A method for monitoring oak defoliating insect pests by using a UAV remote sensing

Abstract

The application discloses a kind of unmanned plane remote sensing oak class leaf-eating insect pest monitoring method, specifically comprising the following steps: S1, obtain the unmanned plane visible light and multispectral image of target area, to visible light image using convolution neural network method in deep learning, land use classification is carried out, entire area is classified according to woodland, farmland, construction land, water area, unused land, obtains woodland distribution boundary range;The application relates to the technical field of forest pest remote sensing monitoring.This unmanned plane remote sensing oak class leaf-eating insect pest monitoring method, unmanned plane aerial angle is big, observation range is wide, greatly reduces ground investigation workload, can realize the monitoring in forest farm level or county range, unmanned plane aerial image processing analysis result is more intuitive to show oak overall pest degree, can macroscopically control disaster situation on large area scale, reduce the error generated when manual measurement, obtain leaf loss rate from stereoscopic angle, more accurate and objective.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forest pest remote sensing monitoring, in particular to a method for monitoring oak defoliating pests by using unmanned aerial vehicle remote sensing. BACKGROUND

[0002] As one of the important factors that endanger forest resources in China, defoliating pests have the characteristics of large number of occurrence and high degree of damage. They can cause tree leaf reduction or defoliation, and even lead to tree death and affect forest ecological safety.

[0003] At present, the traditional method for monitoring oak defoliating pests is to rely on residents or forest rangers in the forest area to adopt ground reconnaissance to observe oak canopy changes, find abnormalities, and report to the local forest protection station for effective prevention and control measures. However, due to the rugged terrain of the mountainous area, this ground investigation method for confirming the occurrence of pests on site and judging the damage degree by perception is time-consuming and laborious, and has serious subjectivity and hysteresis. The traditional oak defoliating pest investigation and monitoring method has obvious shortcomings and deficiencies, mainly in that:

[0004] (1) Artificial investigation is difficult to accurately quantify the oak pest defoliation condition due to the limitation of the survey angle, and it is difficult to accurately define the boundary of the oak pest occurrence area.

[0005] (2) When calculating the leaf damage rate or pest leaf loss rate by the traditional method, the standard procedure needs to be obtained, which is complicated, has human error, and has low work efficiency, so it is difficult to carry out large-area oak defoliating pest monitoring.

[0006] (3) Oak pest occurrence has no periodicity, and traditional artificial investigation is difficult to monitor year after year, so the monitoring efficiency is low and the accuracy is not high.

[0007] Therefore, it is urgent to explore new technology for oak defoliating pest monitoring to improve the timeliness and effectiveness of monitoring. SUMMARY

[0008] (I) Technical problems solved

[0009] In view of the deficiencies of the prior art, the present application provides a method for monitoring oak defoliating pests by using unmanned aerial vehicle remote sensing, which solves the problem that the method of confirming the occurrence of pests on site and judging the damage degree by perception is time-consuming and laborious, and has serious subjectivity and hysteresis.

[0010] (II) Technical scheme

[0011] In order to achieve the above purpose, the present application is realized by the following technical scheme: a method for monitoring oak defoliating pests by using unmanned aerial vehicle remote sensing, specifically comprising the following steps:

[0012] S1, acquire unmanned aerial vehicle visible light and multispectral images of a target area, use a convolutional neural network method in deep learning on the visible light image to perform land use classification, classify the entire area according to woodland, cultivated land, construction land, water area, and unused land to obtain the distribution boundary range of woodland;

[0013] S2, in combination with the unmanned aerial vehicle image, the ground spectrum library, and the forest resource second-class survey data, identify Quercus aliena, Quercus sessilifolia, Quercus mongolica, and Quercus robur, and calculate the insect density and leaf loss rate at the single tree scale according to the survey method specified in the forestry pest occurrence and disaster standard;

[0014] S3, perform band calculation on the acquired multispectral image to construct vegetation index NDVI, enhanced vegetation EVI, RVI ratio vegetation index, improved red edge normalized vegetation index mNDVI750, leaf area index LAI, and digital elevation model DEM, use the vegetation index and digital elevation model to construct spectral information and topographic information index in Matlab, use the least squares support vector machine method optimized by particle swarm to establish a model according to the measured leaf loss value, determine the oak pest area and judge the oak disaster grade by field investigation and vegetation index judgment;

[0015] S4, periodically acquire unmanned aerial vehicle point cloud data of the pest area, segment three-dimensional models of pest trees in different periods, and calculate the single tree leaf loss rate by using the point cloud difference in different periods.

