CN116563659A - Optical smoke detection method combining priori knowledge and feature classification - Google Patents

Optical smoke detection method combining priori knowledge and feature classification Download PDF

Info

Publication number
CN116563659A
CN116563659A CN202210108504.XA CN202210108504A CN116563659A CN 116563659 A CN116563659 A CN 116563659A CN 202210108504 A CN202210108504 A CN 202210108504A CN 116563659 A CN116563659 A CN 116563659A
Authority
CN
China
Prior art keywords
smoke
image
edge
gray
smoke candidate
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210108504.XA
Other languages
Chinese (zh)
Inventor
赵英海
许静
何良
赵国
张品
刘振
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Huahang Radio Measurement Research Institute
Original Assignee
Beijing Huahang Radio Measurement Research Institute
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Huahang Radio Measurement Research Institute filed Critical Beijing Huahang Radio Measurement Research Institute
Priority to CN202210108504.XA priority Critical patent/CN116563659A/en
Publication of CN116563659A publication Critical patent/CN116563659A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration by the use of histogram techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/155—Segmentation; Edge detection involving morphological operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

The invention relates to an optical smoke detection method combining priori knowledge and feature classification, which comprises the following steps: dividing a smoke candidate region from the target image based on the smoke gray scale characteristics; carrying out morphological treatment on the smoke candidate areas, and screening out first-class smoke candidate areas by combining smoke shape information and smoke height information; extracting multidimensional feature descriptors from the smoke candidate areas based on gray features, texture features and gradient features of the images, sending the multidimensional feature descriptors into a trained classifier to judge, and screening out second-class smoke candidate areas; selecting an intersection of the first type smoke candidate region and the second type smoke candidate region as a confirmed smoke candidate region; and carrying out edge extraction and fusion on the confirmed smoke candidate region to obtain the position information of the smoke region, thereby obtaining the final smoke detection result output. The algorithm has low complexity, occupies less hardware resources, can rapidly realize smoke detection, and is suitable for an operation platform with limited hardware resources.

