CN110675588B

CN110675588B - Forest fire detection device and method

Info

Publication number: CN110675588B
Application number: CN201910942744.8A
Authority: CN
Inventors: 韦海成; 王生营; 胡文锐; 何艳茹; 许洋; 肖明霞
Original assignee: North Minzu University
Current assignee: North Minzu University
Priority date: 2019-09-30
Filing date: 2019-09-30
Publication date: 2021-06-01
Anticipated expiration: 2039-09-30
Also published as: CN110675588A

Abstract

The invention provides a forest fire detection device and a forest fire detection method, wherein the forest fire detection device comprises an image acquisition module, a fire detection module and a control module, wherein the image acquisition module is used for acquiring an original image and then sending the original image to the fire detection module for processing; the fire detection module is used for converting the original image from the RGB color space to the HSV color space to obtain an HSV image, then performing a region feature clustering algorithm and a sample entropy algorithm on the HSV image, and calculating to obtain whether a real fire exists in the image; and the fire alarm module is used for alarming when a real fire disaster exists in the image. Compared with the existing forest fire detection method which focuses on video sequence difference for analysis, the method has the advantages that the data volume processed by the algorithm is small, multi-frame complex video comparison is not needed, the data transmission and calculation amount are reduced, the identification efficiency and accuracy are improved, and the method is more suitable for field large-scale environment detection and long-distance high-speed image data transmission.

Description

Forest fire detection device and method

Technical Field

The invention relates to the technical field of fire detection, in particular to a forest fire detection device and method.

Background

Forest fires are the most common form of forest disasters, cause loss of a large amount of forest resources, personnel and property every year, and cause great damage to the ecological environment and economic development. The forest fire has the characteristics of strong burst, wide spread range, difficult disposal and the like. However, because the forest environment is complex, false detection or missed detection can be caused in the process of fire detection; in addition, because the forest fire has strong burst property and can quickly spread, all-weather and large-range real-time detection needs to be carried out on the forest environment, and the data transmission and the calculation amount are huge. Therefore, it is necessary to design a device for effectively recognizing a fire in a complex forest environment.

At present, the commonly used fire detection method mainly adopts the analysis of a video sequence image to judge whether a fire occurs or not. Flame flicker detection method combining the color feature of fire with time variation and wavelet and statistical frequency is adopted to detect flame through the difference of fire image sequence; a multi-feature fusion video flame detection method is adopted, a Gaussian mixture model is used for extracting a foreground object, and then flame flicker recognition algorithm of flame color filtering and statistical technology is adopted to distinguish real flame.

However, the forest fire detection method focusing on the analysis of difference in video sequence images has a large data volume processed by an algorithm, and is not suitable for field large-scale environment detection and long-distance high-speed video data transmission. The algorithm for recognizing forest fires by adopting the graphic imaging method has limitation on the setting of cutting or recognition threshold values, is difficult to solve the problems of missing detection and false detection, and limits the application of the image recognition algorithm in the aspect.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, and provides a forest fire detection device and a forest fire detection method, which can detect a fire more quickly and accurately.

In order to achieve the above object, the embodiments of the present invention provide the following technical solutions:

a forest fire detection apparatus comprising:

the image acquisition module is used for acquiring an original image and then sending the original image to the fire detection module for processing;

the fire detection module is used for converting the original image from an RGB color space to an HSV color space to obtain an HSV image, then performing a region feature clustering algorithm and a sample entropy algorithm on the HSV image, and calculating to obtain whether a real fire exists in the image;

and the fire alarm module is used for alarming when a real fire disaster exists in the image.

A forest fire detection method is realized by adopting the device, and comprises the following steps:

step S2: segmenting the HSV image by adopting a characteristic region clustering algorithm, setting a threshold value of an HSV color space according to fire color characteristics, and extracting a suspected fire region from the HSV image;

step S3: calculating a sample entropy value of the threshold clustering region based on a K-Means clustering algorithm;

step S4: and setting a sample entropy threshold, judging the sample entropy calculated in the step S3, and judging whether the threshold clustering region is a real fire or not.

