Method for detecting color-changing wood based on hyperspectral image of unattended unmanned aerial vehicle

Abstract

The invention discloses a novel vegetation index for color-changing wood identification based on an unmanned aerial vehicle hyperspectral image and a construction method. By extracting a tree spectral reflection curve, a feature selection algorithm is used for screening a feature wave band to construct an area vegetation index. The novel index constructed by the invention can effectively enhance the difference between discolored wood and healthy wood. Meanwhile, the invention provides a color-changing wood detection method based on an unattended unmanned aerial vehicle hyperspectral image, and relates to the field of remote sensing image processing and vegetation disease and pest detection, in particular to a color-changing wood unmanned aerial vehicle hyperspectral image detection method. Comprising the following steps: 1, automatically collecting a hyperspectral image in a target area and extracting the spectral reflectivity of vegetation; 2, screening characteristic wave bands by using a characteristic selection algorithm to construct an area vegetation index, constructing a PLS-DA partial least square discrimination model, and training a one-dimensional convolutional neural network; 3, constructing an automatic identification module comprising the three methods; and 4, inputting the data to be detected into the module for detection.

Description

Technical Field

[0001] The present invention relates to the fields of remote sensing image processing and vegetation pest and disease detection, and in particular to a method for detecting discolored wood caused by vegetation pest and disease using unmanned aerial vehicle (UAV) hyperspectral image recognition. Background Art

[0002] In recent years, pine wood nematode disease, red pine borer, betel nut yellowing disease, pine cutworm, etc. will cause changes in the pigment content and moisture content of tree leaves, which in turn leads to frequent and recurring important forestry pests and diseases that cause leaf discoloration, posing a serious threat to the safety of forest ecological environment. The manual detection methods used in the past are time-consuming, labor-intensive and inefficient, and often delay the best time for pest control. The development of remote sensing technology has provided a new method for the detection of discolored wood. However, satellite remote sensing images are easily limited by spatial resolution and spectral resolution, resulting in low detection accuracy; UAV remote sensing technology is easily limited by human operation, communication conditions, terrain and other factors. The use of unattended UAV platforms with automatic take-off and landing, data acquisition, storage and return functions, and equipped with hyperspectral sensors can overcome the limitations of the field environment and obtain high-resolution remote sensing image data, providing a data basis for the construction of accurate detection technology for discolored wood.

[0003] In the past, visual interpretation based on remote sensing images was highly subjective and inefficient. However, the advancement of graphics processing algorithms and computer hardware equipment has enabled computer-aided identification of discolored wood to be widely used in the field of pest and disease detection. The currently more commonly used method of identifying discolored wood based on single image features: color, texture, and shape is difficult to meet the accuracy requirements of pest and disease detection. Hyperspectral sensors can capture spectral information in more bands, provide more comprehensive and in-depth characteristics of the target to be detected, and help accurately identify soil nutrient distribution, vegetation health status, and pest and disease infestation.

[0004] In view of this, it is necessary to design a discolored wood detection method based on unattended UAV hyperspectral imaging to provide technical support for the effective prevention and control of diseases and pests. Summary of the invention

[0005] In view of the defects existing in the prior art, the purpose of the present invention is to provide a method for detecting discolored wood based on unattended drone hyperspectral imaging, so as to solve the problems of low efficiency and accuracy and insufficient data utilization of traditional detection methods of pests and diseases.

[0006] The technical solution adopted by the present invention is:

[0007] (1) An unmanned aerial vehicle platform equipped with a hyperspectral sensor obtains hyperspectral images of the target area and performs data preprocessing;

[0008] (2) Apply supervised classification algorithms to classify hyperspectral image objects and delineate vegetation areas;

[0009] (3) Use feature selection algorithm to screen three characteristic bands that are sensitive to detecting discolored wood;

[0010] (4) Calculate the triangular area enclosed by the reflectance values ​​corresponding to the three characteristic bands in the spectral curve and construct the area vegetation index;

[0011] (5) The PLS-DA partial least squares discriminant analysis model was constructed using the spectral data of discolored wood and healthy wood;

[0012] (6) Use the spectral data of discolored wood and healthy wood to train a one-dimensional convolutional neural network;

[0013] (7) Combining the above three methods to build an automatic recognition module;

[0014] (8) Inputting the data to be detected into the automatic recognition module to generate recognition results;

[0015] In a preferred embodiment of the method, the target area image is collected at regular time intervals by an unattended UAV platform, which consists of three parts: an intelligent UAV field, an electric vertical take-off fixed-wing UAV, and a hyperspectral multi-source remote sensing payload. The image is subjected to radiation calibration, atmospheric correction, and geometric correction standardization preprocessing to obtain spectral reflectance data of 240 bands in the wavelength range of 400-1000nm.

[0016] In a preferred embodiment of the method, the supervised classification algorithm is a parallelepiped or maximum likelihood algorithm, which classifies the land cover types and extracts the vegetation areas.

[0017] In a preferred embodiment of the method, the feature selection algorithm is a Relief-F feature selection algorithm, which selects three characteristic bands with large differences in spectral reflectance between discolored wood and healthy wood.

[0018] In a preferred embodiment of the method, the area vegetation index is constructed by connecting the corresponding points of the three characteristic bands on the spectral curve and calculating the area of ​​the enclosed triangle. The calculation formula is as follows:

[0019] PWDAI=|((ρ3-ρ1)*(ρ2-ρ1)*(ρ3-ρ2)-(λ3-λ1)*(λ2-λ1)*(λ3-λ2)) / 2|

[0020] Among them, ρ1, ρ2, and ρ3 represent the reflectivity values ​​of the three characteristic bands respectively; λ1, λ2, and λ3 represent the wavelengths of the three characteristic bands respectively.

[0021] In a preferred embodiment of the method, the spectral data of discolored wood and healthy wood are marked as different categories and a spectral data set is created. The training set and the test set are divided into a ratio of 3:1. The training set data is used to fit the PLS-DA discriminant analysis model, and the test set data is used to test the model accuracy.

[0022] In a preferred embodiment of the method, the Alex-Net framework is referred to and improved, and the spectral data of discolored wood and healthy wood with labels are used to construct a one-dimensional convolutional neural network suitable for one-dimensional spectral data;

[0023] In a preferred embodiment of the method, the area vegetation index detection method, the partial least squares discriminant analysis model detection method, and the one-dimensional convolutional neural network detection method are packaged into docker to construct an automatic recognition module for discolored wood;

[0024] In a preferred embodiment of the method, the drone hyperspectral data of the area to be detected is obtained, the data is input into the automatic identification module of the discolored wood, and the corresponding detection results of different detection methods are output.

[0025] Compared with the prior art, the advantages of the present invention are: 1. Compared with traditional drones, the unattended drone platform can cope with various weather conditions, realize automatic charging, automatic take-off and landing operations, and has the advantages of real-time monitoring; 2. The hyperspectral sensor can provide more comprehensive band information and richer data; 3. The automatic recognition module composed of the constructed new area vegetation index, PLS-DA discriminant analysis model, and AlexNet one-dimensional convolutional neural network has high detection accuracy, correctness, and efficiency for discolored wood. It can provide reliable technical support for the accurate identification of discolored wood. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for describing the embodiments.

[0027] Figure 1 The present invention is a flow chart of a method for detecting discolored wood based on unattended drone hyperspectral imaging. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0029] Please refer to Figure 1 , which is a flow chart of a method for detecting discolored wood based on unattended drone hyperspectral imaging according to the present invention.