Description

Optical smoke detection method combining priori knowledge and feature classification
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an optical smoke detection method combining priori knowledge and feature classification.
Background
At present, smoke detection and recognition technology is mainly focused on the field of fire detection, and the purpose of detecting smoke is to discover fire earlier, so that precious time is won for fire extinguishment. In addition, most of application scenes are video monitoring scenes under a fixed background, and research is mainly focused on video smoke detection and identification technologies utilizing characteristics of smoke such as color, texture, turbulence and waving.
However, in the sea-sky background, there is no effective detection method for detecting smoke interference near the water surface, and the conventional video smoke detection and recognition technology based on characteristics of color, texture, turbulence, drift, etc. because the image features are different, for example, the collected image is only a gray image, and the image cannot be applied to optical smoke detection in the sea-sky background due to the fact that color information and the like cannot be used. In addition, the feature extraction and classification method in the existing smoke recognition algorithm has high calculation complexity, long calculation time and large occupied storage space. And the operation requirement of the existing smoke recognition algorithm cannot be met for a platform with limited hardware resources.
Disclosure of Invention
In view of the above analysis, the invention aims to disclose an optical smoke detection method combining priori knowledge and feature classification, and solve the technical problem of optical smoke detection in the sea-sky background.
The invention discloses an optical smoke detection method combining priori knowledge and feature classification, which comprises the following steps:
for a target image containing optical smoke interference under a sea-sky background, a smoke candidate area is segmented from the target image based on smoke gray scale characteristics;
carrying out morphological treatment on the smoke candidate areas, and screening out first-class smoke candidate areas by combining smoke shape information and smoke height information;
extracting multidimensional feature descriptors from the smoke candidate areas based on gray features, texture features and gradient features of the images, sending the multidimensional feature descriptors into a trained classifier to judge, and screening out second-class smoke candidate areas;
selecting an intersection of the first type smoke candidate region and the second type smoke candidate region as a confirmed smoke candidate region;
and carrying out edge extraction and fusion on the confirmed smoke candidate region to obtain the position information of the smoke region, thereby obtaining the final smoke detection result output.
Further, performing morphological treatment on the segmented smoke candidate region, performing target expansion in the horizontal direction, and closing the segmented smoke candidate region;
removing candidate areas which do not meet smoke characteristics from closed smoke candidate areas by combining prior knowledge including the shape and the height of smoke; and carrying out connected domain marking on the removed smoke candidate areas so as to obtain first-class smoke candidate areas.
Further, in screening the candidate areas for smoke of the second type,
the multi-dimensional feature descriptor contains 8-dimensional features, which are respectively a gray scale mean, a gray scale variance, a gray scale maximum, a gray scale differential mean, a contrast, an entropy, a vertical gradient mean, and a vertical gradient variance.
Further, the classification algorithm adopted by the classifier is a tree classification algorithm.
Further, the segmentation of the smoke candidate region comprises:
1) Acquiring gray scale estimated values of the sea surface and the sky according to the histogram information of the image, and acquiring initial seed points of the sea surface and the sky;
2) Acquiring an initial seed point of the smoke based on a white highlighting gray level statistical empirical value of the smoke area and a gray level maximum value of the image;
3) Determining initial seed points of sea surface, sky and smoke, clustering the image gray scale based on a clustering algorithm, and segmenting and extracting potential areas of the image belonging to a smoke target to obtain smoke candidate areas.
Further, smoothing the histogram and detecting peak points to obtain gray scale estimated values of the sea surface and the sky, and obtaining initial seed points of the sea surface and the sky;
the initial seed points of the sky and the sea surface are the first extreme point and the second extreme point of the histogram of the smoothed image;
the initial seed point of the smoke isWherein V is s For statistical empirical values of smoke gray scale, max (img) is the gray maximum of the image.
Further, the Kmeans gray level clustering algorithm is adopted to cluster the gray level of the image, the potential image area belonging to the smoke target is obtained, and after segmentation and extraction, the smoke candidate area is obtained.
Further, the method for extracting and fusing the edges of the confirmed smoke candidate areas comprises the following steps:
1) Extracting an edge map of smoke by adopting an edge detection algorithm;
2) Confirming smoke and supplementing a segmentation result based on the edge map information to obtain a segmentation map after fusing edges;
3) Carrying out morphological expansion on the segmentation map after the fusion of the edges to obtain a final edge feature fusion map;
4) And (5) carrying out connected domain marking on the edge feature fusion map so as to obtain the final smoke detection result output.
