this, it further combines factor analysis to construct a comprehensive smoke identification index (INDSI) and its discrimination threshold, forming a complete smoke identification system. This solves the problems of poor stability of single-band or single-index methods, large interference from complex backgrounds, and low identification accuracy under thin smoke and variable backgrounds. In particular, it can eliminate spectral variation interference caused by background changes such as clouds, cloud shadows, eutrophication, and vegetation diseases and pests. Simultaneously, the index constructed based on the physical model of this invention can effectively characterize the level of smoke concentration, reflecting the spatial distribution characteristics of smoke concentration while identifying smoke, which is of great significance for forest fire detection, fire source location, and fire situation analysis.

[0053] This invention relates to a multi-temporal remote sensing image smoke recognition method based on a smoke remote sensing physical model, including image acquisition and preprocessing, and R waveband... s The calculation of NDSI values ​​for each band, factor analysis to construct NDSI, classification of smoke recognition thresholds, establishment of recognition models, and smoke recognition are specifically included in the following steps:

[0054] A. Acquire remote sensing images of the area to be identified containing smoke, and preprocess the remote sensing images containing smoke to obtain multispectral surface reflectance data of the image to be identified.

[0055] Specifically, the remote sensing image containing smoke can be Landsat series satellite imagery, Sentinel series satellite imagery, or other remote sensing imagery with multispectral bands. The imagery undergoes preprocessing, including one or more combinations of radiometric calibration, atmospheric correction, geometric correction, band registration, resampling, and study area cropping, to ensure that each band has a uniform spatial extent, spatial resolution, and geographic coordinates.

[0056] B. Acquire smoke-free background images corresponding to the area to be identified, and preprocess the smoke-free background images to obtain smoke-free background reflectance data for the same area. Preprocessing includes one or more combinations of radiometric calibration, atmospheric correction, geometric correction, band registration, resampling, and study area cropping.

[0057] Specifically, the smoke-free background image is preferably a smoke-free image acquired under the same sensor, the same spatial range, and the same or similar seasonal conditions, used to characterize the background reflectance state of the area to be identified. The background reflectance of each pixel in the smoke-free state is extracted from the smoke-free background image to represent the spectral state of the ground objects themselves before the presence of smoke.

[0058] C. Based on statistical analysis of smoke samples, determine the ideal smoke reflectance data corresponding to each band by identifying the smoke limiting spectral characteristics.

[0059] Specifically, the ideal smoke reflectance data can be obtained statistically from high-concentration smoke samples, or from the spectral characteristics of limiting concentrations of typical smoke samples from different regions. Ideal smoke reflectance is used to represent a reference value for band reflectance under conditions of high smoke concentration or significant smoke characteristics.

[0060] D. Select bands B1 to B7 from the multispectral remote sensing image and calculate the smoke index for each band. The calculation formula is as follows:

[0061] ;

[0062] Among them, NDSI b R is the smoke index corresponding to the b-th band, where b∈{B1,B2,B3,B4,B5,B6,B7}; b R represents the surface reflectance of the b-th band in the image to be identified. 0,b R represents the background reflectance of the b-th band in a smoke-free background image of the same area. s,b Let be the ideal smoke reflectance corresponding to the b-th band.

[0063] This index is used to represent the degree to which a pixel to be identified changes from a smoke-free background state to an ideal smoke state. When R b Approaching R 0,b When R indicates that the pixel is close to a smoke-free background state in this band; when R b Approaching R s,b When this occurs, it indicates that the pixel is closer to a smoke state in that band.

[0064] E. Based on the smoke index corresponding to bands B1 to B7, calculate the multispectral smoke index value for each band and construct the multispectral smoke index feature matrix.

[0065] Specifically, the multispectral smoke index feature matrix uses pixels as sample units and NDSI as the basis for its design. B1 NDSI B2 NDSI B3 NDSI B4 NDSI B5 NDSI B6 NDSI B7 Using variables, a multispectral smoke index feature matrix is ​​constructed. This matrix describes the distribution characteristics of each pixel in the seven-band smoke index space.