[0030] A method for detecting discolored wood based on unattended UAV hyperspectral imaging includes the following steps:

[0031] (1) An unmanned aerial vehicle platform equipped with a hyperspectral sensor obtains hyperspectral images of the target area and performs data preprocessing;

[0032] (2) Apply supervised classification algorithms to classify hyperspectral image objects and delineate vegetation areas;

[0033] (3) Use feature selection algorithm to screen three characteristic bands that are sensitive to detecting discolored wood;

[0034] (4) Calculate the triangular area enclosed by the reflectance values ​​corresponding to the three characteristic bands in the spectral curve and construct the area vegetation index;

[0035] (5) The PLS-DA partial least squares discriminant analysis model was constructed using the spectral data of discolored wood and healthy wood;

[0036] (6) Use the spectral data of discolored wood and healthy wood to train a one-dimensional convolutional neural network;

[0037] (7) Combining the above three methods to build an automatic recognition module;

[0038] (8) Inputting the data to be detected into the automatic recognition module to generate recognition results;

[0039] Vegetation of different vegetation types, growth states, and disease infection levels have different spectral information data. Targeted color-changing wood detection is carried out through computer image processing technology and the extracted hyperspectral data.

[0040] Specific implementation cases:

[0041] First, an unattended UAV platform equipped with a hyperspectral sensor is used to collect hyperspectral data of a Masson pine forest infected with pine wood nematode disease in Wuyi County, Jinhua City, Zhejiang Province (N 28°40′30″, W 119°36′48″). Preferably, the hyperspectral sensor has a spectral resolution of 1.5nm and a wavelength range of 400-1000nm. The platform consists of three parts: an intelligent UAV field, an electric vertical fixed-wing UAV, and a hyperspectral multi-source remote sensing payload.

[0042] Furthermore, the acquired images were preprocessed with radiometric calibration, atmospheric correction, and geometric correction standardization to obtain spectral reflectance data of 240 bands within the wavelength range of 400-1000nm.

[0043] Furthermore, supervised classification algorithms such as parallelepiped or maximum likelihood algorithms are used to classify land cover types and extract vegetation areas.

[0044] Furthermore, we obtained the hyperspectral sample data of discolored wood and healthy wood in the target area, and used the Relief-F feature selection algorithm to screen out three characteristic bands with large differences in spectral reflectance between discolored wood and healthy wood: bands 62, 113, and 141.

[0045] Furthermore, the corresponding points of the reflectance values ​​of the three characteristic bands on the spectrum curve are connected to calculate the area of ​​the enclosed triangle. The calculation formula is as follows:

[0046] PWDAI=|((ρ 141 -ρ 62 )*(ρ 141 -ρ 113 )*(ρ 113 -ρ 62 )-(λ 141 -λ 62 )*(λ 141 -λ 113 )*(λ 113 -λ 62 )) / 2|

[0047] where ρ 62 , 113 , 141 Respectively represent the reflectivity values ​​of the three characteristic bands; 62 , 113 , 141 Represent the wavelengths of three characteristic bands respectively.

[0048] Furthermore, two characteristic tree categories, discolored wood and healthy wood, were extracted, each of which included 30 spectral reflectance information. A dataset containing two tree categories was created, and each dataset was randomly divided into a training set and a validation set in a ratio of 3:1. A PLS-DA discriminant model for discolored wood was established by appropriately rotating all bands of the spectral data marked with characteristic types.

[0049] Furthermore, referring to the Alex-Net framework and improving it, a spectral dataset containing two tree categories, discolored wood and healthy wood, was used to construct a one-dimensional convolutional neural network suitable for one-dimensional spectral data;

[0050] Furthermore, the area vegetation index detection method, partial least squares discriminant analysis model detection method, and one-dimensional convolutional neural network detection method were packaged into docker to build an automatic identification module for pine wood nematode diseased color wood;

[0051] Furthermore, unmanned drones are used to automatically obtain drone hyperspectral data in the detection area at certain time intervals, and the data is input into the automatic identification module of discolored wood to output the corresponding detection results of different detection methods. Periodic detection of discolored wood infected with pine wood nematode disease is achieved.

Claims

1. A method for constructing a new area vegetation index for identifying discolored wood, comprising the following steps: (1) An unmanned aerial vehicle platform equipped with a hyperspectral sensor obtains hyperspectral images of the target area and performs data preprocessing; (2) Supervised classification algorithms were used to classify objects in hyperspectral images and delineate vegetation areas. (3) Feature selection algorithms were used to select three characteristic bands that were sensitive to detecting discolored wood. (4) The area of ​​the triangle enclosed by the reflectance values ​​corresponding to the three characteristic bands in the spectral curve was calculated to construct an area vegetation index.

2. The method for constructing a new area index for identifying discolored wood according to claim 1, characterized in that: The step (1) is specifically as follows: the target area image is collected at regular time intervals by an unmanned aerial vehicle platform, the platform consists of three parts: an intelligent drone field, an electric vertical take-off fixed-wing drone, and a hyperspectral multi-source remote sensing payload. The image is subjected to radiation calibration, atmospheric correction, and geometric correction standardization preprocessing to obtain spectral reflectance data of 240 bands within a wavelength range of 400-1000nm.

3. The method for constructing a new area index for identifying discolored wood according to claim 1, characterized in that: The step (2) specifically includes: using a supervised classification algorithm to classify the land cover types based on the preprocessed hyperspectral image and extract the vegetation area.

4. The method for constructing a new area index for identifying discolored wood according to claim 1, characterized in that: The step (3) is specifically as follows: based on the spectral data of discolored wood and healthy wood extracted from the hyperspectral image, the Relief-F feature selection algorithm is used to screen out three characteristic bands with large differences in spectral reflectance between the discolored wood and the healthy wood.

5. The method for constructing a new area index for identifying discolored wood according to claim 1, characterized in that: The step (4) is specifically as follows: connecting the corresponding points of the three characteristic bands on the spectrum curve and calculating the area of ​​the enclosed triangle, and the calculation formula is as follows: PWDAI=|((ρ3-ρ1)*(ρ2-ρ1)*(ρ3-ρ2)-(λ3-λ1)*(λ2-λ1)*(λ3-λ2)) / 2| Among them, ρ1, ρ2, and ρ3 represent the reflectivity values ​​of the three characteristic bands respectively; λ1, λ2, and λ3 represent the wavelengths of the three characteristic bands respectively.

6. The method for constructing a new area index for identifying discolored wood according to claim 3, characterized in that: The supervised classification algorithm in step (2) is specifically: parallelepiped or maximum likelihood algorithm.

7. A method for detecting discolored wood using the vegetation index construction method of claim 1, comprising the following steps: (1) Constructing an area vegetation index based on the spectral characteristics of discolored wood and healthy wood; (2) The PLS-DA partial least squares discriminant analysis model was constructed using the spectral data of discolored wood and healthy wood; (3) Use the spectral data of discolored wood and healthy wood to train a one-dimensional convolutional neural network; (4) Combine the above three methods to build an automatic recognition module; (5) Input the data to be detected into the automatic recognition module to generate the recognition results.

8. The method for detecting discolored wood according to claim 7, characterized in that: The step (2) is specifically as follows: marking the spectral data of discolored wood and healthy wood into different categories and creating a spectral data set, dividing the training set and the test set in a ratio of 3:1, using the training set data to fit the PLS-DA discriminant analysis model, and using the test set data to perform a model accuracy test.

9. The method for detecting discolored wood according to claim 7, characterized in that: The step (3) is specifically as follows: using the labeled spectral data of discolored wood and healthy wood, referring to and improving the Alex-Net framework, and constructing a one-dimensional convolutional neural network suitable for one-dimensional spectral data.

10. The method for detecting discolored wood according to claim 7, characterized in that: The step (4) is specifically as follows: packaging the area vegetation index detection method, the partial least squares discriminant analysis model detection method, and the one-dimensional convolutional neural network detection method into docker to construct an automatic identification module for discolored wood.

11. The method for detecting discolored wood according to claim 7, characterized in that: The step (5) specifically includes: obtaining the drone hyperspectral data of the area to be detected, inputting the data into the automatic identification module of the discolored wood, and outputting the corresponding detection results of different detection methods.

Images