Further, counting the number of edge points in the corresponding position of the edge map for each confirmed smoke candidate area, determining the matching degree of the edge points and the smoke candidate areas, and judging that the current smoke candidate area is a smoke target area if the two areas are considered to have an intersection when the overlapping rate condition is met; and obtaining a union set from the clustered and segmented smoke candidate areas and the edge graphs corresponding to the clustered and segmented smoke candidate areas to obtain a segmented graph after fusing the edges.
Further, the segmentation map after the fusion edge:
in the formula, seg (i, j) is a cluster segmentation map, and Edge (i, j) is an Edge extraction map; match (m) is the matching degree of the edge map of the corresponding position of the mth smoke candidate region and the current mth smoke candidate region;
the overlay (m) is the overlapping rate of the edge map of the position corresponding to the m-th smoke candidate area and the current m-th smoke candidate area; m=1,..boxnum; boxnum is the total number of smoke candidate areas in the cluster segmentation map; i, j are the longitudinal and transverse coordinates of the image.
Specifically, the calculation formula of the matching degree is:
a calculation formula of the overlapping rate;
Overlap(m)＝box[m].w*box[m].h*β
the box [ m ] is coordinate information corresponding to an mth smoke candidate region in a clustering segmentation diagram represented by a rectangular frame, wherein the box [ m ] bottom, the box [ m ] top, the box [ m ] left and the box [ m ] right represent the coordinates of an upper boundary, a lower boundary, a left boundary and a right boundary of a detection frame respectively, and the box [ m ] w and the box [ m ] h represent the width and the height of the mth rectangular frame; beta is a settable scaling factor.
The invention can realize at least one of the following beneficial effects:
according to the optical smoke detection method combining priori knowledge and feature classification, the segmentation of smoke candidate areas is realized by adopting a gray level clustering algorithm, and the screening of the smoke candidate areas is performed by combining the priori knowledge including the shape and the height of smoke and the feature classification of the 8-dimensional feature descriptors, so that the accuracy rate of smoke detection is improved. And the determined smoke candidate areas are subjected to result judgment and fusion by combining edge features, so that a more accurate smoke target range is obtained.
The invention can realize smoke detection without using motion characteristics, and solves the technical problem that the motion characteristics can not be used for completing the smoke detection due to continuous change of image background information caused by continuous motion of an aircraft platform.
The algorithm adopted by the invention is simple and effective, occupies less hardware resources, has high running speed and is suitable for hardware platforms with limited resources.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, like reference numerals being used to refer to like parts throughout the several views.
Fig. 1 is a flow chart of an optical smoke detection method combining prior knowledge and feature classification in an embodiment of the invention.
Fig. 2 is a flowchart of a method for segmenting smoke candidate regions in an embodiment of the invention;
fig. 3 is a flowchart of a method for edge extraction and fusion of smoke candidate areas according to an embodiment of the present invention;
fig. 4 is a flowchart of a training process of a classifier based on feature classification in an embodiment of the invention.
Detailed Description
Preferred embodiments of the present invention are described in detail below with reference to the attached drawing figures, which form a part of the present application and, together with the embodiments of the present invention, serve to explain the principles of the invention.
One embodiment of the present invention discloses an optical smoke detection method combining prior knowledge and feature classification, as shown in fig. 1, comprising:
step S101, for a target image containing optical smoke interference under a sea-sky background, a smoke candidate area is segmented from the target image based on smoke gray scale characteristics;
step S102, performing morphological processing on the smoke candidate region, and screening out a first type of smoke candidate region by combining smoke shape information and smoke height information;
step S103, extracting multidimensional feature descriptors from the smoke candidate areas based on gray features, texture features and gradient features of the images, sending the multidimensional feature descriptors into a trained classifier to judge, and screening out second-class smoke candidate areas;
step S104, selecting an intersection of the first type smoke candidate area and the second type smoke candidate area as a confirmed smoke candidate area;
and step 105, carrying out edge extraction and fusion on the confirmed smoke candidate areas to obtain the position information of the smoke areas, thereby obtaining the final smoke detection result output.
In the application scenario of the embodiment, the image content under the sea-sky background is relatively single, and mainly includes the sea surface, the sky and the 3 partial areas of the target, and the 3 partial areas have larger difference in gray scale. For more accurate smoke segmentation, a clustering segmentation algorithm based on sea surface, sky and smoke 3 classification can be utilized.
In order to cluster images more quickly and accurately, it is necessary to acquire the initial segmentation points (i.e., seed points) of the sea surface, sky and smoke 3 targets more accurately, so that it is necessary to determine the approximate gray scale range of the sea surface, sky and smoke. For an image under a sea-sky background, because the background in the image is simple, the histogram has a bimodal characteristic, and clustering initial segmentation points of the sea surface and the sky can be obtained through detection of peak points of the histogram.
In step S101, a gray scale clustering algorithm is used to segment the smoke candidate region based on the gray scale highlighting feature of the smoke.
As shown in fig. 2, the segmentation of specific smoke candidate regions includes:
step S201, gray level estimated values of the sea surface and the sky are obtained according to histogram information of the image, and initial seed points of the sea surface and the sky are obtained;
preferably, the gray level estimated values of the sea surface and the sky are obtained by smoothing the histogram of the image and detecting peak points; the initial seed points of the sky and the sea surface are the first extreme point and the second extreme point of the histogram of the smoothed image.
When calculating the histogram of the image, since the histogram has burrs (i.e., local peaks), the histogram needs to be smoothed to eliminate the burrs, and the histogram smoothing mode can be selected from gaussian smoothing, mean smoothing, and the like.
Wherein, the Gaussian smoothing calculation formula of the histogramWherein h is i Representing the calculated image histogram, r is the smooth scale size, and σ is the standard deviation.
For the smoothed histogram, the first 2 extreme points of the histogram (i.e. the first 2 maximum peaks are obtained) are calculated, and the extreme point calculation formula is H max ＝max(H i-k ,…,H i ,…,H i+k ) I=0, …,255; wherein H is i The smoothed histogram is represented, and k is a scale interval in which the maximum value is obtained.
2 maximum values can be obtained according to an extreme point calculation formula, the prior knowledge of lower sea surface gray level and higher sky gray level is combined, the corresponding gray level is the gray level estimated value of the sea surface and the sky, the first extreme point (namely the maximum extreme point) is the sky initial seed point val_sky, and the second extreme point (namely the next maximum extreme point) is the sea surface initial seed point val_sea.
Step S202, acquiring an initial seed point of smoke based on a white highlight gray level statistical empirical value of a smoke area and a gray level maximum value of an image;
if smoke exists in the image, according to the imaging characteristics of the smoke, the smoke definitely exists in a white highlight region, and the gray value of the smoke region is basically V through statistical analysis of the white highlight gray range of the smoke region s Above (V) s Statistical value) by calculationAnd comparing the gray level corresponding to the maximum value with the statistic value of the smoke gray level, thereby confirming the seed point of the smoke. Combining the gray scale estimated values of the sea surface and the sky obtained by calculation in the step S201, and confirming that the initial seed points of the sea surface, the sky and the smoke 3 are as follows:
wherein V is s For statistical empirical values of smoke gray scale, max (img) is the gray maximum of the image.
Statistical empirical value of smoke gray scale V s By constructing a smoke image library from all the smoke images to be detected, calculating the gray average value V of the smoke areas in the image library s 。
Step S203, clustering the image gray scale by adopting a clustering algorithm based on the initial seed points of sea surface, sky and smoke, and segmenting and extracting the image potential area belonging to the smoke target to obtain a smoke candidate area.
Specifically, the Kmeans gray level clustering algorithm is adopted to cluster the gray level of the image, the potential image area belonging to the smoke target is obtained, and after segmentation and extraction, the smoke candidate area is obtained.
In this embodiment, since the initial seed point is obtained by calculation, the value is more accurate, and thus, an accurate clustering result can be obtained with fewer iterations. And according to the clustering result, segmenting and extracting the potential image area belonging to the smoke category to obtain a smoke candidate area.
Specifically, the screening of the first type smoke candidate area in step S102 includes:
performing morphological processing on the segmented smoke candidate region, and closing the segmented smoke candidate region;
preferably, when morphological processing is performed on the segmented smoke candidate region, the target expansion is performed only in the horizontal direction so as not to affect the height information of the target, thereby closing the segmented candidate region.
And removing candidate areas which do not meet the smoke characteristics in the closed smoke candidate areas by combining the prior knowledge including the shape and the height of the smoke in the smoke candidate areas subjected to morphological processing. For example, the candidate areas which do not satisfy the smoke characteristics are eliminated by excessively increasing or decreasing the height, excessively decreasing the number of pixels in the candidate areas, and forming straight edges which do not exist in the shape in which excessive smoke exists. And carrying out connected domain marking on the removed smoke candidate areas so as to obtain first-class smoke candidate areas.
Specifically, in step S103, in the screening of the second type of smoke candidate area, the color characteristic of the smoke target processed in the embodiment in the optical image is that there is a gray scale highlight area, the gray scale is gradually changed, the whole gray scale value is higher, the shape is changeable, the shape is similar to an ellipse, and the shape is similar to a cloud. Also, the texture of smoke is simpler than other objects such as boats, there is stronger directionality, the horizontal edges are weaker, and the hull of the boat object has stronger horizontal edges. For the smoke characteristic analysis, the gray level features, the texture features and the gradient features of the smoke image can be extracted, so that the characteristics for classification are finally obtained, the trained classifier is used for judging, and the second type smoke candidate region is screened out.
