JP2012203582A - Determination method, determination device, determination system and determination program of vegetation region in color image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a determination method which can properly determine harmless vegetation areas (lawn areas) and harmful vegetation areas in a vegetation region by reducing an influence of data acquisition conditions such as seasonal variation, sunlight and the like and extracting the vegetation region from a color image.SOLUTION: A determination method of a vegetation region in a color image includes: a step for extracting pixels composing a vegetation region by using magnitude correlations between RGB values of respective pixels in a color image; a step for calculating an R/G value by dividing an R component value by a G component value for each pixel in the extracted vegetation region; a step for creating histogram by using the R/G values; a step for classifying the vegetation region as a single vegetation region or a mixed vegetation region by using a static obtained from the created histogram; a step for, when the vegetation region is classified as the mixed vegetation region, determining an area composed of pixels having R/G values less than a threshold in the vegetation region as a harmful vegetation area.

Description

  The present invention extracts a vegetation region from a color image and can determine a harmless vegetation region (turf region) and a harmful vegetation region in the vegetation region, a determination method, a determination device, a determination system, and a determination program About.

  The river embankment protects the property near the basin by preventing water from flowing out of the river during floods. The slope of the river levee may be covered with turf, and the slope of the river levee is firmly fixed by the turf to prevent the river levee from being destroyed by running water during a flood. On the other hand, harmful vegetation such as weeds grows naturally on the river bank. If the harmful vegetation grows, the lawn will die and the strength of the slope of the river dike will be reduced. For this reason, it is necessary to periodically carry out weeding for harmful vegetation on river dikes. The criteria for weeding are the type of vegetation and the degree of growth (area, height, etc.). At present, weeding decisions are made visually and are a burden on river embankment managers.

  On the other hand, a river management camera may be installed on the river bank. Thereby, it is possible to acquire a still image at an arbitrary point and time. Currently, such images are used for remote monitoring of rivers, and the vegetation situation can also be taken. For this reason, performing image processing on the acquired images, automatically collecting river levee conditions, and quantitatively evaluating harmful vegetation can be a technology for supporting river bank maintenance.

  However, when image data acquired in an outdoor natural environment is used for analysis, color information to be observed is not uniform. Even if the observation target is the same, the color tone in the image differs depending on the data acquisition conditions such as the weather and time, and thus erroneous recognition is likely to occur in the analysis. Therefore, these must be taken into account when analyzing river levee images.

  For example, various methods for automatically determining the vegetation status from aerial photographs and satellite photographs have been proposed (Patent Documents 1 to 3 and the like). In these methods, when a plurality of vegetation are mixed, Whether or not the vegetation is harmful vegetation could not be determined, and it was difficult to detect only harmful vegetation.

  In order to solve the problem, in Patent Document 4, an R component image, a G component image, and a B component image are created from an identification target vegetation image in a vegetation survey target region image obtained by photographing a vegetation survey target region, and texture analysis is performed. From the R component image, the G component image, and the B component image, one or more of an average value of luminance values, an average value of contrast, smoothness, uniformity, and entropy as a feature amount of each image By extracting, only harmful vegetation can be detected from the target image. However, in the technique according to Patent Document 4, a parameter that can be changed according to data acquisition conditions such as weather and time is extracted as a feature amount, and erroneous recognition is likely to occur in the analysis.

JP 2007-303855 A JP 2006-252529 A JP 2007-128141 A JP 2008-210165A

  Therefore, the present invention extracts a vegetation area from a color image while reducing the influence of seasonal changes and data acquisition conditions such as sunlight, and in the vegetation area, a harmless vegetation area (turf area) and a harmful vegetation area are appropriately selected. It is an object of the present invention to provide a determination method, a determination device, and a determination program that can be determined easily.

As a result of researches on a method for discriminating vegetation regions in image data, the present inventors have obtained the following knowledge.
(1) In the vegetation region, among the RGB components, the G component has a larger value than the other components. In addition, there is a predetermined magnitude relationship between the RGB component values in the vegetation region and other regions. That is, it is possible to appropriately extract only the vegetation region from the image by using the magnitude relationship of the RGB values in each pixel of the color image.
(2) When the RGB values are compared and analyzed for the harmful vegetation area and the turf area included in the vegetation area, the (GR) value is large in the harmful vegetation area and small in the turf area. That is, the R / G value obtained by dividing the R component value by the G component value is small in the harmful vegetation region and large in the turf region.
(3) When an R / G value is used as a parameter when analyzing a vegetation region, a relative analysis in which the influence of data acquisition conditions such as seasonal changes and sunlight is reduced is possible. For example, when a harmful vegetation region and a turf region are mixed in a vegetation region, a pixel portion having an R / G value equal to or greater than a threshold is relatively determined as a turf region, and a pixel portion having an R / G value less than the threshold is determined as a harmful vegetation region. Is possible.
(4) When the R / G value is calculated for each pixel in the vegetation region and a histogram for the R / G value is created, the harmful vegetation region and the turf region are mixed in the vegetation region (mixed vegetation region) The shape of the histogram differs between when only the turf area or only the harmful vegetation area exists (single vegetation area). That is, it is possible to appropriately classify whether the vegetation region is a mixed vegetation region or a single vegetation region by using a statistic obtained from the created histogram.