[0016] Preferably, the average single tree insect amount A in step S2 is calculated according to the following formula:

[0017] A=tn / t

[0018] In the formula:

[0019] A is the average single tree insect amount, with the unit of head / tree;

[0020] tn is the total insect amount, with the unit of head;

[0021] t is the total number of trees, with the unit of trees.

[0022] The single tree leaf loss rate calculation formula is:

[0023] LLR=(n i / t i )×100%

[0024] In the formula:

[0025] LLR is the leaf loss rate, with the unit of %;

[0026] n i is the amount of lost leaves on a single tree crown;

[0027] ti - the total amount of leaves on the crown of a single tree.

[0028] Preferably, in the step S3, the NDVI, mNDVI750, LAI, DEM, RVI, EVI are combined, and a model is established according to the measured leaf loss data by using the particle swarm optimization least squares support vector machine method, the learning factors are set as c1=1.7 and c2=5, and the effect is optimal when the number of iterations of the kernel function is 100, then the oak leaf-feeding pest damage grade is judged according to the constructed model, and the standard for judging the oak disaster grade is that when the leaf loss rate is less than 20%, it is determined to be light disaster, when the leaf loss rate is greater than 20% and less than 60%, it is determined to be moderate disaster, and when the leaf loss rate is greater than 60%, it is determined to be heavy disaster.

[0029] Preferably, in the step S4, the unmanned aerial vehicle oblique photography technology is used to periodically obtain the three-dimensional point cloud of the pest area, when the leaf growth of the forest reaches 7 months before the disaster, the three-dimensional point cloud of the standard forest in the area with the canopy density less than 0.3 is obtained by using the unmanned aerial vehicle oblique photography method, the three-dimensional point cloud of the standard forest in the area with the canopy density higher than 0.3 is obtained by using the handheld three-dimensional laser scanner, and the three-dimensional point cloud of the forest in the light, moderate and heavy periods is collected respectively, the point cloud segmentation and clustering are performed by using the PointCNN convolutional neural network method, the forest leaf inclination angle, the lower branch index and the forest fractal dimension parameters are constructed by using the TIN irregular triangle network method, the point cloud data in the light, moderate and heavy periods and the standard point cloud data before the disaster are subtracted respectively, the subtracted point cloud is separated, and the three-dimensional leaf loss of the forest is calculated according to the leaf inclination angle, the lower branch index and the forest fractal dimension.

[0030] Preferably, in the step S4, the Pix4D software is used to process the unmanned aerial vehicle point cloud data, and the point cloud data of the handheld laser radar is imported into the LiDAR360 software for point cloud denoising, ground point separation, DEM generation, normalization processing, seed point generation, seed point import and editing, point cloud segmentation based on the edited point cloud and the like, the point cloud data is input into the PointCNN, and is represented as: F1={(p1,i,f1,i):i=1,2,…,N1}, that is, a group of point sets and a feature set corresponding to each point C1 represents the initial feature channel depth, and the PointCNN core operation is X convolution, which can be simply written as: F p =X-Conv(K,p,P,F)=Conv(K,MLP(P-p)×[MLP δ (P-p)F]), for the segmentation task, high-resolution point-by-point output is required, which is realized by constructing the PointCNN after Conv-DeConv.

[0031] (Three) beneficial effects

[0032] The present application provides a kind of unmanned aerial vehicle remote sensing oak defoliator monitoring method.It has the following beneficial effects:

[0033] (1), the unmanned aerial vehicle remote sensing oak defoliator monitoring method, unmanned aerial vehicle flight angle is large, observation range is wide, image acquisition difficulty is low, greatly reduces the ground investigation workload, can realize the monitoring of forest field or county range.