Specifically, the multidimensional feature descriptors extracted based on gray features are three-dimensional features of gray mean values, variances and maximum values of images;
based on the characteristics that the smoke image has high gray level and gradual gray level, and the whole gray level of the ship target is lower, the three-dimensional characteristics of the mean value, the variance and the maximum value of the image are mainly extracted on the gray level characteristics.
The mean value refers to the mean value of the image gray scale, the variance refers to the variance of the image gray scale value, and the maximum value refers to the maximum value of the image gray scale. The definition is as follows:
average value:
variance:
maximum value:
wherein M, N is the width and height of the image, P ij Is the gray value of the pixel.
Specifically, the multidimensional feature descriptors extracted based on texture features are three-dimensional features of gray level difference average value, contrast and entropy of the image;
texture features describe the surface properties of an image or object to which an image region corresponds, which quantify the characteristics of the gray level variations within the region.
The texture based on smog is simpler than the marine target, and the texture features are extracted by adopting a gray level difference method in the embodiment.
Specifically, a pixel point (m, n) in the image is set, and the pixel point and a neighborhood point thereofThe gray level difference of (2) is:
wherein, the liquid crystal display device comprises a liquid crystal display device,known as gray scale difference. Setting all possible values of the gray differential value to be m levels, and obtaining +.>Is a histogram of (a) of the image. By means of the histogram +.>The probability p (k) of the value, k is the gray level difference. A larger p (k) indicates a coarser texture and a smaller p (k) indicates a finer texture.
The texture features of the embodiment mainly extract three features of gray level difference average value, contrast and entropy. The definition is as follows:
gray level differential mean value:
contrast ratio:
entropy:
the multi-dimensional feature descriptors extracted based on the gradient features are two-dimensional gradient features of a vertical gradient mean value and a vertical gradient variance of the image;
based on the characteristics that the smoke image has stronger directivity, the horizontal edge is weaker, and the hull of the ship target has stronger horizontal edge, the gradient characteristic of the embodiment adopts the vertical gradient characteristic of the image. And calculating the vertical gradient mean value and the vertical gradient variance of the image to be used as extracted 2-dimensional gradient characteristics.
And a template for solving the vertical gradient adopts a sobel operator, as shown in a formula.
Obtaining a vertical gradient G y And the vertical gradient mean value and variance are obtained for the vertical gradient of the image, and the formula is as follows:
vertical gradient mean:
wherein G is y (i, j) is the vertical gradient of coordinate point (i, j); n is the image width;
vertical gradient variance:
the two-dimensional features of the vertical gradient mean and variance are taken as gradient features.
The finally extracted multidimensional feature descriptors comprise 8-dimensional features, namely a gray average value, a gray variance, a gray maximum value, a gray differential average value, contrast, entropy, a vertical gradient average value and a vertical gradient variance.
And respectively extracting the 8-dimensional feature descriptors for each suspected smoke target area block, sending the 8-dimensional feature descriptors into a trained classifier to obtain a judging result of the classifier, wherein the judging method is that the target output is 1, the non-target output is 0, and the target area identified as smoke is taken as a second type smoke candidate area.
In step S104, the intersection of the first type smoke candidate region and the second type smoke candidate region is selected as the confirmed smoke candidate region.
In the step, the accuracy of the smoke candidate region is further improved and the accuracy of smoke detection is improved by combining prior knowledge including the shape and the height of the smoke and the feature classification of the 8-dimensional feature descriptors.
In this embodiment, in the smoke segmentation method based on gray clustering, since information such as the boundary and contrast of the target is not considered, a false alarm may occur when the background of the transitional exposure sky of the camera is bright. Meanwhile, as the smoke is in different forms, the gray level of the smoke area is unevenly changed, the phenomenon that the gray level of the highlight part of the partial area is low is presented, and the situation that the same target is divided into a plurality of sub-blocks can be caused by simply using a clustering and dividing algorithm. For the above reasons, step S105 of the present embodiment confirms the result after cluster division and further supplements and communicates the division result.
Specifically, as shown in fig. 3, the method for extracting and fusing edges of the identified smoke candidate region in step S105 includes:
step S301, extracting an edge map of smoke by adopting an edge detection algorithm;
since the interfering optical smoke usually only appears in the vicinity of the sky and sea-sky, the contrast of the smoke to the background is high, and there is edge information. Thus, an edge detection algorithm (e.g., canny edge extraction method) is used to extract edges from the image to obtain an edge map of the smoke.
Specifically, when the edges of the image are extracted, the hough transformation is adopted to extract the sea-sky-line, and after the extracted sea-sky-line is removed, the rest edges are used as the edge map of the smog.
Step S302, confirming smoke and supplementing a segmentation result based on the edge map information to obtain a segmentation map after fusion of edges;
counting the number of edge points in the corresponding position of the edge map for each smoke candidate area, determining the matching degree of the edge points and the smoke candidate areas, and judging that the current smoke candidate area is a smoke target area if the two areas are considered to have an intersection when the overlapping rate condition is met; and obtaining a union set from the clustered and segmented smoke candidate areas and the edge graphs corresponding to the clustered and segmented smoke candidate areas to obtain a segmented graph after fusing the edges.
Specifically, the segmentation map after the edge is fused:
in the formula, seg (i, j) is a cluster segmentation map, and Edge (i, j) is an Edge extraction map; match (m) is the matching degree of the edge map of the corresponding position of the mth smoke candidate region and the current mth smoke candidate region;
overlay (m) is the overlapping rate of the edge map at the position corresponding to the mth smoke candidate region and the current mth smoke candidate region; m=1,..boxnum; boxnum is the total number of smoke candidate areas in the cluster segmentation map; i, j are the longitudinal and transverse coordinates of the image.
Specifically, the calculation formula of the matching degree is:
a calculation formula of the overlapping rate;
Overlay(m)＝box[m].w*box[m].h*β
the box [ m ] is coordinate information corresponding to an mth smoke candidate region in a cluster segmentation diagram represented by a rectangular frame, wherein the box [ m ] bottom, the box [ m ] top, the box [ m ] left and the box [ m ] right represent the coordinates of an upper boundary, a lower boundary, a left boundary and a right boundary of a detection frame respectively, and the box [ m ] w and the box [ n ] h represent the width and the height of the mth rectangular frame; beta is a settable scaling factor.
Step S303, carrying out morphological expansion on the segmented graph after the edge fusion to obtain a final edge feature fusion graph;
and step S304, carrying out connected domain marking on the edge feature fusion map so as to obtain the final smoke detection result output.
Preferably, the embodiment also discloses a training process of the classifier based on the feature classification; as shown in FIG. 4 comprising
Step S401, a sample library of smoke targets, other targets and background targets for classifier learning is established;
aiming at all training images containing smoke, the smoke image blocks are extracted to construct a smoke image as a positive sample library, and other targets (such as ships and the like) and background targets are extracted as negative sample libraries.
Specifically, the color characteristic of the smoke target processed in the embodiment in the optical image is that there is a gray scale highlight region, the gray scale is graded, and the overall gray scale value is higher. The shape is changeable, has a shape similar to an ellipse, and also has a shape similar to a cloud. Also, the texture of smoke is simpler than other objects such as boats, there is stronger directionality, the horizontal edges are weaker, and the hull of the boat object has stronger horizontal edges. For the above-described smoke characteristic analysis, gray features, texture features, and gradient features of the smoke image may be extracted as features that are ultimately used for classification.
Step S402, extracting multidimensional feature descriptors from all samples in a sample library based on gray features, texture features and gradient features of the image;
the multi-dimensional feature descriptors extracted for each sample in the sample library contain 8-dimensional features, namely a gray scale mean, a gray scale variance, a gray scale maximum, a gray scale difference mean, a contrast, an entropy, a vertical gradient mean and a vertical gradient variance.
Step S403, training a classifier by adopting a multi-dimensional feature descriptor extracted from all samples, so that the classifier can identify a smoke target;
the classification algorithm may employ a tree classification algorithm; such as CART binary tree classification algorithms, etc. The tree classification algorithm is used for realizing rapid classification, has low complexity, occupies less hardware resources, has high running speed and is suitable for hardware platforms with limited resources. And obtaining the smoke target classifier after training.
In summary, the optical smoke detection method combining priori knowledge and feature classification disclosed by the embodiment of the invention adopts the gray level clustering algorithm to realize the segmentation of smoke candidate areas, and the accuracy of smoke detection is improved by combining the priori knowledge including the shape and the height of smoke and the feature classification of the 8-dimensional feature descriptors to screen the smoke candidate areas. And the determined smoke candidate areas are subjected to result judgment and fusion by combining edge features, so that a more accurate smoke target range is obtained.
The embodiment of the invention can realize smoke detection without using the motion characteristics, and solves the technical problem that the motion characteristics cannot be used for completing the smoke detection due to the continuous change of the image background information caused by the continuous motion of the aircraft platform.
The algorithm adopted by the embodiment of the invention is simple and effective, occupies less hardware resources, has high running speed and is suitable for hardware platforms with limited resources.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.