Compared with the existing forest fire detection method which focuses on video sequence difference for analysis, the method has the advantages that the data volume processed by the algorithm is small, and the method is more suitable for field large-scale environment detection and remote high-speed image data transmission.

Furthermore, in order to better implement the invention, the method also comprises the following steps:

step S1: the image acquisition module converts the acquired original RGB color space image into an HSV color space image to obtain the HSV image.

Setting a threshold value of an HSV color space according to the color characteristics of the fire to extract a suspected fire area, and compared with the method for distinguishing the fire area by using an RGB color space, under a large field space, the saturation and brightness of the fire area, sky and trees are greatly different; the distribution characteristics of the image are counted in the HSV color space and used for dividing the flame color characteristics, and the effect is better than that of the RGB color space.

Furthermore, in order to better realize the invention, the characteristic region clustering algorithm adopts a K-Means clustering algorithm.

Furthermore, in order to better implement the present invention, the step of segmenting the HSV image by using the feature region clustering algorithm, setting the threshold of the HSV color space according to the fire color feature, and extracting the suspected fire region from the HSV image includes:

step S2-1: setting K pixel points as initial central points of clustering according to the HSV image;

step S2-2: calculating Euclidean distances of S and V components between the central point of the cluster and the image pixel points, so that the clusters with high similarity of the characteristic values of the image pixel points are formed into a class;

step S2-3: correcting the clustering center point, calculating the mean value of the cluster, and updating the clustering center point according to the mean value of the cluster;

step S2-4: calculating Euclidean distances of S and V components between the new clustering center point and the image pixel points, and re-clustering into k categories;

step S2-5: judging whether the k categories reach the convergence condition of the standard measure function, if so, carrying out the next step, otherwise, repeating the step S2-3;

step S2-6: judging whether the category set meets a threshold condition, outputting the result meeting the threshold condition as a threshold clustering region set result, and rejecting the category set not meeting the threshold condition.

Clustering the HSV images by adopting a K-Means clustering algorithm to obtain a plurality of threshold value clustering area sets, namely sets of suspected fire areas on one image, and then rejecting category sets which do not meet threshold value conditions, namely sets which are not suspected fire areas.

Furthermore, in order to better implement the present invention, the step of calculating the sample entropy value of the threshold clustering region based on the K-Means clustering algorithm includes: and calculating the plurality of threshold value clustering region subsets by using a sample entropy value algorithm based on the plurality of threshold value clustering region subsets obtained by the K-Means clustering algorithm to obtain the entropy value of each subset.

Because the extracted threshold clustering region set calculated in step S2 may be misjudged to have a fire-like region, sample entropy calculation is performed under the K-Means clustering algorithm to obtain the entropy of each threshold clustering region.

Furthermore, in order to better implement the present invention, the step of setting the sample entropy threshold X, determining the sample entropy calculated in step S3, and determining whether the threshold clustering region is a real fire includes: setting the threshold value of the sample entropy as X, detecting a fire disaster if the entropy of the subset calculated in the step S3 is larger than X, and performing fire alarm; and if the entropy of the subset is less than X, detecting that the fire is not a fire, and not performing fire alarm.

Setting the size of an entropy threshold value X in different scenes, and comparing the entropy value of each threshold clustering region obtained in the step S3 with the entropy threshold value X, wherein the threshold clustering region larger than the entropy threshold value X is a real fire, otherwise, the threshold clustering region is a fire-like fire; if the fire disaster exists on one image, the alarm is given.

Compared with the prior art, the invention has the beneficial effects that:

According to the method, a suspected fire area is extracted by setting a threshold value of an HSV color space according to the color characteristics of the fire, and compared with the method for distinguishing the fire area by using an RGB color space, the saturation and brightness of the fire area, sky and trees are greatly different in the field large space; the distribution characteristics of the image are counted in the HSV color space and used for dividing the flame color characteristics, and the effect is better than that of the RGB color space.

The method carries out sample entropy calculation on the threshold value clustering region based on the K-Means clustering algorithm, and judges the image complexity of the image region by adopting the sample entropy because the signal of the flame is complex and has no obvious rule, so that the entropy of the image of the fire region is much larger than that of the surrounding artificial buildings or trees, and whether the fire region exists in the image can be judged more accurately.

The invention overcomes the defect of fire identification by simply using fire color characteristics, can accurately extract a flame area, and has the advantages of strong anti-interference capability and strong applicability; particularly, the K-Means clustering algorithm and the sample entropy algorithm used by the invention adopt static image processing, multi-frame complex video comparison is not needed, the data transmission and the operation amount are reduced, and the identification efficiency and accuracy are improved; the method can play a role in algorithm optimization of the projects of the forest fire monitoring and unmanned aerial vehicle fire inspection system by adopting an image recognition method.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a block diagram of a forest fire detection apparatus according to the present invention;

FIG. 2 is a flow chart of the forest fire detection method of the present invention;

FIG. 3 is an original image acquired in an embodiment of the present invention;

FIG. 4 is a flow chart of the K-Means clustering algorithm of the present invention;

FIGS. 5, 6, and 7 are threshold clustering regions of the original image after calculation processing by the K-Means clustering algorithm.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.

Example 1:

the invention is realized by the following technical scheme, as shown in figure 1, the forest fire detection device comprises an image acquisition module, a fire detection module and a fire alarm module. The image acquisition module is connected with the fire detection module and used for acquiring an original image and transmitting the original image to the fire detection module; the fire detection module is borne by a data processing chip, and whether a real fire exists in the acquired image is calculated and identified based on a sample entropy value under a K-Means clustering algorithm so as to realize a forest fire detection function; and if the real fire is detected, the fire alarm module gives an alarm.

The image acquisition module sends the acquired original image to the fire detection module, and the fire detection module firstly converts the RGB color space of the original image into HSV color space to obtain an HSV image; performing K-Means clustering initialization segmentation on the HSV image, setting a threshold value of H, S, V, and extracting a suspected fire area; dividing the edge of the suspected fire area, cutting the edge into a plurality of subsets, and calculating the entropy value of each subset by using the sample entropy value based on the K-Means clustering algorithm; setting an entropy threshold value, and judging whether a real fire disaster exists in the area of the subset; and if the real fire exists, the fire alarm module gives an alarm.

Based on the detection device, a forest fire detection method is provided, as shown in fig. 2, which specifically comprises the following steps:

The model of the HSV color space image is similar to a conical structure, and in each coordinate axis of the HSV color space, a horizontal X axis represents the tone of the image and is measured by H; the horizontal Y-axis represents the saturation of the image colors, measured by S; the vertical Z-axis represents the color brightness of the image, measured as V. Under a large field space, fire areas, sky and trees have large difference in saturation and brightness, HSV color space is used for counting distribution characteristics of images, and the distribution characteristics are used for dividing flame color characteristics to be better than RGB color space. The conversion formula for converting the image from the RGB color space to the HSV color space is as follows:

V＝max(R`,G`,B`)

wherein, R ', G ', B ' is the normalized result of R, G, B, namely:

step S2: and segmenting the HSV image by adopting a characteristic region clustering algorithm, setting a threshold value of an HSV color space according to the fire color characteristic, and extracting a suspected fire region from the HSV image.

The areas suspected of fire have larger difference in saturation and brightness with other areas, and compared with other areas on the image, the areas suspected of fire are highlighted and extracted by using a characteristic area clustering algorithm. The suspected fire area can be effectively extracted by adopting the characteristic area clustering algorithm, but the characteristic area clustering algorithm belongs to the crude extraction process of the image characteristic area, in the embodiment, as shown in fig. 4, the K-Means clustering algorithm is preferentially adopted in the characteristic area clustering algorithm:

Specifically, the threshold setting of the HSV color space is to set a hue H value [0,0.167], a saturation S value [0.7,1], a brightness V value [0.7,1] according to the color, saturation and brightness of the flame, select a suspected fire area through three set thresholds, and identify the suspected fire area in an original image.

After the image is segmented by the K-Means clustering algorithm, a set of pixels with a threshold value smaller than 20 points is removed as an isolated area, as shown in fig. 3, an acquired original image is segmented by the K-Means clustering algorithm to form three image subsets, and the original image (fig. 5a) is clustered to form a flame subregion (fig. 5b), a forest subregion (fig. 5c) and a background subregion (fig. 5 d). It should be noted that, in the actual use process of the fire detection scene, the threshold value is strictly limited by using the clustering and threshold value algorithm, and in this embodiment, the value of the threshold value is 20, and an incorrect selection of the threshold value may cause image missing or false detection.

It should be noted that after an image is subjected to a K-Means clustering algorithm, the image can be divided into a plurality of threshold value clustering area subsets, namely a plurality of suspected fire images, and a plurality of non-threshold value clustering area subsets, namely images which are not fire, and then the threshold value clustering area subsets are extracted for further calculation, and the non-threshold value clustering area subsets are removed.

The selected threshold can quickly identify the fire area (fig. 6b) when processing the original image (fig. 6a), and as shown in fig. 6c, when the suspected fire area is extracted by using the K-Means clustering algorithm and the threshold method, the tent camping in the forest can be misjudged as the suspected fire area because the H-component range contains many red other objects, such as firefighter's clothing, fire fighting vehicles, red flowers, houses, even insects, and the like, the number of pixel points in the image conforming to the H range can also affect the identification of the fire area, the shape, brightness, color, and other characteristics of the tent are very similar to those of a real fire, and the misjudgment can occur if only the color characteristic and the shape characteristic are identified in fig. 6 d. The threshold margin of the image is very small and varies from image to image, and the practicability is affected, so that the threshold clustering algorithm needs to be improved in the next step to improve the identification rate of the fire image.

Step S3: and calculating the sample entropy value of the threshold clustering region based on a K-Means clustering algorithm.

The fire identification is improved and perfected on the threshold value clustering in the step S2 based on sample entropy value calculation under the K-Means clustering algorithm, and sample entropy value analysis is carried out on a plurality of threshold value clustering region subsets judged by the threshold value clustering algorithm to judge whether the fire regions exist in the images. As shown in fig. 7a, after threshold clustering, two area subsets (fig. 7b) are both suspected fire areas, namely real fire and firefighter clothing, and the dimension of the two suspected fire area images is reduced, and then sample entropy calculation is performed to analyze whether the suspected fire areas are real fire.

The basis for calculating the sample entropy value of the threshold clustering region is as follows: the sample entropy measures the time sequence complexity mainly by measuring the probability of generating a new pattern in a signal, when the probability of generating the new pattern is higher, the more complex the signal is, the higher the value of the sample entropy is, when the value of the sample entropy is lower, the higher the sequence self-similarity is, the simpler the signal is, and the smaller the value of the sample entropy is. The fire image is analyzed to find that the image texture, the details and the edges of the fire area are obviously different from those of the non-fire area. For example, the gray value distribution of the pixel points in the fire area shows complex changes, while the gray value color change of the pixel points in the fire-like area is relatively fixed and the gray value change is relatively single. Therefore, the image complexity of the image area is judged by adopting the sample entropy, so that whether the fire area exists in the image can be judged accurately.

And setting the sample entropy value threshold value X to be 0.85, when the sample entropy value of the suspected fire area is larger than the threshold value, detecting the suspected fire area as a fire area, and performing fire alarm, otherwise, when the sample entropy value of the suspected fire area is smaller than the threshold value, detecting the suspected fire area as a non-fire area, and continuing to perform detection. If the entropy value of the clothes of the firefighters in the threshold clustering region subset calculated by the graph is less than 0.85, rejecting the subset of the clothes of the firefighters; but the entropy value of the fire area subset is larger than 0.85, the image is indicated to have a real fire, and an alarm is required to be given.

The system of the fire detection device is subjected to stability test, and the stability of the system is the fundamental guarantee for supporting whether the whole system can normally operate. The verification is carried out by selecting 30 typical forest fire images and 30 similar fire images with interference for simulation analysis.