[0034] (2), the unmanned aerial vehicle remote sensing oak defoliator monitoring method, unmanned aerial vehicle flight image processing analysis result is more intuitive to show the overall degree of oak insect pests, can macroscopically control disaster situation on large area scale.

[0035] (3), the unmanned aerial vehicle remote sensing oak defoliator monitoring method, by three-dimensional modeling of the unmanned aerial vehicle on the insect pest wood estimation leaf loss rate, greatly reduces the error generated when manual measurement, obtains leaf loss rate from stereoscopic angle, is more accurate and objective.

[0036] (4), the unmanned aerial vehicle remote sensing oak defoliator monitoring method, the harm degree of insect pest area is divided, the sensitive band of heavy disaster area is extracted, can monitor the occurrence of oak area insect pests for years, improves the prevention efficiency. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described clearly and completely below.It is obvious that the described embodiments are only part of the embodiments of the present application, not all the embodiments.Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0038] The present application provides a kind of technical scheme: a kind of unmanned aerial vehicle remote sensing oak defoliator monitoring method, specifically includes the following steps:

[0039] S1, acquire unmanned aerial vehicle visible light and multispectral images of the target area, use convolutional neural network method in deep learning to classify the land use of the visible light image, classify the entire area according to woodland, cultivated land, construction land, water area and unused land, acquire the distribution boundary range of woodland, use the unmanned aerial vehicle equipped with visible light (380nm-760nm), red light (650nm±16nm), green light (560nm±16nm), blue light (450nm±16nm), red edge (730nm±16nm) and near-infrared (840nm±26nm) band lenses to take aerial images of the area suspected to have oak leaf-eating pest damage, set the heading overlap rate to be more than 80% and the lateral overlap rate to be more than 75% during the flight, complete the acquisition of remote sensing image data of the target area, process the acquired unmanned aerial vehicle images according to the structure from motion (sfm algorithm) using DJI ZhiTu software or Pix4D software, acquire the digital orthographic image DOM, digital surface model DSM, point cloud data of each band, red band target area image, green band target area image, blue band target area image, red edge band target area image and near-infrared band target area image of the target area;

[0040] S2, combine the unmanned aerial vehicle images, ground spectrum library and forest resources second-class survey data to identify oak species such as Quercus variabilis, Quercus acutissima, Quercus mongolica and Cyclobalanopsis glauca, calculate the insect density and leaf loss rate at the single tree scale according to the investigation method specified in the forestry harmful organism occurrence and disaster standard, perform land use classification on the acquired unmanned aerial vehicle orthographic image, identify the woodland area, use eCognition software to perform object-oriented target segmentation on the orthographic image, set appropriate scale parameters, shape parameters and compactness parameters for optimal parameter segmentation according to the characteristics of the study area, and complete the extraction of woodland information by using NDVI, spectrum, texture, size and spatial relationship in combination with support vector machine and random forest method;

[0041] S3, perform band calculation on the acquired multispectral image, construct vegetation index NDVI, enhanced vegetation EVI, RVI ratio vegetation index, improved red edge normalized vegetation index mNDVI750, leaf area index LAI and digital elevation model DEM, and the calculation methods of each index are as follows:

[0042] NDVI=(NIR-R) / (NIR+R), EVI=2.5(NIR-R) / (NIR+6R-7.5B+1), RVI=NIR / R, mNDVI 750 =(NIR 750 -R) / (NIR 750 +R),

[0043]

[0044] In Matlab, spectral information and terrain information index are constructed by using vegetation index and digital elevation model, a model is established according to the measured leaf loss value by using the particle swarm optimization least squares support vector machine method, the leaf loss rate is judged by field investigation and vegetation index, the oak insect damage area is determined, and the oak disaster grade is judged.

[0045] S4, periodically acquiring unmanned aerial vehicle point cloud data of the insect damage area, segmenting three-dimensional models of insect damaged trees in different periods, and calculating the single tree leaf loss rate by using point cloud difference in different periods.

[0046] In the embodiment of the application, the average single tree insect amount A in step S2 is calculated according to the following formula:

[0047] A=tn / t

[0048] In the formula:

[0049] A is the average single tree insect amount, and the unit is head / plant;

[0050] tn is the total insect amount, and the unit is head;

[0051] t is the total plant number, and the unit is plant.

[0052] The single tree leaf loss rate calculation formula is:

[0053] LLR=(n i / t i )×100%

[0054] In the formula:

[0055] LLR is the leaf loss rate, and the unit is %;

[0056] n i is the amount of lost leaves on a single tree crown;

[0057] t i is the total amount of leaves on a single tree crown.

[0058] In the embodiment of the application, in step S3, NDVI, mNDVI750, LAI, DEM, RVI and EVI are combined, a model is established according to the measured leaf loss data by using the particle swarm optimization least squares support vector machine method, the learning factors c1 and c2 are set to 1.7 and 5 respectively, the effect is optimal when the kernel function iteration number is 100, then the oak leaf-eating pest damage grade is judged according to the constructed model by using the following method, and the standard for judging the oak disaster grade is that when the leaf loss rate is less than 20%, it is determined that the disaster is light; when the leaf loss rate is greater than 20% and less than 60%, it is determined that the disaster is moderate; and when the leaf loss rate is greater than 60%, it is determined that the disaster is severe.

[0059] In this embodiment of the invention, step S4 utilizes UAV oblique photography technology to periodically acquire 3D point clouds of the pest-affected area. Before the disaster, in July when the tree leaf growth reaches its peak, for areas with a canopy closure of less than 0.3, standard tree 3D point clouds are acquired using UAV oblique photography. To ensure consistent flight parameters for different pest-affected areas, the specific parameters for the UAV are: flight altitude 80m, forward overlap 80%, lateral overlap 70%, flight speed 9.2m / s, and the UAV model is an industry-grade surveying UAV. For areas with a canopy closure higher than 0.3, standard tree 3D point clouds are acquired using a handheld 3D laser scanner. An S-shaped route was set up in the oak area using lidar, with a standard route interval of 3 meters. Three-dimensional point clouds of trees were collected during the light, moderate, and severe periods of disaster. The PointCNN convolutional neural network method was used for point cloud segmentation and clustering. The segmented and clustered point cloud data were imported into ArcGIS 10.2, and the parameters of tree leaf inclination angle, lower branch index, and tree fractal dimension were constructed using the TIN irregular triangular network method. The point cloud data of the light, moderate, and severe periods were subtracted from the standard point cloud data before the disaster, and the subtracted point clouds were separated. The three-dimensional leaf loss of trees was calculated based on the leaf inclination angle, lower branch index, and tree fractal dimension.

[0060] In this embodiment of the invention, step S4 uses Pix4D software to process the UAV point cloud data. Point cloud data from the handheld LiDAR is imported into LiDAR360 software and preprocessed, including point cloud denoising, ground point separation, DEM generation, normalization, seed point generation, importing and editing seed points, and point cloud segmentation based on the edited data. The point cloud data is then input into PointCNN, represented as: F1 = {(p1,i,f1,i):i = 1, 2, ..., N1}, i.e., a set of points. and the feature set corresponding to each point C1 represents the initial feature channel depth. The core operation of PointCNN is the X convolution, which can be abbreviated as: F p =X-Conv(K,p,P,F)=Conv(K,MLP(Pp)×[MLP δ (Pp)F]), for segmentation tasks, high-resolution point-by-point output is required, which is achieved by building PointCNN after Conv-DeConv.

[0061] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0062] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring oak defoliating insect pests by using a UAV remote sensing system, characterized in that: Specifically comprising the following steps: S1, acquiring unmanned aerial vehicle visible light and multispectral images of the target area, using a convolutional neural network method in deep learning to classify the land use of the visible light image, classifying the entire area according to woodland, cultivated land, construction land, water area, and unused land to obtain the distribution boundary range of woodland; S2, combining the unmanned aerial vehicle images, the ground spectral library, and the forest resource second-class survey data to identify Quercus variabilis, Quercus sessilifolia, Quercus mongolica, and Quercus aliena, and calculating the insect density and leaf loss rate at the single tree scale according to the survey method specified in the forestry harmful organism occurrence and disaster standard; S3, performing band calculation on the acquired multispectral images to construct vegetation index NDVI, enhanced vegetation EVI, RVI ratio vegetation index, improved red edge normalized vegetation index mNDVI750, leaf area index LAI, and digital elevation model DEM, constructing spectral information and topographic information index in Matlab, establishing a model according to the measured leaf loss value using the particle swarm optimization least squares support vector machine method, determining the oak insect damage area and judging the oak disaster grade by field investigation and vegetation index judgment; S4, periodically acquiring unmanned aerial vehicle point cloud data in the insect damage area, segmenting three-dimensional models of trees in different periods of insect damage, and calculating the single tree leaf loss rate using point cloud difference in different periods; In the step S4, the three-dimensional point cloud in the insect damage area is periodically acquired using the unmanned aerial vehicle oblique photography technology. When the leaf growth of the trees reaches 7 months before the disaster, the three-dimensional point cloud of standard trees is acquired in the area with a canopy density less than 0.3 using the unmanned aerial vehicle oblique photography method, the three-dimensional point cloud of standard trees is acquired in the area with a canopy density higher than 0.3 using a handheld three-dimensional laser scanner, the three-dimensional point cloud of trees in the light, moderate, and severe periods is acquired, respectively, the point cloud is segmented and clustered using the PointCNN convolutional neural network method, the tree leaf inclination, lower branch index, and tree fractal dimension parameters are constructed by the TIN irregular triangle network method, and the point cloud after the difference is separated by using the point cloud data in the light, moderate, and severe periods and the standard point cloud data before the disaster, and the three-dimensional leaf loss of the trees is calculated according to the leaf inclination, lower branch index, and tree fractal dimension. 2.The method according to claim 1, wherein: The average single worm amount A in the step S2 is calculated by the following equation: In the formula: A is the average single tree insect amount, with the unit of head / plant; tn is the total insect amount, with the unit of head; t is the total number of plants, with the unit of plant; Single tree mortality rate calculation formula: In the formula: LLR is the leaf loss rate, with the unit of %; — the amount of leaves lost on individual tree crowns; - total leaf amount on a single tree crown. 3.The method of claim 1, wherein the method further comprises: determining a location of the UAV; and determining a location of the acorn moth based on the location of the UAV. In the step S3, NDVI, mNDVI750, LAI, DEM, RVI, and EVI are combined, a model is established according to the measured leaf loss value data using the particle swarm optimization least squares support vector machine method, the learning factors c1=1.7 and c2=5 are set, the kernel function iteration number is 100, the effect is optimal, the oak leaf-eating pest damage grade is judged according to the constructed model using the following method, and the oak disaster grade is judged according to the following standard: when the leaf loss rate is less than 20%, it is determined to be light disaster; when the leaf loss rate is greater than 20% and less than 60%, it is determined to be moderate disaster; and when the leaf loss rate is greater than 60%, it is determined to be severe disaster.

4. The method according to claim 1, wherein the method is characterized by: The step S4 uses Pix4D software to process the unmanned aerial vehicle point cloud data, and the point cloud data of the handheld laser radar is imported into LiDAR360 software to perform point cloud denoising, ground point separation, DEM generation, normalization processing, seed point generation, seed point import and editing, segmentation preprocessing based on the edited point cloud, and input of the point cloud data into PointCNN, which is represented as: F1={(p1,i, f1,i): i=1, 2, …, N1}, that is, a set of point sets and a feature set corresponding to each point , C1 represents the initial feature channel depth, and the PointCNN core operation is X convolution, which can be simply written as: F p = X - Conv(K,p,P,F)= Conv(K, MLP(P - p)× [MLP δ (P - p) F]), for the segmentation task, a high-resolution point-by-point output is required, which is realized by constructing PointCNN after Conv-DeConv.

Images