Claims (10)

1. An optical smoke detection method combining prior knowledge and feature classification, comprising:
for a target image containing optical smoke interference under a sea-sky background, a smoke candidate area is segmented from the target image based on smoke gray scale characteristics;
carrying out morphological treatment on the smoke candidate areas, and screening out first-class smoke candidate areas by combining smoke shape information and smoke height information;
extracting multidimensional feature descriptors from the smoke candidate areas based on gray features, texture features and gradient features of the images, sending the multidimensional feature descriptors into a trained classifier to judge, and screening out second-class smoke candidate areas;
selecting an intersection of the first type smoke candidate region and the second type smoke candidate region as a confirmed smoke candidate region;
and carrying out edge extraction and fusion on the confirmed smoke candidate region to obtain the position information of the smoke region, thereby obtaining the final smoke detection result output.
2. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 1,
morphological processing is carried out on the segmented smoke candidate areas, target expansion is carried out in the horizontal direction, and the segmented smoke candidate areas are closed;
removing candidate areas which do not meet smoke characteristics from closed smoke candidate areas by combining prior knowledge including the shape and the height of smoke; and carrying out connected domain marking on the removed smoke candidate areas so as to obtain first-class smoke candidate areas.
3. The method of claim 1, wherein, in screening the second type of smoke candidate region,
the multi-dimensional feature descriptors contain 8-dimensional features, which are respectively gray-scale mean, gray-scale variance, gray-scale maximum, gray-scale differential mean, contrast, entropy, vertical gradient mean, and vertical gradient variance.
4. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 3,
the classification algorithm adopted by the classifier is a tree classification algorithm.
5. The method of claim 1, wherein the segmentation of the smoke candidate region comprises:
1) Acquiring gray scale estimated values of the sea surface and the sky according to the histogram information of the image, and acquiring initial seed points of the sea surface and the sky;
2) Acquiring an initial seed point of the smoke based on a white highlighting gray level statistical empirical value of the smoke area and a gray level maximum value of the image;
3) Based on the sea surface, sky and initial seed points of smoke, clustering the image gray scale by adopting a clustering algorithm, and segmenting and extracting potential areas of the image belonging to a smoke target to obtain smoke candidate areas.
6. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 5,
smoothing the histogram and detecting peak points to obtain gray level estimated values of the sea surface and the sky, and obtaining initial seed points of the sea surface and the sky;
the initial seed points of the sky and the sea surface are the first extreme point and the second extreme point of the histogram of the smoothed image;
the initial seed point of the smoke isWherein V is s For statistical empirical values of smoke gray scale, max (img) is the gray maximum of the image.
7. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 6,
and clustering the gray level of the image by adopting a Kmeans gray level clustering algorithm to obtain a potential image area belonging to a smoke target, and obtaining a smoke candidate area after segmentation and extraction.
8. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 1,
the method for extracting and fusing the edges of the confirmed smoke candidate areas comprises the following steps:
1) Extracting an edge map of smoke by adopting an edge detection algorithm;
2) Confirming smoke and supplementing a segmentation result based on the edge map information to obtain a segmentation map after fusing edges;
3) Carrying out morphological expansion on the segmentation map after the fusion of the edges to obtain a final edge feature fusion map;
4) And (5) carrying out connected domain marking on the edge feature fusion map so as to obtain the final smoke detection result output.
9. The method for optical smoke detection combining a priori knowledge and feature classification as set forth in claim 5,
counting the number of edge points in the corresponding position of the edge map for each confirmed smoke candidate region, determining the matching degree of the edge points and the smoke candidate region, and judging that the current smoke candidate region is a smoke target region if the matching degree of the edge points and the smoke candidate region is considered to have an intersection when the overlapping rate condition is met; and obtaining a union set from the clustered and segmented smoke candidate areas and the edge graphs corresponding to the clustered and segmented smoke candidate areas to obtain a segmented graph after fusing the edges.
10. The method of optical smoke detection combining a priori knowledge and feature classification as set forth in claim 9,
the segmentation map after the fusion edge comprises:
in the formula, seg (i, j) is a cluster segmentation map, and Edge (i, j) is an Edge extraction map; match (m) is the matching degree of the edge map of the corresponding position of the mth smoke candidate region and the current mth smoke candidate region;
the overlay (m) is the overlapping rate of the edge map of the position corresponding to the m-th smoke candidate area and the current m-th smoke candidate area; m=1,..boxnum; boxnum is the total number of smoke candidate areas in the cluster segmentation map; i, j are the longitudinal and transverse coordinates of the image.
Specifically, the calculation formula of the matching degree is:
a calculation formula of the overlapping rate;
Overlap(m)＝box[m].w*box[m].h*β
the box [ m ] is coordinate information corresponding to an mth smoke candidate region in a clustering segmentation diagram represented by a rectangular frame, wherein the box [ m ] bottom, the box [ m ] top, the box [ m ] left and the box [ m ] right represent the coordinates of an upper boundary, a lower boundary, a left boundary and a right boundary of a detection frame respectively, and the box [ m ] w and the box [ m ] h represent the width and the height of the mth rectangular frame; beta is a settable scaling factor.
CN202210108504.XA 2022-01-28 2022-01-28 Optical smoke detection method combining priori knowledge and feature classification Pending CN116563659A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210108504.XA CN116563659A (en) 2022-01-28 2022-01-28 Optical smoke detection method combining priori knowledge and feature classification