The present invention has been made based on the above findings. That is,
The first aspect of the present invention is a vegetation region extraction step of extracting pixels constituting a vegetation region using the magnitude relationship of RGB values in each pixel of a color image;
For each pixel of the extracted vegetation region, a R / G value is calculated by dividing the R component value by the G component value, and a histogram using the R / G value is created.
A class classification process for classifying whether the vegetation region is a single vegetation region or a mixed vegetation region using a statistic obtained from the created histogram,
In the class classification step, when the vegetation region is classified as a mixed vegetation region, a determination step of determining a region composed of pixels having an R / G value less than a threshold in the vegetation region as a harmful vegetation region;
A method for determining a vegetation area included in a color image.

  In the present application, “vegetation” means a group of plants. In particular, vegetation on river banks is roughly divided into turf and harmful vegetation. In river dikes, harmful vegetation is what causes lawn to die by overgrowth. As a specific example of harmful vegetation, for example, a Japanese perennial plant, a Japanese knotweed.

  In the histogram creating step according to the first aspect of the present invention, it is preferable that the R / G value is normalized to a predetermined gradation and the number of pixels in each gradation is histogrammed.

  In the classification process according to the first aspect of the present invention, it is preferable to use the average value of R / G values, the skewness and kurtosis of the histogram as the statistics.

In the histogram creation process according to the first aspect of the present invention, the R / G value is normalized to a predetermined gradation (for example, 256 gradations are desirable), and the number of pixels in each gradation is histogrammed. When the statistics obtained from the histogram satisfy all of the following conditions (1) to (3), the vegetation area is classified as a mixed vegetation area, and any one of the following conditions (1) to (3) However, it is particularly preferable to classify the vegetation region into a single vegetation region when it is not satisfied.
(1) The average value of R / G values is equal to or less than a threshold value (when normalized to 256 gradations, for example, 120 or less is desirable).
(2) The skewness of the histogram is equal to or less than a threshold value (when normalized to 256 gradations, for example, 1 or less is desirable).
(3) The kurtosis of the histogram is equal to or less than a threshold (when normalized to 256 gradations, for example, 14 or less is desirable).

In the vegetation area extracting step according to the first aspect of the present invention, an RGB value is obtained for each pixel of the color image, and an area composed of pixels satisfying the following condition (I) or (II) is defined as the vegetation area. It is preferable to extract.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value

According to a second aspect of the present invention, there is provided a vegetation area extracting means for extracting pixels constituting the vegetation area using a magnitude relationship of RGB values in each pixel of the color image.
For each pixel of the extracted vegetation region, a histogram creating means for calculating an R / G value by dividing the R component value by the G component value and creating a histogram using the R / G value;
Class classification means for classifying whether the vegetation region is a single vegetation region or a mixed vegetation region using statistics obtained from the created histogram,
In the class classification means, when the vegetation area is classified as a mixed vegetation area, a determination means for determining an area composed of pixels having an R / G value less than a threshold in the vegetation area as a harmful vegetation area;
A vegetation area determination device included in a color image.

  In the second aspect of the present invention, the histogram creating means is preferably means for normalizing the R / G value to a predetermined gradation and making the number of pixels in each gradation a histogram.

  In the class classification means according to the second aspect of the present invention, it is preferable that an average value of R / G values, a skewness of a histogram, and a kurtosis are used as statistics.

In the second aspect of the present invention, the histogram creating means is means for normalizing the R / G value to a predetermined gradation (for example, 256 gradations are desirable) and forming a histogram of the number of pixels in each gradation. When the statistic obtained from the histogram satisfies all of the following conditions (1) to (3), the classification means classifies the vegetation area into a mixed vegetation area, and any of the following conditions (1) to (3) It is particularly preferable that the vegetation region is classified into a single vegetation region when any one of them is not satisfied.
(1) The average value of R / G values is equal to or less than a threshold value (when normalized to 256 gradations, for example, 120 or less is desirable).
(2) The skewness of the histogram is equal to or less than a threshold value (when normalized to 256 gradations, for example, 1 or less is desirable).
(3) The kurtosis of the histogram is equal to or less than a threshold (when normalized to 256 gradations, for example, 14 or less is desirable).

In the second aspect of the present invention, the vegetation area extracting means obtains an RGB value for each pixel of the color image, and extracts an area composed of pixels satisfying the following condition (I) or (II) as the vegetation area. It is preferable that it is a means to do.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value

  A third aspect of the present invention is a system for discriminating a vegetation area included in a color image, comprising an imaging device that captures a color image and a determination device according to the second aspect of the present invention.

According to a fourth aspect of the present invention, there is provided the discrimination device according to the second aspect of the present invention, in which a color image is input.
Using the magnitude relationship of RGB values in each pixel of the color image, the vegetation area included in the color image is extracted,
For the RGB values related to each pixel of the extracted vegetation region, the R component value is divided by the G component value to calculate the R / G value, and a histogram using the R / G value is created.
Using the statistics obtained from the created histogram, classify whether the vegetation area is a single vegetation area or a mixed vegetation area,
When the vegetation region is classified as a mixed vegetation region, a pixel portion having an R / G value less than a threshold in the vegetation region is determined as a harmful vegetation region.
This is a program for discriminating a vegetation area included in a color image.

  In the fourth aspect of the present invention, it is preferable that the discriminating apparatus normalizes the R / G value to a predetermined gradation and makes the number of pixels in each gradation a histogram.

  In the fourth aspect of the present invention, it is preferable that the discriminating device described above calculates an average value of R / G values, skewness and kurtosis of a histogram as statistics.

In the fourth aspect of the present invention, the discriminating device described above normalizes the R / G value to a predetermined gradation (for example, 256 gradations are desirable) to form a histogram of the number of pixels in each gradation, and obtains it from the histogram. When the statistic to be satisfied satisfies all the following conditions (1) to (3), the vegetation area is classified as a mixed vegetation area, and any one of the following conditions (1) to (3) is not satisfied It is particularly preferable to classify the vegetation region into a single vegetation region.
(1) The average value of R / G values is equal to or less than a threshold value (when normalized to 256 gradations, for example, 120 or less is desirable)
(2) The skewness of the histogram is equal to or less than a threshold value (when normalized to 256 gradations, for example, 1 or less is desirable).
(3) The kurtosis of the histogram is equal to or less than a threshold (when normalized to 256 gradations, for example, 14 or less is desirable).

In the fourth aspect of the present invention, it is preferable that the discrimination device obtains an RGB value for each pixel of the color image and extracts a pixel portion in which the RGB value satisfies the following condition (I) or (II) as a vegetation region. .
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value

  In the present invention, the vegetation region is appropriately extracted from the image in consideration of the magnitude relationship of the RGB values, and the vegetation region in the image is analyzed using the R / G value. It is assumed that other than that is discriminated. When an R / G value is used as a parameter when analyzing a vegetation region, a relative analysis in which the influence of data acquisition conditions such as seasonal changes and sunlight is reduced is possible. That is, according to the present invention, a vegetation region is extracted from a color image while reducing the influence of seasonal data and data acquisition conditions such as sunlight, and in the vegetation region, a harmless vegetation region (turf region) and a harmful vegetation region are extracted. It is possible to provide a determination method, a determination device, a determination system, and a determination program.

It is a figure which shows the discrimination method of this invention which concerns on one Embodiment. It is a figure for demonstrating the image data before using for the determination method which concerns on this invention. It is a figure for demonstrating the state from which the vegetation area | region was extracted. It is a figure which shows an example of a histogram. It is a figure for demonstrating a single vegetation area | region and a mixed vegetation area | region. It is a figure for demonstrating the discrimination process of a vegetation area. It is a figure for demonstrating the image data after being provided to the determination method which concerns on this invention. It is a figure for demonstrating the structure of the discrimination device 100 of this invention which concerns on one Embodiment. It is a figure which shows the result which concerns on an Example.

1. Method for Determining Vegetation Regions Included in Color Image FIG. 1 shows an example of a method for determining harmful vegetation regions included in a color image. As shown in FIG. 1, the discrimination method S10 inputs color image data and arbitrarily preprocesses the image data, and forms a vegetation region using the magnitude relationship of RGB values in each pixel of the color image. A vegetation region extraction step S1 that extracts pixels, calculates an R / G value by dividing the R component value by the G component value for each pixel of the extracted vegetation region, and creates a histogram using the R / G value Histogram creation step S2, classifying whether the vegetation region is a single vegetation region or a mixed vegetation region using statistics obtained from the created histogram, class classification step S3, and class classification step When the vegetation region is classified into the mixed vegetation region in S3, a determination step of determining, from the vegetation region, a region composed of pixels having an R / G value less than a threshold value as a harmful vegetation region. It is equipped with a 4.

(Process S0)
Step S0 is a step in which color image data is input to the discriminator and optionally preprocessed. The size, gradation, etc. of the input image are not particularly limited. For example, a color image with 256 gradations may be used for both RGB. As for preprocessing, for example, in the input image data, there may be a character string indicating distortion due to the specifications of the imaging device, the shooting date of the image, or the like at the end of the image. It is preferable to remove unnecessary distortions and character strings by removing the portion by a predetermined number of pixels. Thereby, the discrimination | determination precision of a harmful vegetation area | region can be improved.

(Vegetation region extraction step S1)
The vegetation region extraction step S1 is a step of extracting pixels constituting the vegetation region using the magnitude relationship of the RGB values in each pixel of the color image. In the image to be processed, for example, a vegetation region 1 (turf region 1a, harmful vegetation region 1b) and other regions (soil region 2) are mixed, as in an image 10 schematically shown in FIG. May exist. In step S1, only the vegetation region 1 is extracted from the plurality of regions mixed in this way.

  As a result of intensive studies by the present inventors and investigating the RGB values of the vegetation region 1 (turf region 1a, harmful vegetation region 1b) and other regions (soil region 2) included in the image, the turf region 1a and soil It was found that the values of the R component and the G component in the region 2 and the values of the B component in the harmful vegetation region 1b and the soil region 2 are approximated, respectively. More specifically, the R component value and the G component value are higher in the harmful vegetation region 1b than in the turf region 1a and the soil region 2, and the B component value is in the harmful vegetation region 1b and the soil region 2. It was found to be higher than 1a. That is, it was considered difficult to extract only the vegetation region 1 from the image only by paying attention to any one of the RGB values. On the other hand, in the vegetation area | region 1, it turned out that G component value becomes larger than another component value among RGB values.

  That is, instead of focusing on only one of the RGB values, only the vegetation region 1 can be appropriately extracted from the image by using the magnitude relationship of the RGB values. Preferably, it is as follows.

As a result of considering the magnitude relationship of RGB values, there is a difference in the “B / (GB) value” that is a value obtained by dividing the B value by the (GB) value between the vegetation region 1 and the soil region 2. Admitted. More specifically, the vegetation area | region 1 can be extracted appropriately by extracting the area | region comprised from the pixel whose B / (GB) value satisfy | fills the following conditions (I-1).
(I-1) 0 <B / (GB) <N

  The upper limit value N of the numerical range is any value between 1% and 6% of the maximum gradation value of the color image to be determined. When the maximum gradation is 256 gradations, the value can be any value from 5 to 15. In particular, the upper limit value N that can best extract the vegetation region 1 is about 4% of the maximum gradation value (10 in the case of 256 gradations). Moreover, since the vegetation area | region 1 shows a value with high G component compared with B component as above-mentioned, a lower limit can be made into 0.

However, a partial region may be erroneously extracted only with the above condition (I-1). From this viewpoint, in order to more appropriately extract only the vegetation region 1, it is particularly preferable to extract a pixel region that simultaneously satisfies the following condition (I-2) together with the above condition (I-1).
(I-2) R ≦ G or B + M <G

  The condition (I-2) is set by paying attention to the fact that, in the vegetation region 1, the G component value is larger than the R component value and the B component value. In addition, since the difference between the G component value and the B component value is larger than the difference between the G component value and the R component value in the vegetation region 1, it is preferable that B + M <G. Here, the value M to be added to the B component value is any value between 9% and 14% of the maximum gradation value. When the maximum gradation value is 256 gradations, it can be any value between 25 and 35. In particular, the value M that can extract the vegetation region 1 best is about 12% of the maximum gradation value (30 in the case of 256 gradations).

Furthermore, as a result of intensive studies by the present inventors, it has been found that the harmful vegetation region 1b is affected by sunlight more than other regions, and there is a portion where the gradation value is significantly increased. That is, it can be said that the region where both the RGB values are 78% or more of the maximum gradation value (for example, 200 or more in the case of 256 gradations) is the harmful vegetation region 1b affected by sunlight. From this viewpoint, a pixel region that satisfies the following condition (II) can also be extracted as the vegetation region 1.
(II) 78% or more of the maximum gradation value for R, G and B

In summary, in the vegetation region extraction step S1, in order to appropriately extract only the vegetation region 1 included in the image, the RGB values are configured from pixels that satisfy the following condition (I) or (II). Most preferably, the region is extracted.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(Here, N is any value from 1% to 6% of the maximum gradation value, preferably about 4% of the maximum gradation value, and M is 9% to 14% of the maximum gradation value. Any of the following values, preferably about 12% of the maximum gradation value)
(II) For all of R, G, and B, 78% or more of the maximum gradation value

When the RGB of the color image to be discriminated has 256 gradations, it is desirable that the conditions (I) and (II) are specifically as follows.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(Here, N is any value from 5 to 15, preferably 10, and M is any value from 25 to 35, preferably 30.)
(II) R ≧ 200, G ≧ 200, and B ≧ 200

  Thus, in step S1, by extracting the vegetation region 1 using the magnitude relationship of the RGB values, for example, an image in which a region (soil region 2) other than the vegetation region 1 is rejected as shown in FIG. 20 can be obtained.

(Histogram creation step S2)
The histogram creation step S2 is a step of calculating an R / G value by dividing the R component value by the G component value for each pixel of the extracted vegetation region 1, and creating a histogram using the R / G value. .

  When the RGB values of the turf area 1a and the harmful vegetation area 1b included in the vegetation area 1 are compared and analyzed, the (GR) value may be large in the harmful vegetation area 1b and small in the turf area 1a. I understood. That is, the R / G value obtained by dividing the R component value by the G component value is small in the harmful vegetation region 1b and large in the turf region 1a. Here, when the R / G value is used as a parameter, a relative analysis in which the influence of data acquisition conditions such as seasonal changes and sunlight is reduced is possible. This is because the influence of the data acquisition condition is included in each of the R component value and the G component value, but the influence of the data acquisition condition is offset / reduced by dividing the R component value by the G component value. Here, when the R / G value is calculated for each pixel in the vegetation region 1 and a histogram for the R / G value is created, the grass region 1a and the harmful vegetation region 1b are mixed in the vegetation region 1 ( The shape of the histogram differs between the mixed vegetation region) and the case where only the turf region or only the harmful vegetation region exists (single vegetation region).

  In step S2, it is particularly preferable to normalize the R / G value to a predetermined gradation and form a histogram of the number of pixels in each gradation. The “predetermined gradation” is usually 256 gradations, but may be other gradations (64 gradations, 128 gradations, etc.). However, it is preferable to normalize the R / G value to 256 gradations from the viewpoint of facilitating analysis and creating a more suitable histogram. FIG. 4 shows an example of the histogram (histogram 30).

(Class classification process S3)
The class classification step S3 is a step of classifying whether the vegetation region 1 is a single vegetation region or a mixed vegetation region, using a statistic obtained from the created histogram. An example of the single vegetation region is a vegetation region composed of only the turf region 1a as shown in the image 10a in FIG. In addition, a vegetation region including only the harmful vegetation region 1b as shown in the image 10b of FIG. 5B is also classified as a single vegetation region. Examples of the mixed vegetation region include a vegetation region in which the turf region 1a and the harmful vegetation region 1b are mixed as shown in FIG.

  Examples of the “statistic obtained from the histogram” used in the class classification step S3 include an average value of R / G values and values obtained from the shape of the histogram, such as histogram dispersion, skewness, and kurtosis. . In particular, it is preferable to use any one of an average value of R / G values, histogram skewness and kurtosis, more preferably at least histogram skewness and kurtosis, and at least an average value of R / G values, Most preferably, histogram skewness and kurtosis are used. In the present application, the “dispersion degree”, “distortion degree”, and “kurtosis” of the histogram can be calculated using a conventionally known calculation formula.

As described above, since the R / G value is a small value in the harmful vegetation region 1b, when the vegetation region 1 is a mixed vegetation region, the average value of the R / B values is a single value consisting only of the turf region 1a. The value is relatively smaller than that in the case of the vegetation region, and tends to be relatively larger than that in the case of the single vegetation region including only the harmful vegetation region 1b. In addition, according to the analysis by the present inventors, it was found that the mixed vegetation region has a larger variation in gradation value than the single vegetation region. The variation in the gradation value affects, for example, the skewness and kurtosis of the histogram when the number of pixels is histogrammed for each gradation. That is, when the skewness or kurtosis of the histogram is equal to or less than the threshold value, the vegetation region can be classified as a mixed vegetation region. In summary, in the class classification step S3, when all of the following conditions (1) to (3) are satisfied, the vegetation area is classified as a mixed vegetation area, and any of the following conditions (1) to (3) It is particularly preferable to classify the vegetation region into a single vegetation region when any one of them is not satisfied.
(1) Average R / G value is below threshold (2) Histogram skewness is below threshold (3) Histogram kurtosis is below threshold

  In particular, when the histogram is created by normalizing the R / G value to a predetermined gradation in the histogram creation step S2, the threshold value related to the average value of the R / G value is set to 40% to 50% of the maximum gradation value ( When normalized to 256 gradations, 120) is most preferable. Further, when a histogram is created by normalizing the R / G value to 256 gradations, the threshold relating to the skewness of the histogram is set to 1, and the threshold relating to the kurtosis of the histogram is set to 14, so that the vegetation region is most appropriately applied. One classification can be performed.

(Determination step S4)
When the vegetation region 1 is classified as the mixed vegetation region in the classification step S3, the determination step S4 defines a region composed of pixels having an R / G value less than a threshold in the vegetation region 1 as the harmful vegetation region 1b This is a step of discriminating. As described above, the R / G value is small in the harmful vegetation region 1b and large in the turf region 1a. Therefore, as shown in FIG. 6, an area composed of pixels having an R / G value equal to or greater than a threshold is defined as a turf area 1a, and an area composed of pixels having an R / G value less than the threshold is defined as a harmful vegetation area 1b. It can be determined appropriately.

  For example, when the R / G value is normalized to a predetermined gradation, the threshold value of the R / G value in the determination step S4 is preferably set to a value corresponding to 41% to 46% of the maximum gradation value. For example, when normalized to 256 gradations, 104 to 119 can be set as the threshold value. However, the threshold value is merely an example. The threshold value can be corrected as appropriate. Further, the threshold value may be set by a known discriminant analysis method (Otsu's method) using a histogram.

  In the harmful vegetation region 1b in the vegetation region 1, as described above, sunlight may be reflected on the upper part of the harmful vegetation, and the color information may be changed. In this case, in some pixels of the harmful vegetation region 1b, the R / B value may be equal to or greater than the threshold value, and it may be erroneously determined that a part of the harmful vegetation region 1b is the turf region 1b. From the viewpoint of preventing this, it is preferable to perform an expansion / contraction process on the portion related to the harmful vegetation region 1b. Specifically, eight pixels in the vicinity of the pixel determined as the harmful vegetation region 1b are regarded as being the same harmful vegetation region 1b, and subjected to an expansion process for correction, and further, the eight pixels in the vicinity of the harmful vegetation region 1b If there is another area such as the turf area 1a, the turf area may be regarded as a turf area, and a contraction process may be performed to correct the harmful vegetation area 1b to the turf area.

  Through the determination step S4, for example, as schematically shown in FIG. 7, an image 40 in which the turf area, the harmful vegetation area, and the soil area are appropriately determined can be obtained.

  As described above, in the discrimination method S10, the vegetation region 1 is appropriately extracted from the image in consideration of the magnitude relationship of the RGB values (step S1), and the vegetation region 1 in the image is determined using the R / G value. Analysis and histogram formation (step S2), and using the statistics obtained from the histogram, the harmful vegetation region 1b included in the vegetation region 1 and the other (turf region 1a) are determined (steps S3 and S4). . When the R / G value is used as a parameter when analyzing the vegetation region 1, it is possible to perform a relative analysis in which the influence of data acquisition conditions such as seasonal changes and sunlight is reduced. That is, according to the discrimination method S10, the vegetation region 1 is extracted from the color image while reducing the influence of the data acquisition conditions such as seasonal change and sunlight, and the harmless vegetation region (turf region 1a) is extracted from the vegetation region 1. And harmful vegetation region 1b can be appropriately determined. In addition, the ratio of the vegetation region 1 in the image and the ratio of the harmful vegetation region 1b in the vegetation region 1 can be appropriately calculated.

2. Discriminating device, discriminating system, and discriminating program for vegetation area included in color image A discriminating device capable of executing the discriminating method S10 according to the present invention will be described. FIG. 8 schematically shows a determination apparatus 100 according to an embodiment of the present invention. As shown in FIG. 8, the discrimination apparatus 100 includes a preprocessing unit that performs the step S0 related to the preprocessing, a vegetation region extraction unit that performs the vegetation region extraction step S1, a histogram generation unit that performs the histogram generation step S2, and the class. The CPU 11 functions as a class classification unit that performs the classification step S3, a determination unit that performs the determination step S4, and a storage device for the CPU 11. The CPU 11 is configured by combining a microprocessor unit and various peripheral circuits necessary for the operation thereof, and a storage device for the CPU 11 includes, for example, programs and various data (for example, determination targets) necessary for the determination and execution of the above steps S1 to S4. ROM 12 that stores a plurality of image data), a RAM 13 that functions as a work area of the CPU 11, and the like. In addition to the configuration, the determination device 100 functions by combining the CPU 11 with software stored in the ROM 12.

  A user who discriminates the vegetation area of the image data inputs data to the discriminating apparatus 100 via the input means 16 (for example, a keyboard of a personal computer). An input signal from the input means 16 reaches the CPU 11 as an input signal via the input port 14. The CPU 11 performs the above steps S1 to S4 based on the signal from the input means 16 and the program stored in the ROM 12, and outputs a signal related to the discrimination result via the output port 15 to the output means (for example, the screen of the personal computer). And the vegetation area discrimination result is displayed on the output means 17. For example, the vegetation area determination method S10 is executed for each of a plurality of image data to classify whether the vegetation area in the image data is a single vegetation area or a mixed vegetation area, and occupies the classification result or image. In addition to the form in which the ratio of harmful vegetation areas is displayed on the output means 17, only images with a ratio of harmful vegetation areas of a certain value or more are displayed in the image data, or a plurality of harmful vegetation areas are in descending order. The image data is rearranged and displayed on the output unit 17, so that the user can easily determine whether or not weeding of harmful vegetation at the imaging point is necessary based on the result displayed on the output unit 17. can do.

  Specifically, signals and information (image data selection signal, gradation setting for creating a histogram, and statistics obtained from the histogram are necessary for executing the discrimination method S10 via the input unit 16. Calculation formulas, various threshold values, etc.) are input to the discrimination device 100. On the other hand, the ROM 12 stores a determination program capable of executing the determination method S10 according to the present invention, and the CPU 11 can execute the determination method S10 by executing the program. The determination program is not particularly limited as long as it can cause the determination apparatus 100 to perform the determination method S10. For example, the determination program may be as follows.

  That is, in the discriminating apparatus 100 to which a color image has been input, the CPU 11 is caused to extract a vegetation region included in the color image using an RGB signal value in each pixel of the color image, based on an input signal from the user, For the RGB values related to each pixel of the extracted vegetation region, the R component value is divided by the G component value to calculate the R / G value, and a histogram using the R / G value is created, and the created histogram If the vegetation area is classified as a single vegetation area or a mixed vegetation area using the statistics obtained from the above, and the vegetation area is classified as a mixed vegetation area, the R / G value in the vegetation area is The discrimination device 100 can be caused to execute the discrimination method S10 by a discrimination program that discriminates a pixel portion that is less than the threshold value as a harmful vegetation region.

  Here, it is preferable that the discriminating apparatus 100 normalizes the R / G value to a predetermined gradation and makes the number of pixels in each gradation a histogram. As described above, by creating a histogram using R / G values normalized to a predetermined gradation, a vegetation region is a single vegetation region or a mixed vegetation region based on statistics such as the shape of the histogram. It can be classified more appropriately.

  Further, from the viewpoint of more appropriately classifying the vegetation regions, it is preferable that the determination device 100 calculate the average value of R / G values, the skewness and kurtosis of the histogram as the statistics. It is more preferable that the discriminating apparatus 100 classify the vegetation region into the mixed vegetation region when the average value of the R / G value, the skewness of the histogram, and the kurtosis are all equal to or less than the threshold value.

In particular, the discriminating apparatus 100 normalizes the R / G value to a predetermined gradation (for example, 256 gradations are desirable) to make the number of pixels in each gradation a histogram, and the statistics obtained from the histogram are as follows ( When all of the conditions 1) to (3) are satisfied, the vegetation area is classified into a mixed vegetation area, and when any one of the following conditions (1) to (3) is not satisfied, the vegetation area is single If the vegetation region is classified, the vegetation region can be more appropriately classified into a single vegetation region or a mixed vegetation region.
(1) The average value of R / G values is equal to or less than a threshold value (when normalized to 256 gradations, for example, 120 or less is desirable).
(2) The skewness of the histogram is equal to or less than a threshold value (when normalized to 256 gradations, for example, 1 or less is desirable).
(3) The kurtosis of the histogram is equal to or less than a threshold (when normalized to 256 gradations, for example, 14 or less is desirable).

Further, when the discrimination device 100 determines an RGB value for each pixel of the color image and extracts an area composed of pixels satisfying the following condition (I) or (II) as the vegetation area, the color image It is possible to extract only the vegetation area more appropriately.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(Here, N is any value from 1% to 6% of the maximum gradation value, preferably about 4% of the maximum gradation value, and M is 9% to 14% of the maximum gradation value. Any of the following values, preferably about 12% of the maximum gradation value)
(II) For all of R, G, and B, 78% or more of the maximum gradation value

When the RGB of the color image to be discriminated has 256 gradations, it is desirable that the conditions (I) and (II) are specifically as follows.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(Here, N is any value from 5 to 15, preferably 10, and M is any value from 25 to 35, preferably 30.)
(II) R ≧ 200, G ≧ 200, and B ≧ 200

  As described above, the discriminating apparatus 100 includes the vegetation region extracting means (CPU 11) for extracting the pixels constituting the vegetation region using the magnitude relationship of the RGB values in each pixel of the color image, and the extracted vegetation region. For each pixel, the R component value is divided by the G component value to calculate an R / G value, and a histogram using the R / G value is created. Class classification means (CPU11) for classifying whether the vegetation area is a single vegetation area or a mixed vegetation area, and the class classification means, the vegetation area is classified into the mixed vegetation area A determination means (CPU 11) for determining an area composed of pixels having an R / G value less than a threshold value as a harmful vegetation area in the vegetation area; By the determination program, and it is capable of executing the determination method S10. That is, according to the discriminating apparatus 100, the vegetation region 1 is extracted from the color image while reducing the influence of data acquisition conditions such as seasonal changes and sunlight, and the harmless vegetation region (turf region 1a) is extracted from the vegetation region 1. ) And the harmful vegetation region 1b can be appropriately determined.

  In addition, what is necessary is just to use a well-known imaging device about acquisition of image data. For example, images sequentially obtained from a fixed point camera used for river bank monitoring or the like can be used. In the above description, the image data is described as being input by the user. However, it is assumed that the determination device 100 can automatically and sequentially receive information from the fixed point camera via the network. On the other hand, a program may be set so that the determination method S10 is performed on the received image data at regular intervals.

<Evaluation method>
The discrimination method according to the present invention was evaluated by the following method.
(1) For 54 images (720 pixels × 480 pixels, 256 gradations for each RGB) acquired from an imaging device (multicast CCTV image providing system) installed on a river bank, the image is recorded using the determination device according to the present invention. The discriminating method according to the invention was carried out to classify whether the vegetation area included in the image is a single vegetation area or a mixed vegetation area.
(2) Discrimination results (examples) for 18 images including mixed vegetation regions as vegetation regions, and discriminating images into harmful vegetation regions, turf regions, and other regions by the discrimination method according to the present invention. , RGB components or a * components as feature quantities, and a discrimination result (comparative example) obtained by applying the k-means method (k = 3) to the target image, and a harmful vegetation region, a turf region, and other The discrimination results (reference examples) discriminated into areas were respectively acquired and the similarity of the discrimination results was compared. For the comparative example, the a * component was used as a feature value because a value of a * in the turf was higher than that of harmful vegetation.
(3) Discrimination results (examples) for 18 images including mixed vegetation regions as vegetation regions, and discriminating images into harmful vegetation regions, turf regions, and other regions by the discrimination method according to the present invention. Then, the discrimination results (reference examples) discriminate | determined into the harmful vegetation area | region, the turf area | region, and the other area | region by visual observation were acquired, respectively, and the coincidence rate of the discrimination result was computed. The “match rate” is a ratio of pixels determined to be the same region between the example and the reference example.

<Evaluation results>
As a result of the class classification processing according to the discrimination method of the present invention, it is possible to correctly classify whether the vegetation area included in the levee image is a mixed vegetation area or a single vegetation area in 54 out of 54 target image data (100%). It was.

  FIG. 9 shows a discrimination result (reference example: FIG. 9B) obtained by visually discriminating a harmful vegetation region, a turf region, and other regions with respect to one image data (FIG. 9A). Results of discrimination of harmful vegetation area and turf area obtained by the discrimination method of the invention (Example: FIG. 9C) and the k-means method (k = 3) using the a * component as a feature amount And the discrimination result obtained by applying the k-means method (k = 3) to the target image using the RGB components as feature quantities (comparative example: FIG. 9D). FIG. 9E) is shown. As is clear from FIG. 9, in the results (FIGS. 9D and 9E) determined using the k-means method, the turf area is misclassified as other areas, and the harmful vegetation area is classified as turf. It can be seen that it is misclassified as a region. On the other hand, it can be seen that when the vegetation region is determined by the determination method according to the present invention (FIG. 9C), a result similar to that obtained when visually determined (FIG. 9B) is obtained. That is, it can be said that the discrimination method according to the present invention is useful for classification of river bank images and discrimination of vegetation regions.

  By the discrimination method according to the present invention, the discrimination result (Example) in which the image is discriminated into the harmful vegetation region, the turf region, and other regions, and the visual discrimination into the harmful vegetation region, the turf region, and other regions are made. When the discrimination results (reference examples) were obtained and the coincidence rate of the discrimination results was calculated, the coincidence rate of all pixels (average value of 18 images) was 81.3% and a good result was obtained. The discrepancy rate (18.7%) is as shown in Table 1 below. The main reason for the discrepancy rate is that in the vegetation region extraction step (step S1), other regions included in the image such as poles and chords are erroneously extracted as harmful vegetation regions and turf regions. It was. Therefore, it is considered that the matching rate can be further improved by improving the extraction accuracy in the vegetation region extraction step.

  Although the present invention has been described with reference to the most practical and preferred embodiments at the present time, the invention is not limited to the embodiments disclosed herein. The vegetation area determination method, determination apparatus, determination system, and determination program can be changed as appropriate without departing from the spirit or concept of the invention that can be read from the claims and the entire specification. It should be understood as encompassed within the scope of the present invention.

  According to the present invention, for example, the situation of a river levee can be automatically collected using an image acquired from a camera provided on the river levee, and a quantitative evaluation of harmful vegetation can be performed. Can support the maintenance.

1 Vegetation (vegetation area)
1a Turf (turf area)
1b Harmful vegetation (harmful vegetation area)
2 soil (soil area)

Claims (16)

A vegetation region extracting step of extracting pixels constituting the vegetation region using the magnitude relationship of RGB values in each pixel of the color image;
For each pixel of the extracted vegetation region, a R / G value is calculated by dividing the R component value by the G component value, and a histogram using the R / G value is created.
Classifying whether the vegetation region is a single vegetation region or a mixed vegetation region using a statistic obtained from the created histogram,
In the class classification step, when the vegetation region is classified as a mixed vegetation region, a determination step of determining, from the vegetation region, a region composed of pixels having the R / G value less than a threshold value as a harmful vegetation region. When,
A method for determining a vegetation area included in a color image.   The determination method according to claim 1, wherein in the histogram creation step, the R / G value is normalized to a predetermined gradation, and the number of pixels in each gradation is histogrammed.   The discrimination method according to claim 1 or 2, wherein, in the class classification step, an average value of the R / G values, skewness and kurtosis of the histogram are used as the statistics. In the histogram creation step, the R / G value is normalized to a predetermined gradation, and the number of pixels in each gradation is converted into a histogram. In the class classification step, the statistics obtained from the histogram are the following (1) to ( When all the conditions of 3) are satisfied, the vegetation area is classified into a mixed vegetation area, and when any one of the following conditions (1) to (3) is not satisfied, the vegetation area is defined as a single vegetation area. The discrimination method according to claim 1, which is classified into:
(1) Average R / G value is below threshold (2) Histogram skewness is below threshold (3) Histogram kurtosis is below threshold 2. The vegetation region extracting step obtains an RGB value for each pixel of the color image and extracts a region composed of pixels satisfying the following condition (I) or (II) as a vegetation region. The discrimination | determination method of any one of -4.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value A vegetation region extracting means for extracting pixels constituting the vegetation region using the magnitude relationship of RGB values in each pixel of the color image;
For each pixel of the extracted vegetation region, a R / G value is calculated by dividing the R component value by the G component value, and a histogram creating means for creating a histogram using the R / G value;
Classifying means for classifying whether the vegetation region is a single vegetation region or a mixed vegetation region, using a statistic obtained from the created histogram,
In the class classification means, when the vegetation area is classified as a mixed vegetation area, a determination means that determines an area composed of pixels having the R / G value less than a threshold in the vegetation area as a harmful vegetation area; ,
An apparatus for discriminating a vegetation area included in a color image.   The discrimination apparatus according to claim 6, wherein the histogram creating means is means for normalizing the R / G value to a predetermined gradation and forming a histogram of the number of pixels in each gradation.   The discriminating apparatus according to claim 6 or 7, wherein in the class classification means, an average value of the R / G values, skewness and kurtosis of the histogram are used as the statistics. The histogram creating means is means for normalizing the R / G value to a predetermined gradation and forming a histogram of the number of pixels in each gradation, and the class classification means has a statistic obtained from the histogram as follows: When all the conditions of 1) to (3) are satisfied, the vegetation area is classified into a mixed vegetation area, and when any one of the following conditions (1) to (3) is not satisfied, the vegetation area is The discriminating apparatus according to claim 6, which is means for classifying into a single vegetation region.
(1) Average R / G value is below threshold (2) Histogram skewness is below threshold (3) Histogram kurtosis is below threshold The vegetation area extracting means is means for obtaining an RGB value for each pixel of the color image and extracting an area composed of pixels satisfying the following condition (I) or (II) as a vegetation area: The discrimination device according to any one of claims 6 to 9.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value An imaging device for imaging a color image;
The discrimination device according to any one of claims 6 to 10,
A vegetation area discrimination system included in a color image. The discrimination apparatus according to any one of claims 6 to 10, wherein a color image is input,
Using the magnitude relationship of RGB values in each pixel of the color image, the vegetation region included in the color image is extracted,
For the RGB value related to each pixel of the extracted vegetation region, the R component value is divided by the G component value to calculate the R / G value, and a histogram using the R / G value is created.
Using the statistics obtained from the created histogram, classify whether the vegetation region is a single vegetation region or a mixed vegetation region,
When the vegetation region is classified as a mixed vegetation region, the pixel portion in which the R / G value is less than a threshold in the vegetation region is determined as a harmful vegetation region.
A program for discriminating vegetation areas included in color images.   The program according to claim 12, wherein the discriminating apparatus normalizes the R / G value to a predetermined gradation and causes the number of pixels in each gradation to be a histogram.   The program according to claim 12 or 13, which causes the discrimination device to calculate an average value of the R / G values, skewness and kurtosis of the histogram as the statistics. The discriminating apparatus normalizes the R / G value to a predetermined gradation and makes the number of pixels in each gradation a histogram, and the statistics obtained from the histogram satisfy all the following conditions (1) to (3): In this case, the vegetation region is classified into a mixed vegetation region, and if any one of the following conditions (1) to (3) is not satisfied, the vegetation region is classified into a single vegetation region. The program described in.
(1) Average R / G value is below threshold (2) Histogram skewness is below threshold (3) Histogram kurtosis is below threshold The discrimination device is caused to obtain an RGB value for each pixel of the color image, and an area composed of pixels satisfying the following condition (I) or (II) is extracted as a vegetation area: 15. The program according to any one of 15.
(I) 0 <B / (GB) <N and R ≦ G or B + M <G
(N is any value between 1% and 6% of the maximum gradation value, and M is any value between 9% and 14% of the maximum gradation value.)
(II) For all of R, G, and B, 78% or more of the maximum gradation value