A forest fire unmanned aerial vehicle inspection image processing system

Abstract

The application relates to the field of fire-fighting early warning technology, in particular to a forest fire-fighting unmanned aerial vehicle inspection image processing system. For the initial connected domain in the infrared image, whether the initial connected domain has fire area with different states is judged through comparison between high-temperature points, and then the initial connected domain is segmented to obtain an accurate fire connected domain. Based on the accurate fire connected domain, a pixel point with more obvious features is found out through classification and comparison, the pixel point is taken as a pixel point needing to be amplified in obviousness and the obviousness is adjusted, and then an obviousness image with eliminated errors is obtained. The initial connected domain needing to be segmented is segmented through matching in the initial connected domain and the edge shape outside the initial connected domain, the accurate fire area is determined, then the pixel point needing to be adjusted in obviousness is determined by using the classification and comparison method, and an obviousness image with strong information reference is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire warning, in particular to a forest fire unmanned aerial vehicle inspection image processing system. BACKGROUND

[0002] Forest fires are a global environmental problem that poses a significant threat to ecosystems, economies, and human social safety. Prevention and timely response to forest fires are crucial to reducing the impact of this disaster. However, forests are often vast and complex in terrain, making traditional ground monitoring and warning methods limited. In the prior art, unmanned aerial vehicles are used to locate fire locations and fire areas through aerial inspection.

[0003] Significance detection algorithms play an important role in forest environment monitoring by unmanned aerial vehicles, which can help quickly locate key areas in images, such as abnormal temperature concentration areas, color or texture changes, and potential fire sources. In the CA significance detection algorithm, context information is used to compare with the nearest significant area when obtaining the significance of the region. This method of obtaining significance is relatively simple, and some feature significant pixel points cannot obtain more significant significance, resulting in a large error in the significance detection result. In addition, in the fire scene, the fire will continue to spread, causing the determination of the fire area to be chaotic, which will lead to different fire progress states being identified as one area due to the spread factor. Because the region is not clearly defined, it causes errors in the acquisition of significance when comparing significance, which further affects the effectiveness of forest fire warning data. SUMMARY

[0004] In order to solve the technical problem of the prior art that the significance has errors when analyzing the infrared image of forest fire due to unclear fire area definition and other factors, the purpose of the present application is to provide a forest fire unmanned aerial vehicle inspection image processing system, and the technical solution adopted is as follows:

[0005] The present application provides a forest fire unmanned aerial vehicle inspection image processing system, which comprises:

[0006] An infrared image acquisition module is configured to obtain an infrared image collected by an unmanned aerial vehicle during a forest fire event. The infrared image includes a plurality of initial connected domains.

[0007] The fire conduction domain division module is configured to: for each initial conduction domain, obtain an edge recess degree of the initial conduction domain; select high-temperature points in the initial conduction domain and combine the high-temperature points in pairs to obtain a high-temperature point pair; obtain a matching degree of pixel value variation characteristics of two high-temperature points in the high-temperature point pair in different directions in the initial conduction domain; determine whether the initial conduction domain needs to be divided according to the matching degree and the edge recess degree; and for the initial conduction domain that needs to be divided, determine a division line according to variation differences between adjacent pixel values, taking the edge recess point as a starting point, to obtain a divided fire conduction domain.

[0008] The saliency adjustment module is configured to: obtain different conduction domain clusters according to areas of the fire conduction domains; in one conduction domain cluster, determine a relationship between a pixel value difference between each pixel point and a neighborhood pixel point and a preset fluctuation change range of the conduction domain cluster, select a pixel point that needs to be amplified in saliency, and adjust the saliency to obtain an adjusted saliency image.

[0009] Further, the edge recess degree obtaining method comprises:

[0010] obtaining a curvature of each edge pixel point position on the initial conduction domain, selecting a recess position according to the curvature, and performing normalization processing on an absolute value of an average curvature of the recess position to obtain the edge recess degree.

[0011] Further, the high-temperature point selection method comprises:

[0012] In the initial conduction domain, a pixel point with a pixel value greater than a preset threshold value is selected as the high-temperature point.

[0013] Further, the matching degree obtaining method comprises:

[0014] In one high-temperature point pair, for each high-temperature point, a straight line with an edge pixel point as an end point is drawn in different directions, taking each high-temperature point as a starting point, and pixel values on the straight line form a pixel value variation sequence in each direction; the pixel value variation sequences between the two high-temperature points are matched by using a DTW algorithm to obtain a sequence similarity, all matched sequence pairs are obtained, and an average value of the sequence similarity of all the matched sequence pairs is taken as the matching degree.

[0015] Further, the determination of whether the initial conduction domain needs to be divided according to the matching degree and the edge recess degree comprises:

[0016] For one high-temperature point pair, an overall confusion weight of each high-temperature point pair is obtained according to fluctuation of the pixel value variation sequence of the high-temperature point; after the matching degree is negatively correlated and normalized, the overall confusion weight is multiplied by the matching degree to obtain a non-connection of each high-temperature point pair.

[0017] The product of the average non-connection of all high-temperature point pairs and the edge recess degree is taken as the spread degree of the initial connected domain; if the spread degree is greater than a preset spread degree threshold, it is determined that the initial connected domain needs to be segmented.

[0018] Further, the method for obtaining the overall confusion weight comprises:

[0019] For each high-temperature point in a high-temperature point pair, the average variance of all pixel value change sequences is normalized to obtain a high-temperature point confusion weight; and the average high-temperature point confusion weight of the two high-temperature points in a high-temperature point pair is taken as the overall confusion weight of the high-temperature point pair.

[0020] Further, the method for obtaining the edge recess point comprises:

[0021] The recess position with the minimum curvature is taken as the edge recess point.

[0022] Further, the method for determining the segmentation line comprises:

[0023] Growth is performed in the eight-neighbor domain with the edge recess point as an initial growth point; the neighbor pixel points in the initial growth point neighborhood are taken as first-neighbor pixel points, and the pixel points in a preset neighborhood of the first-neighbor pixel points are taken as second-neighbor pixel points; in the preset neighborhood of the first-neighbor pixel points, a plurality of sets of symmetric directions are obtained with the first-neighbor pixel points as the symmetric points, for each direction, the pixel value difference of the adjacent pixel points in each direction is obtained with the first-neighbor pixel points as the initial points, the numerical difference degree of the pixel value difference and the pixel value difference of the symmetric position in the symmetric direction is obtained, and the cumulative sum of all the numerical difference degrees in a set of symmetric directions is taken as the initial segmentation degree in the symmetric direction; the average initial segmentation degree of all the sets of symmetric directions is taken as the growth feature of the first-neighbor pixel points; the first-neighbor pixel points with the maximum growth feature in the eight-neighbor domain of the initial growth point are taken as new growth points, and continuous growth is performed based on the new growth points until the edge pixel points of the initial connected domain are reached, and all the growth points constitute the segmentation line.

[0024] Further, the method for screening out the pixel points needing to be amplified in saliency comprises:

[0025] The connected domain cluster with the maximum area is taken as a standard connected domain cluster, and the average value of the pixel value difference between all the adjacent pixel points in the standard connected domain cluster is taken as a reference value;

[0026] For each connected domain cluster, the average area in the connected domain cluster is normalized to obtain an area weight, and the product of the area weight and the reference value is taken as the fluctuation change range.

[0027] For each connected domain cluster, obtain the average pixel value difference of each pixel point and all neighboring pixel points in the eight neighborhoods, and if the average pixel value difference is greater than the fluctuation range, determine that the pixel point is a pixel point that needs to be amplified in saliency.

[0028] Further, the method for obtaining the adjusted saliency image comprises:

[0029] The difference value of the average pixel value difference minus the fluctuation range is normalized to obtain an amplification ratio.

[0030] An initial saliency image of the infrared image is obtained by using the CA saliency detection algorithm, and for the pixel point that needs to be amplified in saliency, the initial saliency is amplified by using the amplification ratio to obtain an adjusted saliency image.

[0031] The present application has the following beneficial effects:

[0032] The present application first considers that the fire spread will cause different state fire areas to be fused into one fire area, resulting in errors in saliency detection, so first for the initial connected domain in the infrared image, by comparing the high temperature points, it is judged whether there is a different state fire area in the initial connected domain, that is, the smaller the matching degree and the greater the edge depression degree, the more the initial connected domain is formed by the fusion of different state fire areas, and the more accurate the fire connected domain can be obtained by segmentation. Based on the accurate fire connected domain, by classification and comparison, the pixel point with more prominent features is found out as the pixel point that needs to be amplified in saliency and the saliency is adjusted, and then a saliency image without errors is obtained. The present application segments the initial connected domain that needs to be segmented by the matching condition inside the initial connected domain and the edge shape outside, determines the accurate fire area, and then uses the classification and comparison method to determine the pixel point that needs to be adjusted in saliency, and obtains a saliency image with strong information reference. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0034] Figure 1 A forest fire unmanned aerial vehicle inspection image processing system block diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0035] In order to further clarify the technical means and effects taken by the present application to achieve the predetermined inventive purpose, the specific implementation, structure, features and effects of a forest fire unmanned aerial vehicle inspection image processing system according to the present application are described in detail as follows in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0037] The specific scheme of a forest fire unmanned aerial vehicle inspection image processing system provided by the present application is described in detail below in combination with the drawings.

[0038] Please refer to Figure 1 which shows a forest fire unmanned aerial vehicle inspection image processing system block diagram provided by an embodiment of the present application, the system comprises: an infrared image acquisition module 101, a fire connected domain division module 102, a saliency adjustment module 103.

[0039] It should be noted that when the unmanned aerial vehicle performs the inspection task in the forest inspection area, the collected infrared image can be sent to the infrared image acquisition module 101 in real time, and then the existing connected domain detection algorithm can be used to obtain a plurality of initial connected domains on the infrared image. The pixel value in the infrared image represents the temperature value of the corresponding area, that is, the higher the pixel value, the higher the temperature at the corresponding position.

[0040] Since in the CA saliency detection process, the image collected by the unmanned aerial vehicle contains different regions, different regions have different fire states, that is, different fire periods, which causes the distribution characteristics of different regions to be different. In the CA saliency detection algorithm, when obtaining the saliency of the region, it is necessary to compare with the nearest salient region. Due to the characteristics of fire spread, regions with different fire states are regarded as a connected domain, and comparison based on this connected domain will cause large errors in the saliency detection result, that is, comparison should be made between regions of the same type. Therefore, for the initial connected domain, it is necessary to analyze the internal temperature characteristics to determine whether it belongs to a region with different fire states, and then segment it to obtain the accurate fire connected domain in the infrared image.

[0041] In the fire connected domain division module 102, for each initial connected domain, the same method analysis is needed first, that is, whether the initial connected domain needs to be divided, that is, whether it is a fusion area under different fire states. The embodiment of the application analyzes from two dimensions of the internal state and the external state of the initial connected domain. For the internal state of the initial connected domain, the temperature distribution under different fire states is different, and the temperature change under different fire states is different, so the embodiment of the application adopts a matching form to screen out high temperature points and combine them two by two to obtain the matching degree of the pixel value change characteristics of the two high temperature points in different directions, that is, the smaller the matching degree, the more the two high temperature points belong to high temperature points under different fire states, and the more the initial connected domain belongs to the fusion area under different fire states, and the more it needs to be divided. For the external state of the initial connected domain, the shape of the fusion area will be irregular during area fusion, and obvious concave features will be generated on the edge, so the edge concave degree is obtained, and the greater the edge concave degree, the more the initial connected domain belongs to the fusion area under different fire states, and the more it needs to be divided. Therefore, whether the initial connected domain needs to be divided can be determined according to the matching degree and the edge concave degree; for the initial connected domain that needs to be divided, the edge concave point is taken as the starting point to determine the division line according to the change difference between adjacent pixel values, and the initial connected domain is divided into two connected domains. It should be noted that after the initial connected domain is divided, the divided connected domains still need to be further analyzed whether they need to be divided, until all the connected domains do not need to be divided, at which time the obtained connected domain set is the set of fire connected domains, and all the divided fire connected domains can be obtained.

[0042] Preferably, in the embodiment of the application, the edge concave degree acquisition method comprises:

[0043] The curvature of each edge pixel point position on the initial connected domain is obtained, the concave position is screened out according to the curvature, the absolute value of the average curvature of the concave position is normalized to obtain the edge concave degree. The curvature on the edge can capture the local set features of the edge, and the curvature is a positive number, which means that the corresponding position is a convex position, and the curvature is a negative number, which means that the corresponding position is a concave position. For the concave position, the smaller the curvature, that is, the greater the absolute value of the curvature, the more obvious the concave position, so the absolute value of the average curvature of all the concave positions is normalized to obtain the edge concave degree. In the embodiment of the application, the normalization method selects the range standardization, which normalizes by obtaining the maximum value and the minimum value in the data dimension, and the specific content is a technical means familiar to those skilled in the art, which is not described here.

[0044] Preferably, in the embodiment of the present application, in the initial connected domain, the pixel points with pixel values greater than a preset threshold are selected as the high temperature points. The threshold is set to be four-thirds of the average pixel value in the initial connected domain, that is, the pixel points greater than the threshold are considered as high temperature points with obvious high temperature features, and subsequent analysis of the high temperature points can reduce the data analysis amount and increase the analysis efficiency.

[0045] Preferably, in an embodiment of the present application, considering that the main change of the region in different fire states is different in the temperature distribution in the region, for example, in the middle period of the fire, the region is in a period of intense fire, and the overall region is at a high temperature, and the temperature distribution is relatively uniform; and in the late period of the fire, the fire gradually declines due to the reduction of flammable materials, and only some positions continue to burn, and local positions are at a high temperature, and other positions are relatively low. Based on this, in the embodiment of the present application, the matching degree of the change of the pixel values of the high temperature points in each direction is analyzed, and the method for obtaining the matching degree includes:

[0046] In one of the pairs of high temperature points, for each high temperature point, a straight line with the high temperature point as the starting point and the edge pixel point as the end point is drawn in different directions, and the pixel values on the straight line constitute a pixel value change sequence in each direction. In the embodiment of the present application, the direction of the straight line is the eight-neighborhood direction with the high temperature point as the center, that is, the straight line is extended in eight directions from the high temperature point until it extends to the edge pixel point, and eight pixel value change sequences are obtained for each high temperature point.

[0047] The pixel value change sequences between two high temperature points are matched by using the DTW algorithm to obtain the sequence similarity, all matching sequence pairs are obtained, and the average value of the sequence similarity of all matching sequence pairs is taken as the matching degree. The smaller the matching degree between two high temperature points is, the more different the temperature features of the two high temperature points in different directions are, and the more likely the two high temperature points are located in regions in different fire states.

[0048] In the embodiment of the present application, the similarity of the sequences is quantified by the DTW algorithm, and the specific sequence similarity can be the result of negative correlation mapping and normalization of the DTW distance. By KM matching, a one-to-one pixel value change sequence matching result between two high temperature points can be obtained, and then the matching degree can be obtained.

[0049] In the embodiment of the present application, the method of negative correlation mapping and normalization is to take the reciprocal of the DTW distance as the power of the exponential function with the natural constant as the base, and the output result of the exponential function is the data after negative correlation mapping and with the value range between 0 and 1.

[0050] Further, in the embodiment of the present application, whether the initial connected domain needs to be segmented is determined according to the matching degree and the edge recess degree, including:

[0051] For a high temperature point pair, an overall chaos weight of each high temperature point pair is obtained according to fluctuation of the pixel value change sequence of the high temperature point. The greater the overall chaos weight is, the more likely the high temperature point in the high temperature point pair is a chaos change point, which is probably caused by temperature chaos of fire spread, and the greater the overall chaos weight is, the more the high temperature point pair reflects that the initial connected domain should be segmented.

[0052] The matching degree is weighted according to the overall chaos weight, and non-connection of each high temperature point pair is obtained. That is, for each high temperature point pair, after the matching degree is negatively correlated and normalized, the non-connection of each high temperature point pair is obtained by multiplying the overall chaos weight. Wherein, the greater the overall chaos weight is and the smaller the matching degree is, the more likely the high temperature point in the high temperature point pair is a chaos change point, and the greater the probability of the two high temperature points being in different fire state positions is, and the greater the non-connection is, which means that the two high temperature points are less likely to be region points of the same fire state.

[0053] It should be noted that the method of negative correlation mapping and normalization can also use the above exponential function mapping method, which will not be described in detail.

[0054] The product of the average non-connection of all high temperature point pairs and the edge recess degree is taken as the spread degree of the initial connected domain. That is, the greater the average non-connection is and the greater the edge recess degree is, the more likely the initial connected domain is a fusion region of multiple fire states formed by fire spread, and the greater the spread degree is. If the spread degree is greater than a preset spread degree threshold, it is determined that the initial connected domain needs to be segmented.

[0055] In the embodiment of the present application, the spread degree is further normalized, and the spread degree threshold is set to 0.62. The normalization method can also use range standardization, which will not be described in detail.

[0056] Further, in the embodiment of the present application, the fluctuation of the pixel value change sequence can be represented by the element variance in the sequence, and therefore the method for obtaining the overall chaos weight comprises: for each high temperature point in a high temperature point pair, the average variance of all pixel value change sequences is normalized to obtain a high temperature point chaos weight; and the average high temperature point chaos weight of the two high temperature points in a high temperature point pair is taken as the overall chaos weight of the high temperature point pair.

[0057] Preferably, in an embodiment of the present application, the method for obtaining the recess position based on the curvature takes the recess position with the minimum curvature as the starting point when the initial connected domain is to be segmented.

[0058] Preferably, in an embodiment of the present application, the method for determining the segmentation line comprises:

[0059] Growth is performed in the eight-neighborhood with the edge recess point as an initial growth point; a neighborhood pixel point in the initial growth point neighborhood is taken as a first neighborhood pixel point, and a pixel point in a preset neighborhood of the first neighborhood pixel point is taken as a second pixel point. It should be noted that the embodiment of the present application analyzes the neighborhood of the first neighborhood pixel point in order to analyze the pixel value change of the first neighborhood pixel point in the symmetric direction, and therefore a sufficient number of data points are required, and therefore the neighborhood of the first neighborhood pixel point in the embodiment of the present application is set as a region in a 7*7 window.

[0060] In the preset neighborhood of the first neighborhood pixel point, a plurality of groups of symmetric directions are obtained with the first neighborhood pixel point as a symmetric point. In the embodiment of the present application, the upper, lower, left, right, upper left, upper right, lower left, and lower right eight directions are analyzed, that is, there are four groups of symmetric directions, which are: up and down, left and right, upper left and lower right, and upper right and lower left.

[0061] The purpose of the embodiment of the present application in analyzing the symmetric direction is to consider that the point on the segmentation line should be the intersection between two different fire state regions, and therefore the two sides of the segmentation line should present different pixel value changes, and therefore the embodiment of the present application analyzes the pixel value change in the symmetric direction. For each direction, the pixel value difference of the adjacent pixel points in each direction is obtained with the first neighborhood pixel point as an initial point, the numerical difference degree of the pixel value difference and the pixel value difference of the symmetric position in the symmetric direction is obtained, and the cumulative sum of all numerical difference degrees in a group of symmetric directions is taken as the initial segmentation degree in the symmetric direction. The greater the numerical difference degree, the more asymmetric the pixel value change in the group of symmetric directions, and the more likely the first neighborhood pixel point is a point on the segmentation line.

[0062] In the embodiment of the present application, the numerical difference degree is the absolute value of the difference between the two pixel value differences. It should be noted that because the pixel value change has a direction, the pixel value difference in the embodiment of the present application is the pixel value of the previous pixel point minus the pixel value of the next pixel point in the corresponding direction, that is, the pixel value difference has a positive or negative sign.

[0063] There are a plurality of groups of symmetric directions for a first neighborhood pixel point, and therefore the average initial segmentation degree of all groups of symmetric directions is taken as the growth feature of the first neighborhood pixel point. The greater the growth feature, the more likely the first neighborhood pixel point is a segmentation point on the segmentation line.

[0064] The first neighborhood pixel point with the maximum growth feature in the eight-neighborhood of the initial growth point is taken as a new growth point, and the new growth point is used for continuous growth until the edge pixel point of the initial connected domain is reached, and all the growth points constitute the segmentation line. The segmentation line divides the connected domain to be segmented into two connected domains, and further segmentation analysis is performed on the segmented connected domains, and finally all the fire connected domains can be obtained.

[0065] After all the fire connected domains are obtained, the saliency adjustment can be performed. In order to filter out the pixel points with more salient features, the saliency adjustment module 103 first classifies the fire connected domains according to the areas of the fire connected domains to obtain different connected domain clusters. The fire connected domains in one connected domain cluster can be regarded as the same fire degree and can be uniformly analyzed.

[0066] In one connected domain cluster, first, the pixel value difference between each pixel point and the neighboring pixel points is obtained. The greater the pixel value difference is, the more the pixel point belongs to the temperature prominent point. Therefore, the relationship between the pixel value difference and the preset fluctuation change range of the connected domain cluster to which the pixel point belongs is judged. The greater the pixel value difference of the pixel point is, the more likely the pixel point is a pixel point that needs to be salient. Therefore, the pixel points with more salient features can be filtered out. Such pixel points are usually concentrated in the feature salient region such as the boundary of the fire region, and the saliency of the pixel points is further improved to obtain an adjusted saliency image. The adjusted saliency image has a higher saliency value for the pixel points with more salient features, which facilitates subsequent fire analysis, so that the final fire analysis result is more accurate and the efficiency of the fire analysis is higher.

[0067] Preferably, in the embodiment of the present application, the pixel points that need to be amplified in saliency are screened out, including:

[0068] The connected domain cluster with the largest area is taken as a standard connected domain cluster, and the average value of the pixel value difference between all adjacent pixel points in the standard connected domain cluster is taken as a reference value.

[0069] For each connected domain cluster, the average area in the connected domain cluster is normalized to obtain an area weight, and the product of the area weight and the reference value is taken as the fluctuation change range. That is, the fluctuation change range can be regarded as the overall pixel value fluctuation in the cluster. The pixel value difference greater than the fluctuation range is determined as a feature salient pixel point.

[0070] For each connected domain cluster, the average pixel value difference between each pixel point and all neighboring pixel points in the eight-neighborhood is obtained. If the average pixel value difference is greater than the fluctuation change range, the pixel point is determined as a pixel point that needs to be amplified in saliency.

[0071] Further, the method for obtaining the adjusted saliency image includes:

[0072] The difference between the average pixel value difference and the fluctuation change range is normalized to obtain an increase ratio.

[0073] An initial saliency image of the infrared image is obtained by using a CA saliency detection algorithm, and for a pixel point requiring saliency amplification, the initial saliency is amplified by using the amplification ratio to obtain an adjusted saliency image.

[0074] In the embodiment of the present application, since the amplification ratio is a data between 0 and 1, the amplification ratio is obtained by adding the positive integer 1 to the amplification ratio as an amplification multiple, and the amplified saliency is obtained by directly multiplying the amplification multiple by the initial saliency, and then the adjusted saliency image is obtained.

[0075] In summary, for the initial connected domain in the infrared image, whether there is a fire area of different states in the initial connected domain is determined by comparison between high-temperature points, and then the initial connected domain is segmented to obtain an accurate fire connected domain. Based on the accurate fire connected domain, a pixel point with more significant features is found out by classification and comparison, and is taken as a pixel point requiring saliency amplification, and then the saliency is adjusted to obtain a saliency image with errors eliminated. The initial connected domain requiring segmentation is segmented by matching inside the initial connected domain and the edge shape outside, the accurate fire area is determined, and then the pixel point requiring saliency adjustment is determined by the classification and comparison method, and a saliency image with strong information reference is obtained.

[0076] It should be noted that the above-mentioned embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0077] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

Claims

1. A forest fire unmanned aerial vehicle inspection image processing system, characterized in that, The system comprises: An infrared image acquisition module, configured to obtain infrared images collected by a UAV on a forest fire event; the infrared images comprise a plurality of initial connected domains; A fire connected domain division module, configured to, for each initial connected domain, obtain an edge indentation degree of the initial connected domain; screen out high-temperature points in the initial connected domain and combine the high-temperature points in pairs to obtain a pair of high-temperature points; obtain a matching degree of pixel value variation characteristics of the two high-temperature points in the pair of high-temperature points in different directions in the initial connected domain; determine whether the initial connected domain needs to be divided according to the matching degree and the edge indentation degree; for the initial connected domain that needs to be divided, determine a division line according to variation differences between adjacent pixel values, taking the edge indentation point as a starting point, to obtain a divided fire connected domain; A saliency adjustment module, configured to obtain different connected domain clusters according to areas of the fire connected domains; in one connected domain cluster, determine a relationship between a pixel value difference between each pixel point and a neighborhood pixel point and a preset fluctuation change range of the connected domain cluster, screen out pixel points that need to be amplified in saliency, and adjust the saliency to obtain an adjusted saliency image. 2.The forest fire unmanned aerial vehicle inspection image processing system of claim 1, wherein, The edge indentation degree acquisition method comprises: Obtaining a curvature of each edge pixel point position on the initial connected domain, screening out a indentation position according to the curvature, and performing normalization processing on an absolute value of an average curvature of the indentation position to obtain the edge indentation degree. 3.The forest fire unmanned aerial vehicle inspection image processing system of claim 1, wherein, The high-temperature point screening method comprises: In the initial connected domain, selecting a pixel point with a pixel value greater than a preset threshold as the high-temperature point.