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210108504.XA CN116563659A (en) 2022-01-28 2022-01-28 Optical smoke detection method combining priori knowledge and feature classification

Publications (1)

Publication Number Publication Date
CN116563659A true CN116563659A (en) 2023-08-08

Family

ID=87493429

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210108504.XA Pending CN116563659A (en) 2022-01-28 2022-01-28 Optical smoke detection method combining priori knowledge and feature classification

Country Status (1)

Country Link
CN (1) CN116563659A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116977327A (en) * 2023-09-14 2023-10-31 山东拓新电气有限公司 Smoke detection method and system for roller-driven belt conveyor

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116977327A (en) * 2023-09-14 2023-10-31 山东拓新电气有限公司 Smoke detection method and system for roller-driven belt conveyor
CN116977327B (en) * 2023-09-14 2023-12-15 山东拓新电气有限公司 Smoke detection method and system for roller-driven belt conveyor

Similar Documents

Publication Publication Date Title
US11922615B2 (en) Information processing device, information processing method, and storage medium
CN106651872B (en) Pavement crack identification method and system based on Prewitt operator
CN107610114B (en) optical satellite remote sensing image cloud and snow fog detection method based on support vector machine
CN113538433B (en) Mechanical casting defect detection method and system based on artificial intelligence
CN109961049B (en) Cigarette brand identification method under complex scene
JP7113657B2 (en) Information processing device, information processing method, and program
US8340420B2 (en) Method for recognizing objects in images
CN109427055B (en) Remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy
CN109242870A (en) A kind of sea horizon detection method divided based on image with textural characteristics
CN108985170A (en) Transmission line of electricity hanger recognition methods based on Three image difference and deep learning
CN110717896A (en) Plate strip steel surface defect detection method based on saliency label information propagation model
CN110175556B (en) Remote sensing image cloud detection method based on Sobel operator
CN112580447B (en) Edge second-order statistics and fusion-based power line detection method
CN107742113A (en) One kind is based on the posterior SAR image complex target detection method of destination number
CN110689003A (en) Low-illumination imaging license plate recognition method and system, computer equipment and storage medium
CN110348307B (en) Path edge identification method and system for crane metal structure climbing robot
CN111091071A (en) Underground target detection method and system based on ground penetrating radar hyperbolic wave fitting
CN111476074A (en) Human body foreign matter detection method based on millimeter wave image
CN116563659A (en) Optical smoke detection method combining priori knowledge and feature classification
CN107977608B (en) Method for extracting road area of highway video image
CN109961065B (en) Sea surface ship target detection method
CN111695373A (en) Zebra crossing positioning method, system, medium and device
CN116563591A (en) Optical smoke detection method based on feature extraction under sea-sky background
CN107704864B (en) Salient object detection method based on image object semantic detection
CN116228757B (en) Deep sea cage and netting detection method based on image processing algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination