Intelligent classification method for trees with diseases and insect pests and withered trees based on multispectral images

Abstract

The invention discloses an intelligent classification method for trees with diseases and insect pests and withered trees based on a multispectral image, and belongs to the technical field of image processing, and the method comprises the following steps: S1, collecting the multispectral image of a to-be-classified region, carrying out the geometric correction and radiation correction of the multispectral image, and carrying out the classification of the to-be-classified region; extracting a pixel value of a pixel point in each wave band in the latest multispectral image; s2, determining a nonlinear adjustment value of each wave band according to a pixel value of a pixel point in the newest multispectral image in each wave band, and generating a color entropy for the newest multispectral image; and S3, determining a pest and disease damage area of the newest multispectral image, and determining a withered tree area in the remaining area according to the color entropy of the newest multispectral image. According to the method, the robustness to the linear change of the spectrum is kept, and the processing efficiency of large-scale images is greatly improved on the premise of ensuring the classification precision.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and particularly relates to an intelligent classification method for pest-damaged trees and dead trees based on multispectral images. Background Art

[0002] With the intensification of climate change and pests and diseases, the number of dead trees in forests has been increasing year by year, which has had a serious impact on the ecosystem and forestry resource management. Traditional methods for monitoring dead trees mainly rely on manual field surveys, which are inefficient, costly, and difficult to cover large areas. In recent years, unmanned aerial vehicle (UAV) remote sensing technology has gradually been applied to forestry monitoring. However, existing technologies mostly rely on single-spectral information and are commonly used for dead tree monitoring, but there are the following problems: 1) Single feature limitation: Traditional methods rely on a single vegetation index (such as NDVI), making it difficult to distinguish between pest-damaged (local lesions) and dead trees (overall death); 2) Insufficient classification accuracy: Existing algorithms have a high misjudgment rate for pest-damaged trees and dead trees with similar spectral characteristics, especially in complex forest environments. Summary of the Invention

[0003] In order to solve the above problems, the present invention proposes an intelligent classification method for pest-damaged trees and dead trees based on multispectral images.

[0004] The technical solution of the present invention is as follows: An intelligent classification method for pest-damaged trees and dead trees based on multispectral images includes the following steps:

[0005] S1. Collect multispectral images of the area to be classified. After geometric correction and radiometric correction of the multispectral images, extract the pixel values of the pixel points in each band of the latest multispectral image.

[0006] S2. According to the pixel values of the pixel points in each band of the latest multispectral image, determine the non-linear adjustment value of each band, and generate color entropy for the latest multispectral image.

[0007] S3. Determine the pest-damaged area of the latest multispectral image, and in the remaining area, determine the dead tree area according to the color entropy of the latest multispectral image.

[0008] Further, S2 includes the following sub-steps:

[0009] S21. Add the ratio of the pixel value of the pixel point in the red band of the latest multispectral image to the maximum red band pixel value, the ratio of the pixel value of the pixel point in the blue band to the maximum blue band pixel value, and the ratio of the pixel value of the pixel point in the green band to the maximum green band pixel value as the color rate of the pixel point.

[0010] S22. Calculate the color entropy of the latest multispectral image according to the color rates of all pixel points.

[0011] The beneficial effects of the above further solution are as follows: In the present invention, by comparing the pixel values of the red, yellow, and green bands with the maximum pixel values of their respective bands and summing them, the color rate of the pixel point is obtained, which can comprehensively reflect the relative intensity of the pixel point on different color channels and provides a comprehensive index for further analyzing color features.

[0012] Further, S22 includes the following sub-steps:

[0013] S221. Calculate the non-linear adjustment value of each band according to the lower quartile and upper quartile of the pixel values of all pixel points in the latest multispectral image in each band.

[0014] S222. Determine the adjustment range of the latest multispectral image according to the non-linear adjustment values of each band.

[0015] S223. Determine the proportion of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multispectral image.

[0016] S224. Determine the color entropy of the latest multispectral image according to the proportion of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multispectral image.

[0017] The beneficial effects of the above further solution are as follows: In the present invention, the non-linear adjustment value is calculated using the upper and lower quartiles, which can avoid the interference of extreme values, make the data more robust, and adapt to the complex features of the image. The adjustment range combines the adjustment values of each band with the maximum pixel value to limit the effective range of the color rate and screen out representative pixel points. The color entropy is calculated through the proportion of pixel points whose color rate belongs to the adjustment range, converting the pixel distribution proportion into a measure of the uncertainty of color distribution, and providing effective features for the classification of pests and dead trees.

[0018] Further, in S221, the non-linear adjustment value of the band The calculation formula is:

[0019] ;

[0020] In the formula, represents the upper quartile of the pixel values of all pixel points in the latest multispectral image in the band, represents the lower quartile of the pixel values of all pixel points in the latest multispectral image in the band, represents the maximum pixel value of all pixel points in the latest multispectral image in the band, represents the minimum pixel value of all pixel points in the latest multispectral image in the band, represents the exponent.

[0021] Further, in S222, the expression of the adjustment range of the latest multispectral image is , where represents the non - linear adjustment value of the red band, represents the non - linear adjustment value of the green band, represents the non - linear adjustment value of the blue band, represents finding the minimum value, represents the maximum pixel value of all pixels in all bands in the latest multispectral image.

[0022] Furthermore, in S224, the color entropy of the latest multispectral image is calculated by the formula:

[0023] ;

[0024] In the formula, represents the logarithmic function, represents the ratio of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multispectral image.

[0025] Furthermore, S3 includes the following sub - steps:

[0026] S31: Take the pixel points in the latest multispectral image whose normalized red - edge index is less than the set threshold as the diseased and pest - infested tree areas;

[0027] S32: Based on the remaining area in the latest multispectral image except the diseased and pest - infested tree areas, use the latest multispectral image to generate a color entropy - based screening objective function;

[0028] S33: Use the screening objective function to determine the optimal target pixel points;

[0029] S34: Take the pixel points in the remaining area whose Manhattan distance from the optimal target pixel points is less than the set threshold as the dead tree areas.

[0030] The beneficial effects of the above further solution are as follows: In the present invention, the Normalized Difference Red Edge Index (NDRE) is a remote sensing index used to evaluate the health and growth status of vegetation. The decrease in the NDRE value (Normalized Difference Red Edge Index) is directly related to the reduction of chlorophyll caused by pests and diseases, and can be used as an effective screening index. Pixel points with values less than the set threshold are regarded as the areas of pest-infected trees, enabling the rapid location of areas that may be affected by pests and diseases using specific indices, providing a basis for further differentiating dead trees subsequently. A screening objective function is constructed based on the remaining area excluding the areas of pest-infected trees. This function comprehensively considers the color entropy and the relationship between the mean pixel values of different bands, and can highlight points with significant differences in color characteristics from other pixel points in the remaining area, providing an effective mathematical model for screening the areas of dead trees. The best target pixel point is determined using the screening objective function, which can be regarded as the representative pixel point in the remaining area that is most relevant to the characteristics of dead trees. Pixel points in the remaining area with a Manhattan distance less than the set threshold from the best target pixel point are regarded as the areas of dead trees. By measuring the color distance to determine the range of dead trees, the similarity in color of dead trees is taken into account, enabling the relatively accurate identification of the areas of dead trees and avoiding the uncertainty and randomness of individual pixel point judgments.

[0031] Further, in S32, the expression of the screening objective function is:

[0032] ;

[0033] In the formula, represents the color entropy of the latest multispectral image, represents the mean pixel value of the th pixel point in all bands in the remaining area, represents the best target pixel point determined by the screening objective function, represents the mean pixel value of all pixel points in the remaining area in the red band, represents the mean pixel value of all pixel points in the remaining area in the green band, represents the mean pixel value of all pixel points in the remaining area in the blue band.

[0034] The beneficial effects of the present invention are:

[0035] (1) The present invention effectively eliminates image geometric distortion and radiation error through geometric correction and radiometric correction, ensuring that pixel values truly reflect the spectral characteristics of ground objects; by calculating the non-linear adjustment values of each band, the spectral responses of bands such as red, green, and blue are optimized in terms of dynamic range, and the extreme values are adaptively compressed and the intermediate values are stretched using quartile statistics, highlighting the subtle spectral differences between pest-infected trees and dead trees in key bands;

[0036] (2)The color entropy index proposed by the present invention innovatively integrates multi-band information and quantifies the vegetation health status by statistically analyzing the uniformity of the color rate distribution: based on the normalized red-edge index, the pest and disease areas are quickly located. In the remaining areas, through the combination of the color entropy threshold and Manhattan distance clustering, the dead trees with homogenized spectral features are accurately identified;

[0037] (3)The present invention is applicable to the vegetation monitoring in different regions and seasons, has strong generalization ability, significantly improves the sensitivity to the changes in the physiological state of vegetation, and at the same time maintains the robustness to the spectral linear changes. On the premise of ensuring the classification accuracy, the processing efficiency of large-scale images is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of an intelligent classification method for pest and disease trees and dead trees based on multi-spectral images. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The embodiments of the present invention will be further described below with reference to the drawings.

[0040] As Figure 1 shown, the present invention provides an intelligent classification method for pest and disease trees and dead trees based on multi-spectral images, including the following steps:

[0041] S1. Collect the multi-spectral images of the area to be classified, and after geometric correction and radiometric correction of the multi-spectral images, extract the pixel values of the pixel points in each band of the latest multi-spectral images;

[0042] S2. Determine the non-linear adjustment values of each band according to the pixel values of the pixel points in each band of the latest multi-spectral images, and generate the color entropy for the latest multi-spectral images;

[0043] S3. Determine the pest and disease areas of the latest multi-spectral images, and in the remaining areas, determine the dead tree areas according to the color entropy of the latest multi-spectral images.

[0044] In the embodiments of the present invention, S2 includes the following sub-steps:

[0045] S21. Add the ratio of the pixel value of the pixel point in the red band of the latest multi-spectral image to the maximum pixel value in the red band, the ratio of the pixel value in the blue band to the maximum pixel value in the blue band, and the ratio of the pixel value in the green band to the maximum pixel value in the green band as the color rate of the pixel point;

[0046] S22. Calculate the color entropy of the latest multi-spectral images according to the color rates of all pixel points.

[0047] In the present invention, by comparing the pixel values of the red, yellow, and green bands with the maximum pixel values of their respective bands and summing them, the color rate of the pixel point is obtained, which can comprehensively reflect the relative intensity of the pixel point on different color channels and provides a comprehensive index for further analyzing the color characteristics.

[0048] In the embodiment of the present invention, S22 includes the following sub-steps:

[0049] S221. Calculate the non-linear adjustment value of each band according to the lower quartile and upper quartile of the pixel values of all pixel points in the latest multi-spectral image in each band;

[0050] S222. Determine the adjustment range of the latest multi-spectral image according to the non-linear adjustment values of each band;

[0051] S223. Determine the proportion of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multi-spectral image;

[0052] S224. Determine the color entropy of the latest multi-spectral image according to the proportion of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multi-spectral image.

[0053] In the present invention, using the upper and lower quartiles to calculate the non-linear adjustment value can avoid the interference of extreme values, make the data more robust, and adapt to the complex characteristics of the image. The adjustment range combines the adjustment values of each band with the maximum pixel value to limit the effective range of the color rate and screen out representative pixel points. Calculating the color entropy through the proportion of pixel points whose color rate belongs to the adjustment range transforms the pixel distribution proportion into a measure of the uncertainty of color distribution and provides effective features for the classification of pests and dead trees.

[0054] In the embodiment of the present invention, in S221, the non-linear adjustment value of the band The calculation formula is:

[0055] ;

[0056] In the formula, represents the upper quartile of the pixel values of all pixel points in the latest multi-spectral image in the band, represents the lower quartile of the pixel values of all pixel points in the latest multi-spectral image in the band, represents the maximum pixel value of all pixel points in the latest multi-spectral image in the band, represents the minimum pixel value of all pixel points in the latest multi-spectral image in the band, represents the exponent.

[0057] In the embodiment of the present invention, in S222, the expression of the adjustment range of the latest multi-spectral image is , where represents the non - linear adjustment value of the red band, represents the non - linear adjustment value of the green band, represents the non - linear adjustment value of the blue band, represents finding the minimum value, represents the maximum pixel value of all pixels in all bands in the latest multi - spectral image.

[0058] In the embodiment of the present invention, in S224, the color entropy of the latest multi - spectral image has the following calculation formula:

[0059] ;

[0060] In the formula, represents the logarithmic function, represents the proportion of the number of pixels whose color rate belongs to the adjustment range to the total number of pixels in the latest multi - spectral image.

[0061] In the embodiment of the present invention, S3 includes the following sub - steps:

[0062] S31: Take the pixels in the latest multi - spectral image whose normalized red - edge index is less than the set threshold as the pest - damaged tree area;

[0063] S32: Based on the remaining area in the latest multi - spectral image except the pest - damaged tree area, use the latest multi - spectral image to generate a screening objective function for the color entropy;

[0064] S33: Use the screening objective function to determine the optimal target pixel;

[0065] S34: Take the pixels in the remaining area whose Manhattan distance from the optimal target pixel is less than the set threshold as the dead tree area.

[0066] In the present invention, the Normalized Difference Red Edge Index (NDRE) is a remote sensing index used to evaluate the health and growth status of vegetation. The decrease in the NDRE value is directly related to the reduction of chlorophyll caused by pests and diseases, and can be used as an effective screening index. Pixel points with values less than a set threshold are regarded as the areas of pest-infected trees. It is possible to quickly locate the areas that may be affected by pests and diseases by using specific indices, providing a basis for further distinguishing dead trees subsequently. A screening objective function is constructed based on the remaining areas except for the areas of pest-infected trees. This function comprehensively considers the relationship between color entropy and the mean pixel values of different bands, and can highlight the points in the remaining areas that have significant differences in color characteristics from other pixel points, providing an effective mathematical model for screening the areas of dead trees. The best target pixel point is determined using the screening objective function. This point can be regarded as the representative pixel point in the remaining areas that is most relevant to the characteristics of dead trees. Pixel points in the remaining areas with a Manhattan distance less than the set threshold from the best target pixel point are regarded as the areas of dead trees. By measuring the color distance to determine the range of dead trees, the similarity in color of dead trees is taken into account, enabling the relatively accurate identification of the areas of dead trees and avoiding the uncertainty and randomness of individual pixel point judgments.

[0067] In the embodiment of the present invention, in S32, the expression of the screening objective function is:

[0068] ;

[0069] In the formula, represents the color entropy of the latest multispectral image, represents the mean pixel value of the th pixel point in all bands in the remaining area, represents the best target pixel point determined by the screening objective function, represents the mean pixel value of all pixel points in the remaining area in the red band, represents the mean pixel value of all pixel points in the remaining area in the green band, represents the mean pixel value of all pixel points in the remaining area in the blue band.

[0070] Those of ordinary skill in the art will realize that the embodiments described herein are for helping the reader understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. An intelligent classification method for pest-infected trees and dead trees based on multispectral images, characterized in that, It includes the following steps: S1. Collect the multi-spectral images of the area to be classified. After geometric correction and radiometric correction of the multi-spectral images, extract the pixel values of the pixel points in each band of the latest multi-spectral image; S2. Determine the non-linear adjustment values for each band according to the pixel values of the pixel points in each band of the latest multi-spectral image, and generate the color entropy for the latest multi-spectral image; S3. Determine the pest and disease area of the latest multi-spectral image, and determine the dead tree area in the remaining area according to the color entropy of the latest multi-spectral image.

2. The intelligent classification method for pest and disease damaged trees and dead trees based on multispectral images according to claim 1, wherein The S2 includes the following sub-steps: S21. Add the ratio of the pixel value of the pixel point in the red band of the latest multi-spectral image to the maximum red band pixel value, the ratio of the pixel value in the blue band to the maximum blue band pixel value, and the ratio of the pixel value in the green band to the maximum green band pixel value as the color rate of the pixel point; S22. Calculate the color entropy of the latest multi-spectral image according to the color rates of all pixel points.

3. The intelligent classification method for pest-infected trees and dead trees based on multispectral images according to claim 2, wherein The S22 includes the following sub-steps: S221. Calculate the non-linear adjustment values for each band according to the lower quartile and upper quartile of the pixel values of all pixel points in each band of the latest multi-spectral image; S222. Determine the adjustment range of the latest multi-spectral image according to the non-linear adjustment values for each band; S223. Determine the proportion of the number of pixel points whose color rates belong to the adjustment range to the total number of pixel points of the latest multi-spectral image; S224. Determine the color entropy of the latest multi-spectral image according to the proportion of the number of pixel points whose color rates belong to the adjustment range to the total number of pixel points of the latest multi-spectral image.

4. The intelligent classification method for pest and disease trees and dead trees based on multispectral images according to claim 3, wherein In the S221, the non-linear adjustment value of the waveband The calculation formula is as follows: ； Wherein, represents the upper quartile of the pixel values of all pixel points of the latest multispectral image in the band, represents the lower quartile of the pixel values of all pixel points of the latest multispectral image in the band, represents the maximum value of the pixel values of all pixel points of the latest multispectral image in the band, represents the minimum value of the pixel values of all pixel points of the latest multispectral image in the band, represents an exponent.

5. The intelligent classification method for pest-infected trees and dead trees based on multi-spectral images according to claim 3, characterized in that In S222, the expression for the adjustment range of the latest multispectral image is , where represents the non - linear adjustment value of the red band, represents the non - linear adjustment value of the green band, represents the non - linear adjustment value of the blue band, represents finding the minimum value, represents the maximum pixel value of all pixel points in all bands of the latest multispectral image.

6. The intelligent classification method for pest-infected trees and dead trees based on multispectral images according to claim 3, characterized in that In S224, the color entropy of the latest multispectral image is calculated by the formula: ； In the formula, represents the logarithmic function, represents the ratio of the number of pixel points whose color rate belongs to the adjustment range to the total number of pixel points in the latest multispectral image.

7. The intelligent classification method for pest-infected trees and dead trees based on multispectral images according to claim 1, characterized in that, The S3 includes the following sub-steps: S31. Take the pixel points with the normalized red edge index less than the set threshold in the latest multi-spectral image as the pest and disease tree area; S32. Based on the remaining area in the latest multi-spectral image except the pest and disease tree area, use the color entropy generated by the latest multi-spectral image to construct a screening objective function; S33. Use the screening objective function to determine the optimal target pixel points; S34. Take the pixel points with the Manhattan distance less than the set threshold from the optimal target pixel points in the remaining area as the dead tree area.

8. The intelligent classification method for pest-damaged trees and dead trees based on multispectral images according to claim 7, wherein In the S32, the expression of the screening objective function is: ； In the formula, represents the color entropy of the latest multispectral image, represents the average pixel value of the -th pixel point in all bands in the remaining area, represents the best target pixel point determined by the screening objective function, represents the average pixel value of all pixel points in the remaining area in the red band, represents the average pixel value of all pixel points in the remaining area in the green band, represents the average pixel value of all pixel points in the remaining area in the blue band.

Images