4. The forest fire unmanned aerial vehicle inspection image processing system according to claim 1, wherein, The matching degree acquisition method comprises: In one pair of high-temperature points, for each high-temperature point, a straight line with an edge pixel point as a terminal point is drawn in different directions, taking each high-temperature point as a starting point; pixel values on the straight line form a pixel value variation sequence in each direction; the pixel value variation sequences between the two high-temperature points are matched by using a DTW algorithm to obtain sequence similarity, to obtain all matching sequence pairs; and an average value of the sequence similarity of all the matching sequence pairs is taken as the matching degree.

5. The forest fire unmanned aerial vehicle inspection image processing system according to claim 4, wherein, The determination of whether the initial connected domain needs to be divided according to the matching degree and the edge indentation degree comprises: For one pair of high-temperature points, an overall chaos weight of each pair of high-temperature points is obtained according to fluctuation of the pixel value variation sequence of the high-temperature points; after the matching degree is negatively correlated and normalized, the overall chaos weight is multiplied by the matching degree to obtain a non-connection of each pair of high-temperature points; A product of an average non-connection of all the pairs of high-temperature points and the edge indentation degree is taken as a spread degree of the initial connected domain; if the spread degree is greater than a preset spread degree threshold, it is determined that the initial connected domain needs to be divided. 6.The forest fire unmanned aerial vehicle inspection image processing system of claim 5, wherein, The overall chaos weight acquisition method comprises: For each high-temperature point in one pair of high-temperature points, a high-temperature point chaos weight is obtained by normalizing an average variance of all the pixel value variation sequences; and an average high-temperature point chaos weight of the two high-temperature points in one pair of high-temperature points is taken as an overall chaos weight of the pair of high-temperature points.

7. The forest fire unmanned aerial vehicle inspection image processing system according to claim 2, wherein, The edge indentation point acquisition method comprises: The indentation position with the minimum curvature is taken as the edge indentation point. 8.The forest fire unmanned aerial vehicle inspection image processing system of claim 1, wherein, The division line determination method comprises: Taking the edge recess point as an initial growth point, growth is performed in an eight-neighbor domain; taking a neighbor pixel point in a neighborhood of the initial growth point as a first neighborhood pixel point, a pixel point in a preset neighborhood of the first neighborhood pixel point is a second pixel point; in a preset neighborhood of the first neighborhood pixel point, a plurality of groups of symmetric directions are obtained by taking the first neighborhood pixel point as a symmetric point, for each direction, a pixel value difference of adjacent pixel points in each direction is obtained by taking the first neighborhood pixel point as an initial point, a numerical difference degree of the pixel value difference and a pixel value difference of a symmetric position in the symmetric direction is obtained, and an accumulation of all numerical difference degrees in a group of symmetric directions is taken as an initial segmentation degree in the symmetric direction; an average initial segmentation degree of all groups of symmetric directions is taken as a growth feature of the first neighborhood pixel point; a first neighborhood pixel point with a maximum growth feature in the eight-neighbor domain of the initial growth point is taken as a new growth point, and continuous growth is performed based on the new growth point until the edge pixel point of the initial connected domain is reached, and all growth points constitute the segmentation line. 9.The forest fire unmanned aerial vehicle inspection image processing system of claim 1, wherein, The pixel points requiring to be amplified in significance are screened out, including: Taking the connected domain cluster with the largest area as a standard connected domain cluster, and taking an average value of pixel value differences between all adjacent pixel points in the standard connected domain cluster as a reference value; For each connected domain cluster, the average area in the connected domain cluster is normalized to obtain an area weight, and a product of the area weight and the reference value is taken as the fluctuation change range; For each connected domain cluster, an average pixel value difference between each pixel point and all neighborhood pixel points in the eight-neighbor domain is obtained, and if the average pixel value difference is greater than the fluctuation change range, the pixel point is determined as a pixel point requiring to be amplified in significance. 10.The forest fire unmanned aerial vehicle inspection image processing system of claim 9, wherein, The method for obtaining the adjusted significance image includes: The difference between the average pixel value difference and the fluctuation change range is normalized to obtain an amplification ratio; An initial significance image of the infrared image is obtained by using the CA significance detection algorithm, and for the pixel point requiring to be amplified in significance, the initial significance is amplified by using the amplification ratio to obtain the adjusted significance image.

Images