[0066] F. Factor analysis was performed on the multispectral smoke index feature matrix to extract common components characterizing changes in the smoke spectrum. Based on factor loadings, the smoke indices corresponding to bands B1 to B7 were divided into a first smoke index combination and a second smoke index combination. Factor analysis included correlation testing of the multispectral smoke index feature matrix, extraction of common components, and calculation of factor loadings.

[0067] Specifically, correlation tests and common component extraction were performed on the smoke indices corresponding to bands B1 to B7 to obtain factor loading results. Factor analysis results show that two components contribute significantly to the smoke indices across the seven bands. Specifically, the smoke indices corresponding to bands B1, B2, B3, and B4 contribute strongly to the first component, while the smoke indices corresponding to bands B5, B6, and B7 contribute strongly to the second component.

[0068] Based on the factor analysis results above, the smoke indices of bands B1 to B4 were determined as the first smoke index combination, and the smoke indices of bands B5 to B7 were determined as the second smoke index combination. Considering that the first smoke index combination has a more obvious response effect on smoke recognition, bands B1 to B4 were selected to construct the main smoke recognition index; at the same time, the index combination constructed from bands B5 to B7 was used as an auxiliary constraint indicator.

[0069] G. Construct a multispectral fusion smoke recognition index based on bands B1 to B4. The first smoke index combination is used as the main smoke recognition combination. The smoke indices corresponding to bands B1, B2, B3, and B4 are determined as the first smoke index combination. The multispectral fusion smoke recognition index based on bands B1 to B4 is constructed using the first smoke index combination.

[0070] To simplify the calculation process, the coefficients of the smoke index for each band from B1 to B4 are all set to 1, resulting in:

[0071] ;

[0072] Among them, INDSI B1_B4NDSI is a multispectral fusion smoke recognition index based on bands B1 to B4. B1 NDSI B2 NDSI B3 NDSI B4 These are the smoke indices for the B1, B2, B3, and B4 bands, respectively.

[0073] H. Construct a multispectral auxiliary discrimination index based on bands B5 to B7. Using the second smoke index combination as the auxiliary discrimination combination, the smoke indices corresponding to bands B5, B6, and B7 are determined as the second smoke index combination, and a multispectral auxiliary discrimination index based on bands B5 to B7 is constructed using this second smoke index combination.

[0074] To simplify the calculation process, the coefficients of the smoke index for each band from B5 to B7 are all set to 1, resulting in:

[0075] ;

[0076] Among them, INDSI B5_B7 NDSI is a multispectral auxiliary discriminant index based on the B5 to B7 bands. B5 NDSI B6 NDSI B7 These are the smoke indices for the B5, B6, and B7 bands, respectively.

[0077] I. Set smoke discrimination thresholds based on the multispectral fusion smoke recognition index based on the B1 band to the B4 band and the multispectral auxiliary discrimination index based on the B5 band to the B7 band, and identify smoke pixels according to the threshold rules.

[0078] Specifically, INDSI will be enabled when a pixel meets the following conditions. B1_B4 >0.21 and INDSI B5_B7 When the value is less than 1.3, the pixel is identified as a candidate smoke pixel. In a further preferred embodiment, to enhance the suppression of complex background interference, the smoke index corresponding to the B7 band is introduced as an auxiliary constraint. When a pixel simultaneously satisfies INDSI... B1_B4 >0.21, INDSI B5_B7 <1.3 and NDSI B7 When the value is less than 0.5, the pixel is identified as a smoke pixel.

[0079] J. Perform spatial post-processing on smoke pixels to remove non-smoke interference areas, obtaining the smoke identification results from the remote sensing image. Spatial post-processing includes one or more of the following: non-smoke interference area removal, outlier screening, and boundary smoothing.

[0080] Specifically, the identified smoke pixels are analyzed to remove non-smoke interference areas, and the smoke recognition result is finally obtained.

[0081] The results of smoke recognition in remote sensing images include one or more of the following: smoke binary mask, smoke spatial distribution map, and smoke area statistics.

[0082] Example 1:

[0083] like Figure 1 As shown, a method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model includes the following steps:

[0084] A. Select the multispectral remote sensing image of the area to be identified as a smoke-containing image, perform radiometric calibration, atmospheric correction and study area cropping on the image to obtain the surface reflectance data of the B1 to B7 bands of the image to be identified.

[0085] B. Select smoke-free remote sensing images of the same or similar seasonal conditions in the same area as smoke-free background images. Perform the same preprocessing on the smoke-free background images as on the smoke-containing images to obtain the background reflectance data of the B1 to B7 bands of the smoke-free background images.

[0086] C. Based on the limiting spectral characteristics of high-concentration smoke samples or typical smoke samples from different regions, determine the ideal smoke reflectance data corresponding to bands B1 to B7.

[0087] D. Calculate the smoke index for bands B1 to B7 based on the reflectance of the image to be identified, the reflectance of the smoke-free background, and the ideal smoke reflectance:

[0088] ;

[0089] Where b∈{B1,B2,B3,B4,B5,B6,B7}.

[0090] E. NDSI of each pixel B1 NDSI B2 NDSI B3 NDSI B4 NDSI B5 NDSI B6 NDSI B7 Using as the variable, calculate the multispectral smoke index value for each band.

[0091] F. Factor analysis was performed on the multispectral smoke index feature matrix. The factor analysis results showed that the seven band smoke indices mainly formed two contributing components, with bands B1 to B4 forming the first smoke index combination and bands B5 to B7 forming the second smoke index combination. Considering smoke response capability, computational complexity, and recognition stability, bands B1 to B4 were selected to construct the main smoke recognition index.

[0092] G. Calculate the multispectral fusion smoke recognition index based on bands B1 to B4:

[0093] ;

[0094] H. Calculate the multispectral auxiliary discrimination index based on the B5 band to the B7 band:

[0095] ;

[0096] I. Identify smoke pixels using threshold rules. If a pixel satisfies INDSI... B1_B4 >0.21 and INDSI B5_B7 When the value is less than 1.3, the pixel is identified as a candidate pixel for smoke; furthermore, when the pixel simultaneously satisfies INDSI... B1_B4 >0.21, INDSI B5_B7 <1.3 and NDSI B7 When the value is less than 0.5, the pixel is identified as a smoke pixel.

[0097] J. Perform connected component analysis and spatial post-processing on the smoke pixel recognition results to remove non-smoke interference areas and obtain the final smoke recognition results of the remote sensing image.

[0098] In this embodiment, Figure 1 This paper illustrates the overall technical flow of the multi-temporal remote sensing image smoke recognition method based on a smoke remote sensing physical model according to the present invention. Figure 1 As can be seen, the method includes the following steps in sequence: acquisition and preprocessing of remote sensing images with smoke, acquisition and preprocessing of smoke-free background images, determination of ideal smoke reflectance, calculation of smoke index from B1 band to B7 band, construction of multispectral smoke index feature matrix, factor analysis, construction of main smoke identification index combination, construction of auxiliary discrimination index combination, threshold discrimination, and spatial post-processing.

[0099] Figure 2 The Landsat 8 / OLI remote sensing image used in this embodiment is shown. This image illustrates the original remote sensing image features of the area to be identified, where the smoke area exhibits a certain continuous diffusion pattern in the image. Complex background features coexist with the smoke area, providing basic data for subsequent smoke index calculation and smoke identification.

[0100] Figure 3 The calculation results of the pixel smoke index NDSI in this embodiment are shown. By introducing the smoke-free background reflectance and the ideal smoke reflectance, NDSI can reflect the degree to which a pixel changes from a smoke-free background state to an ideal smoke state. Figure 3 The difference in index values ​​between the smoke region and the non-smoky background region indicates that the smoke index can enhance the separability between smoke pixels and background pixels.

[0101] Figure 4 The calculation results of the Pixel Multispectral Fusion Smoke Recognition Index (INDSI) in this embodiment are shown. This result is obtained by fusing the smoke indices corresponding to bands B1 to B4, and is used to highlight the main smoke response area. Compared with single-band smoke indices, INDSI can comprehensively utilize smoke response information from multiple bands, improving the continuity of the smoke area and the stability of recognition.

[0102] Figure 5 The image shows the smoke candidate pixel identification results based on dual threshold conditions in this embodiment. Specifically, when a pixel meets the INDSI... B1_B4 >0.21 and INDSI B5_B7 When the value is less than 1.3, it is identified as a candidate pixel for smoke. Figure 5 As can be seen, this step can initially extract the main smoke area, but may still retain a small number of non-smoke interference pixels caused by complex background.

[0103] Figure 6 The image shows the smoke pixel recognition results after incorporating B7 band auxiliary constraints in this embodiment. Specifically, in Figure 5 Based on the smoke candidate pixels shown, NDSI_ is further introduced. B7 A value <0.5 is used as an auxiliary discrimination condition to eliminate some non-smoke interference areas. Figure 5 compared to, Figure 6 The results shown can more accurately preserve the main area of ​​the smoke and reduce misjudgments caused by background factors such as clouds, cloud shadows, water changes, and abnormal vegetation.

[0104] Figure 7 The results of the comparison of smoke recognition advantages in the embodiments of the present invention are shown. The images are Landsat8 / OLI remote sensing images of the study area, and the red boxes represent the selected typical non-smoke interference areas. Figure 8 (a1) in the image is a satellite image of a bare area. Figure 8 (b1) in the image represents the smoke candidate pixel identification result obtained after threshold discrimination in the bare land area. Figure 8 (c1) in the figure represents the smoke pixel recognition result obtained after further combining auxiliary discrimination conditions; Figure 8 (a2) in the image is a satellite image of the lake area. Figure 8(b2) in the image represents the smoke candidate pixel recognition result obtained after threshold discrimination in the lake area. Figure 8 (c2) in the figure represents the smoke pixel recognition result obtained after further combining auxiliary discrimination conditions.

[0105] As can be seen from this embodiment, the present invention utilizes the reflectance of a smoke-free background and the reflectance of an ideal smoke to construct a smoke index, which can highlight the degree of change of smoke pixels relative to background pixels. By determining the B1 to B4 bands as the main smoke recognition combination through factor analysis, and combining the B5 to B7 bands and the B7 band index for auxiliary constraints, the stability of smoke recognition in remote sensing images can be effectively improved and misjudgments caused by complex backgrounds can be reduced.

[0106] This invention constructs smoke indices for each band by introducing smoke-free background reflectance and ideal smoke reflectance. Based on this, factor analysis is performed on the smoke indices of bands B1 to B7 to extract the main components of smoke spectral changes. According to the factor analysis results, the smoke indices corresponding to bands B1 to B4 are used as the main smoke identification combination, and the smoke indices corresponding to bands B5 to B7 are used as the auxiliary discrimination combination. Finally, a threshold rule is used to achieve smoke pixel recognition in remote sensing images.

[0107] It will be understood by those skilled in the art that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model, characterized in that, Includes the following steps: A. Acquire remote sensing images of the area to be identified containing smoke, and obtain multispectral surface reflectance data of the image to be identified after preprocessing. B. Obtain the smoke-free background image corresponding to the area to be identified, and obtain the smoke-free background reflectance data of the same area after preprocessing. C. Based on the smoke sample or the limiting spectral characteristics of smoke, determine the ideal smoke reflectance data corresponding to each band; D. Select bands B1 to B7 from the multispectral remote sensing image and calculate the smoke index corresponding to each band. E. Construct a multispectral smoke index feature matrix based on the smoke index; F. Perform factor analysis to extract common components characterizing changes in the smoke spectrum. Based on the factor loading, divide the smoke indices corresponding to the B1 to B7 bands into a first smoke index combination and a second smoke index combination. G. The smoke indices corresponding to the B1 band to the B4 band are determined as the first smoke index combination, and a multispectral fusion smoke recognition index based on the B1 band to the B4 band is constructed using the first smoke index combination. H. Determine the smoke index corresponding to the B5 band to the B7 band as the second smoke index combination, and construct a multispectral auxiliary discrimination index based on the B5 band to the B7 band using the second smoke index combination. I. Based on the multispectral fusion smoke recognition index obtained in step G and the multispectral auxiliary discrimination index obtained in step H, set the smoke discrimination threshold and determine the smoke candidate pixels; J. Analyze and process candidate smoke pixels, remove non-smoke interference areas, and obtain smoke recognition results from remote sensing images.

2. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In steps A and B, preprocessing includes one or more of the following: radiometric calibration, atmospheric correction, geometric correction, band registration, resampling, and study area cropping.

3. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step B, the smoke-free background image is a smoke-free image of the area to be identified, acquired under the same sensor, the same spatial range, and the same or similar seasonal conditions, and is used to characterize the background reflectivity of the area to be identified.

4. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step C, the ideal smoke reflectance data is determined by the reflectance statistics of high-concentration smoke samples in the smoke sample set, or by the statistical analysis of the limiting concentration spectral characteristics of smoke samples from different regions. The smoke index for each band is calculated using the following formula: ； Among them, NDSI b R is the smoke index corresponding to the b-th band, where b∈{B1,B2,B3,B4,B5,B6,B7}; b R represents the surface reflectance of the b-th band in the image to be identified. 0,b R represents the background reflectance of the b-th band in a smoke-free background image of the same area. s,b Let be the ideal smoke reflectance corresponding to the b-th band.

5. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step E, the multispectral smoke index feature matrix is ​​constructed using pixels as sample units and the smoke indices corresponding to bands B1 to B7 as variables.

6. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step F, factor analysis includes correlation testing, common component extraction, and factor loading calculation of the multispectral smoke index feature matrix; the first smoke index combination is the smoke index corresponding to bands B1, B2, B3, and B4, and the second smoke index combination is the smoke index corresponding to bands B5, B6, and B7.

7. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In steps G and H, the fusion coefficient of the smoke indexes of each band involved in the fusion is 1. The multispectral fusion smoke recognition index is calculated using the following formula: ； Among them, INDSI B1_B4 NDSI is a multispectral fusion smoke recognition index based on bands B1 to B4. B1-4 These are the smoke indices for the B1, B2, B3, and B4 bands, respectively. The multispectral-assisted discrimination index is calculated using the following formula: ； Among them, INDSI B5_B7 NDSI is a multispectral auxiliary discriminant index based on the B5 to B7 bands. B5-B7 These are the smoke indices for the B5, B6, and B7 bands, respectively.

8. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step I, the smoke discrimination threshold includes a first threshold and a second threshold. The first threshold is INDSI. B1_B4 >0.21, the second threshold is INDSI B5_B7 <1.3; When determining candidate pixels for smoke detection, the smoke index corresponding to the B7 band is further used for auxiliary discrimination. When a pixel simultaneously meets the INDSI... B1_B4 >0.21, INDSI B5_B7 <1.3 and NDSI B7 When the value is less than 0.5, the pixel is identified as a smoke pixel.

9. The method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: In step J, spatial post-processing includes one or more of the following: exclusion of non-smoke interference areas, outlier screening, and boundary smoothing.

10. A method for smoke recognition in multi-temporal remote sensing images based on a smoke remote sensing physical model according to claim 1, characterized in that: The results of smoke recognition in remote sensing images include one or more of the following: smoke binary mask, smoke spatial distribution map, and smoke area statistics.

Images