In the verification process, a suspected fire area in the image is extracted by using a threshold value clustering algorithm, sample entropy calculation is carried out on the extracted suspected fire area, whether the image is a real fire or not is judged, and the detection result is shown in table 1:

index (I)	Fire area
		Total number of images	60
Identifying the number of errors	2
		Identifying the correct quantity	58
Rate of correct recognition/%)	96.67％
		Run time/s	16.03

TABLE 1

Wherein, when 30 forest fire images are used for fire identification, 28 forest fire images are used for detecting fire and alarming; no alarm is given when 30 similar fire images with interference are detected as no fire images. The accuracy of the identification of all 60 images is 96.67%, but the error is in a reasonable range, and the stability requirement of the system is met.

In summary, the invention provides a forest fire detection device and method aiming at the problems of missing detection, false detection and the like in the existing image recognition forest fire, a characteristic region clustering algorithm and a sample entropy value calculation method are used, firstly, the color gamut space conversion is carried out on the collected forest fire image, namely, the RGB color space is converted into the HSV color space, and the influence of visual deviation in the image recognition process is reduced; then, performing subset clustering on a suspected fire area appearing in the image by adopting a K-Means clustering algorithm of a characteristic area clustering algorithm and according to an HSV color space component distance criterion; on the basis, the clustered region subset weight is distinguished through a sample entropy algorithm, the statistical difference of entropy values of a real fire region and a fire-like region is distinguished, and whether a real fire exists in a suspected fire region subset screened out by clustering or not is confirmed. The device of the invention adopts static image processing, does not need multi-frame complex video comparison, reduces data transmission and calculation amount, improves the identification efficiency and accuracy, and meets the requirements of fire detection reliability and real-time performance.

The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A forest fire detection device, its characterized in that: the method comprises the following steps:

the fire detection module is used for converting the original image from the RGB color space to the HSV color space to obtain an HSV image, then performing a region feature clustering algorithm and a sample entropy algorithm on the HSV image, and calculating to obtain whether a real fire exists in the image;

the fire alarm module is used for alarming when a real fire disaster exists in the image;

the fire detection module calculates the threshold clustering region subsets by using a sample entropy algorithm based on the threshold clustering region subsets obtained by the K-Means clustering algorithm to obtain the entropy of each subset; setting a threshold value of the sample entropy value as X, and detecting a fire when the entropy value of the subset is greater than X; and if the entropy value of the subset is less than X, detecting as a non-fire.

2. A forest fire detection method implemented using the apparatus of claim 1, characterized in that: the method comprises the following steps:

step S2: the method comprises the steps of adopting a characteristic region clustering algorithm to segment HSV images, setting thresholds of HSV color spaces according to fire color characteristics, and extracting suspected fire regions from the HSV images;

step S4: setting a sample entropy threshold value X, judging the sample entropy calculated in the step S3, and judging whether a threshold value clustering region is a real fire;

the step of calculating the sample entropy value of the threshold clustering region based on the K-Means clustering algorithm comprises the following steps: calculating the plurality of threshold clustering region subsets by using a sample entropy algorithm based on the plurality of threshold clustering region subsets obtained by the K-Means clustering algorithm to obtain an entropy value of each subset;

the step of setting a sample entropy threshold value X, determining the sample entropy calculated in step S3, and determining whether a threshold clustering region is a real fire, includes:

setting the threshold value of the sample entropy as X, detecting a fire disaster if the entropy of the subset calculated in the step S3 is larger than X, and performing fire alarm; and if the entropy of the subset is less than X, detecting that the fire is not a fire, and not performing fire alarm.

3. A forest fire detection method as claimed in claim 2, wherein: further comprising the steps of:

4. A forest fire detection method according to claim 1, characterised in that: the characteristic region clustering algorithm adopts a K-Means clustering algorithm.

5. A forest fire detection method as claimed in claim 4, wherein: the steps of adopting a characteristic region clustering algorithm to segment the HSV image, setting a threshold value of an HSV color space according to fire color characteristics, and extracting a suspected fire region from the HSV image comprise the following steps: