Fruit tree ecological index assessment method based on orthoimage and three-dimensional point cloud data

Abstract

The invention discloses a fruit tree ecological index assessment method based on an orthoimage and three-dimensional point cloud data, and the method comprises the following steps: obtaining an orthoimage map and three-dimensional point cloud data of a fruit tree region, carrying out the individual tree segmentation of the orthoimage map, and determining the individual tree region of each fruit tree; constructing a three-dimensional point cloud data set on the single tree area, performing point cloud denoising and layering, and outputting an independent layered three-dimensional point cloud of a single fruit tree; and calculating the core ecological index of the fruit tree by using the independent layered three-dimensional point cloud, so that the scheme is suitable for efficient monitoring and evaluation of a large-scale orchard, and the evaluation precision and efficiency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of remote sensing and agricultural informatization, and more particularly relates to a fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data. BACKGROUND

[0002] With the continuous expansion of fruit tree planting area in China, efficient evaluation of fruit tree ecological environment has become a key link to promote sustainable agricultural development. Traditional orchard ecological monitoring mainly relies on manual field investigation or simple vegetation index analysis, which is not only time-consuming and labor-intensive, but also the evaluation results are often low in quantitative degree and lack of description of the internal structure characteristics of the canopy, making it difficult to meet the needs of modern agricultural fine management. In recent years, unmanned aerial survey and three-dimensional point cloud acquisition technology have been widely used in the management of agricultural and forestry resources. Some studies use high-resolution aerial images to extract fruit tree canopy profile, estimate leaf area index, and monitor pests and diseases, etc. However, these methods are mainly based on two-dimensional images or simple statistical features, and pay insufficient attention to the three-dimensional structure and light transmittance of the fruit tree canopy. Most of the existing technology for fruit tree ecological evaluation is still limited to indicators such as chlorophyll content and coverage, lacking quantitative analysis of the ventilation and light transmittance characteristics of the canopy. At the same time, traditional methods are difficult to cover large-scale orchards quickly and automatically. Therefore, there is an urgent need for a new fruit tree ecological index evaluation method that can fully utilize orthophoto and three-dimensional point cloud data to extract multi-dimensional information of the fruit tree canopy, in order to achieve efficient and quantitative evaluation of the ecological status of fruit trees.

[0003] The existing technology with patent number CN119831158A proposes a forestry surveying and mapping method based on remote sensing big data. This method first integrates remote sensing images, unmanned aerial images, temperature sensors and humidity sensors data for spatio-temporal alignment processing, fills in missing data through time interpolation, detects outliers, calculates forest area ecological indicators, and outputs forest health ecological evaluation results. Then, based on the results, the forest species are identified, the contribution of species to ecological function is analyzed in combination with environmental factors, the distribution of key species is identified, and a species distribution ecological function contribution report is output. Then, based on the above two reports, the dynamic changes of forest ecology and resources are analyzed, the future forest ecological changes are predicted in combination with climate factors, and a forest resource future prediction report is obtained. Then, the spatial distribution of forest resources is analyzed, the resource utilization efficiency is evaluated, the high and low efficiency areas are identified and the difference reasons are analyzed, and a forest resource difference report is output. Finally, for the resource utilization low efficiency area, the resources are reconfigured, the potential ecological risk area is evaluated in combination with climate change and environmental monitoring data, the early warning and risk management are stimulated, and the forest resource management optimization and risk control scheme is formed. This method is based on two-dimensional images or simple statistical features, and pays insufficient attention to the three-dimensional structure and light transmittance of the fruit tree canopy, and lacks quantitative analysis of the ventilation and light transmittance characteristics of the canopy for fruit tree ecological evaluation, which is low in quantitative degree and cannot fully reflect the ecological status of fruit trees. SUMMARY

[0004] The present application provides a fruit tree ecological index evaluation method based on orthographic images and three-dimensional point cloud data to overcome the problems of insufficient quantitative evaluation of fruit tree ecology, ignoring the three-dimensional structure of the canopy based on two-dimensional data in the prior art.

[0005] The primary object of the present application is to solve the above technical problems, and the technical solution of the present application is as follows: The present application provides a fruit tree ecological index evaluation method based on orthographic images and three-dimensional point cloud data, comprising the following steps: Obtain the orthographic image of the fruit tree area and the three-dimensional point cloud data, perform single tree segmentation on the orthographic image, and determine the single tree area of each fruit tree; Construct a three-dimensional point cloud dataset on the single tree area, and perform point cloud denoising and layering to output an independent layered three-dimensional point cloud of a single fruit tree; Calculate the core ecological index of the fruit tree using the independent layered three-dimensional point cloud.

[0006] Further, the orthographic image is segmented to determine the single tree area of each fruit tree, comprising the following steps: Perform preliminary single tree segmentation on the orthographic image using a segmentation model, and perform vector boundary correction combined with manual visual interpretation to generate a fruit tree single tree classification map; cut the fruit tree single tree classification map into a plurality of image blocks of a predetermined size to obtain a single tree segmentation remote sensing image dataset; Train a dense instance segmentation network using the remote sensing image dataset to obtain a trained dense instance segmentation network; Input the orthographic image of the region to be detected into the trained dense instance segmentation network to output the instance segmentation result of the fruit tree; According to the instance segmentation result, extract the center point pixel coordinates and crown pixel area of each fruit tree, combine the image control point coordinates, and calculate and output the geographic coordinates and actual crown area of each fruit tree single tree area using an affine transformation model to determine the single tree area of each fruit tree.

[0007] Further, the dense instance segmentation network is a YOLO model, which is composed of the following modules: Feature extraction module: ResNet101 residual network is used to extract low-level spatial features and high-level semantic features, and a more accurate feature positioning is achieved through a dilated convolution pyramid pooling layer; ROI encoder module: including a convolutional neural network and a residual connection structure, used to enhance the feature expression ability of the features extracted by the feature extraction module; Candidate region generator: extracts regions that may contain targets from the ROI encoder, and selects multiple final candidate boxes through a non-maximum suppression (NMS) algorithm; The dense instance segmentation module: instance prediction is performed on each final candidate box, and class label and position information are output, and the matching quality of the instance set is optimized through the earth moving distance (EMD) loss function, and the segmentation accuracy is improved.

[0008] Further, a three-dimensional point cloud dataset is constructed on a single tree area, and point cloud denoising and layering are performed, and an independent layered three-dimensional point cloud of a single fruit tree is output, including the following steps: The geographical boundary of the single tree area is used to crop the corresponding single fruit tree point cloud region from the original three-dimensional point cloud data, and the original point cloud of the single fruit tree is output; The statistical outlier elimination algorithm based on mean and variance is used to determine and remove isolated abnormal points; the height threshold elimination method is used to remove abnormal high or low weed and ground noise points, and the denoised single tree point cloud is output; The hollow area in the denoised point cloud is filled by using surface fitting or voxel interpolation method to restore the integrity of the tree crown structure; Using the denoised point cloud, the highest point and the lowest point of the single fruit tree point cloud are determined, the total height difference of the tree crown is calculated, and the tree crown is vertically divided into n height layers; Repeat the above layering process for each fruit tree until all single fruit trees generate independent layered three-dimensional point cloud data.

[0009] Further, the tree crown is vertically and equally divided into six height layers, of which the first layer is the top crown layer, and the top crown layer is subdivided into five equal height sublayers; the second to fifth layers are projected horizontally according to eight sectors to count the average radius of the point cloud; the sixth layer is the ground layer.

[0010] Further, the core ecological indicators of the fruit tree are calculated using the independent layered three-dimensional point cloud, including the light transmittance index and the ventilation rate index of the fruit tree, and the calculation of the light transmittance index includes the following steps: Calculate the average radius of each independent layer except the ground layer and the overall average radius, and calculate the vertical difference index using the average radius of each layer and the overall average radius; Using the eight directional sectors of the second to fifth layer point cloud, the number of points in each sector is counted to determine the number of sparse directions in each sector, and the horizontal heterogeneity sparse index is obtained; The average radius difference degree of the five sublayers of the first layer top crown layer is calculated according to the calculation method of the vertical difference index to obtain the top sparseness index; The vertical difference index, the horizontal heterogeneity sparse index and the top sparseness index are summed to obtain a sum result T, which is nonlinearly converted by using a Sigmoid function to output the light transmittance index of the single fruit tree , the expression is as follows:

[0011] wherein, and respectively the first control function slope and the first central threshold; The calculation of the ventilation rate index comprises the following steps: The crown diameter of each fruit tree is used to construct a spatial buffer zone with the geometric center of the fruit tree as the center, and the height threshold is determined according to the difference between the digital terrain model and the digital elevation model to distinguish ground points and non-ground points, and calculate the ground exposure index , the expression is as follows:

[0012] wherein, is the number of ground points in the spatial buffer zone, is the number of non-ground points; The ground exposure index is nonlinearly mapped using the Sigmoid function to obtain the ventilation rate index , the expression is as follows:

[0013] wherein, and respectively the second control function slope and the second central threshold; The ventilation rate of a single fruit tree is obtained by correcting the ventilation rate using the light transmittance index.

[0014] Further, the vertical difference index The expression is as follows:

[0015] wherein, n is the total number of layers, is the weight corresponding to the ith layer, is the average radius of the ith layer, is the overall average radius.

[0016] Further, the calculation method of the horizontal heterogeneity sparse index comprises the following steps: Divide the point cloud of the second layer to the fifth layer into eight horizontal direction sectors; For each sector, count the total number of points N of the layer and the number of points If , it is determined that there is a sparse area in the sector, and the count is 1; if , the count is 0; Count the number of sectors with sparse areas; The horizontal heterogeneity sparse index is calculated using the number of sectors with sparse areas and the total number of sectors, and the expression is as follows:

[0017] wherein, is the total number of sectors of the second layer to the fifth layer, and 32 is the total number of sectors of the sparse area.

[0018] Further, the method for correcting the ventilation rate by using the light transmittance index is as follows: If the light transmittance of the fruit tree is , the ventilation rate is corrected and enlarged to obtain the enlarged ventilation rate The expression is as follows:

[0019] wherein, b is the enlargement multiple; If the light transmittance of the fruit tree is , the original value is maintained.

[0020] Compared with the prior art, the beneficial effects of the technical scheme of the present application are as follows: By fusing the unmanned aerial vehicle orthographic image and the three-dimensional point cloud data, the present method can fully utilize the two-dimensional and three-dimensional information to accurately depict the structural features of the fruit tree crown layer; the present application innovatively proposes to calculate the core ecological index of the fruit tree by using the independent layered three-dimensional point cloud, so that the fruit tree ecological evaluation is more comprehensive and quantitative; based on the fully automatic data processing flow, the present application is suitable for efficient monitoring and evaluation of large-scale orchards, and significantly improves the evaluation accuracy and efficiency, and overcomes the shortcomings of qualitative evaluation and limited coverage of the traditional method. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to make the purpose, technical scheme of the present application more clear, the present application provides the following drawings and makes the following description: Figure 1 The method flowchart provided for the embodiment of the present application; Figure 2 The data processing and analysis process schematic diagram provided for the embodiment of the present application; Figure 3 The fruit tree single tree segmentation schematic diagram provided for the embodiment of the present application; Figure 4 The fruit tree single tree orthographic image diagram provided for the embodiment of the present application; Figure 5 The fruit tree crown point cloud layering schematic diagram provided for the embodiment of the present application; Figure 6 The orchard fruit tree ecological environment thermal diagram schematic diagram provided for the embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable one skilled in the art to better understand the above objects, features and advantages of the present application, the application will be described in further detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0023] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can be practiced in other manners different from those described herein, and therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below.

[0024] Embodiment 1 The present application provides a fruit tree ecological index evaluation method based on orthographic images and three-dimensional point cloud data, as shown in Figure 1 As shown in the flow chart of the fruit tree ecological index evaluation method based on orthographic images and three-dimensional point cloud data, as shown in Figure 2 As shown in the data processing and analysis process schematic diagram of the present application, the specific steps are as follows: S1: Obtain the orthographic image of the fruit tree area and the three-dimensional point cloud data, perform single tree segmentation on the orthographic image, and determine the single tree area of each fruit tree.

[0025] In this embodiment, the fruit orchard of Feng'er base in Conghua District, Guangzhou City, Guangdong Province is selected as the research area of the embodiment. Under good lighting and windless weather conditions, multiple flights and multiple routes of automatic aerial survey are carried out on the target orchard by a UAV carrying a high-resolution RGB camera, and multi-view aerial images with an overlap rate of more than 80% are collected. Through the SfM (Structure-from-Motion) three-dimensional reconstruction method, combined with a multi-view stereo matching algorithm, a high-precision digital elevation model (DEM), a digital surface model (DSM), an orthographic image (Orthomosaic), and dense three-dimensional point cloud data of the orchard area are generated. The spatial resolution of the orthographic image reaches 0.05 meters, the point cloud density can reach 100 points / m², and it has the ability to describe the single tree crown structure in detail.

[0026] More specifically, the segmentation method is a single tree segmentation method based on multi-scale dense instance detection: the high spatial resolution of UAV images and the strong representation ability of deep learning models are used to realize efficient, automatic and fine processing of litchi single tree remote sensing recognition and spatial information extraction.

[0027] As shown in Figure 3 , Figure 4 The orthographic image is segmented to determine the single tree area of each fruit tree, including the following steps: First, the unmanned aerial vehicle remote sensing image of the research area is acquired and orthophoto map is generated. The flight route and flight parameters (including heading overlap rate, lateral overlap rate and flight height) of the target area are set by the task planning software, and the unmanned aerial vehicle automatically performs the flight task to acquire multiple images, and the image reconstruction software is used to generate the two-dimensional orthophoto map of the research area.

[0028] Secondly, a litchi tree single tree remote sensing image dataset is constructed. A segmentation model is used to preliminarily segment the generated orthophoto map, in this embodiment, the segmentation model adopts a YOLOv11 model with fixed parameters, and a vector boundary correction is performed in combination with visual interpretation by a surveying and mapping professional to generate a fruit tree single tree classification map; the fruit tree single tree classification map is cropped into a plurality of image blocks of a preset size, and is divided into a training set, a verification set and a test set according to a 7:1:2 ratio to obtain a single tree segmentation remote sensing image dataset.

[0029] The dense instance segmentation network is trained by using the remote sensing image dataset, and a trained dense instance segmentation network is obtained. The dense instance segmentation network is a YOLO model, the network takes ResNet101 as a feature extraction module backbone, realizes feature positioning in combination with a dilated convolution pyramid pooling layer, and uses an earth mover distance (EMD) loss function to optimize the matching quality of the instance set.

[0030] The orthophoto map of the to-be-detected area is input into the trained dense instance segmentation network, and the instance segmentation result (rectangular detection frame and corresponding pixel-level segmentation result) of the fruit tree is output. According to the instance segmentation result, the center point pixel coordinates and the crown pixel area of each fruit tree are extracted, the image control point coordinates are combined, the affine transformation model is used to calculate and output the geographic coordinates and the actual crown area of each fruit tree single tree area, and the single tree area of each fruit tree is determined.

[0031] The YOLO model framework is composed of the following modules: The feature extraction module: a ResNet101 residual network is used to extract low-level spatial features and high-level semantic features, and a dilated convolution pyramid pooling layer is used to realize more accurate feature positioning; The ROI encoder module: including a convolutional neural network and a residual connection structure, used to enhance the feature expression ability of the features extracted by the feature extraction module; The candidate region generator: extracts regions that may contain targets from the ROI encoder, and selects multiple final candidate frames through a non-maximum suppression (NMS) algorithm; The dense instance segmentation module: performs instance prediction on each final candidate frame, outputs class labels and position information, and optimizes the matching quality of the instance set through an earth mover distance (EMD) loss function to improve the segmentation accuracy.

[0032] After completion, the above network is trained using the constructed data set, in this embodiment, the number of training is set to 1000, the optimizer is stochastic gradient descent (SGD), the initial learning rate is 0.01, the training batch size is 4, the momentum is set to 0.9, and the weight decay is set to 0.0001.

[0033] S2: Construct a three-dimensional point cloud data set on the single tree area, and perform point cloud denoising and layering to output an independent layered three-dimensional point cloud of a single fruit tree.

[0034] The specific process is as follows: The geographical boundary of the single tree area is used to crop the corresponding single fruit tree point cloud area from the original three-dimensional point cloud data, and the original point cloud of the single fruit tree is output. To improve the quality of the point cloud, the following method is used for denoising: The statistical outlier elimination algorithm based on mean and variance includes: taking each point as the center, searching for a certain number of neighborhood points, in this embodiment, 20 neighborhood points are searched, and the average distance between the point and the neighborhood points is calculated and the standard deviation If the average distance of the point exceeds the global mean plus twice the standard deviation , it is considered as an outlier and is removed.

[0035] Utilize height threshold elimination: in this embodiment, according to experience, the lowest height of litchi tree leaves will be higher than 1m, and the near-ground height threshold is set to 1m, filtering points below this height, excluding grass, bare soil and other non-crown structure points, and outputting the denoised single tree point cloud; Use surface fitting or voxel interpolation method to fill the hollow area in the denoised point cloud, and restore the integrity of the tree crown structure; Using the denoised point cloud, determine the highest point and the lowest point of the single fruit tree point cloud, calculate the total height difference of the tree crown, and vertically divide the tree crown into n height layers; In this embodiment, it is found through experiments that it is more reasonable to vertically divide the tree crown into 6 height layers (L1-L6) with equal intervals, about 0.5-0.9m per layer, as shown in Figure 5 The tree crown point cloud layering diagram with an altitude of 74.9-80.1m is shown, the horizontal axis represents the longitude in the projection coordinates, and the vertical axis represents the latitude in the projection coordinates, wherein: The first layer L1 is the top crown layer, which is subdivided into five equal height sub-layers to enhance the description accuracy of the top structure; L2–L5 layer adopts the fan-shaped direction projection method to divide the horizontal plane into 8 sectors, and the average radius and density of the horizontal projection statistical point cloud are calculated according to the eight sectors as the cross-sectional structural characteristics of the layer to reflect the horizontal expansion trend and sparsity characteristics of the crown layer; The sixth layer L6 is a ground layer, which assists in identifying the trunk and the connecting structure of the bottom of the crown layer.

[0036] The above layering treatment is repeated for each fruit tree until all single fruit trees generate independent, hierarchical and clear layered three-dimensional point cloud data, which is used for crown structure modeling, biomass estimation and tree vigor evaluation and other orchard management application scenarios.

[0037] S3: calculating the core ecological indicators of the fruit tree using the independent layered three-dimensional point cloud.

[0038] The core ecological indicators of the fruit tree include the light transmittance indicator and the ventilation rate indicator. Based on the point cloud data, a calculation method for the two types of ecological indicators is proposed. The calculation of the light transmittance indicator includes the following steps: In order to quantitatively reflect the permeability and light utilization efficiency of the fruit tree crown structure, a light transmittance evaluation method based on three-dimensional point cloud data is proposed, which comprehensively describes the fruit tree shape and crown sparsity characteristics from three aspects of vertical distribution, horizontal heterogeneity and top sparsity.

[0039] Vertical difference index extraction: first, the average radius of each layer (L2–L5) of the fruit tree three-dimensional point cloud which has completed layering is calculated in turn. The calculation method is as follows: Taking the center of each layer as the reference, the point cloud of the layer is divided into 8 equal-angle fan directions (such as east, northeast, north, etc.), and the maximum horizontal distance of the points to the projection center of each layer is calculated in each direction. The average value of the 8 directions is taken as the average radius of the layer; The first layer (L1, top) is further divided into 5 equal-height sub-layers, and the average radius of each sub-layer is calculated by the above method; The average radius of the above layers (5 top sub-layers + L2–L5, a total of 9 layers) is denoted as , and the average radius of all sub-layers is calculated; The average radius of each layer and the overall average radius are calculated using the average radius of each height layer and the top crown layer sub-layer. The difference between the average radius of each layer and the overall average radius is obtained, which is combined with the layer sequence weight to accumulate layer by layer to obtain the vertical difference index, which is used to reflect the structural changes in the vertical direction of the fruit tree. The expression of the vertical difference index is as follows:

[0040] Wherein, n is the total number of layers, is the weight corresponding to the i-th layer. In this embodiment, the value of n is 9 (5 top sub-layers + L2–L5, a total of 9 layers). Set to equal weight or adjust by floor height, is the average radius of the i-th layer, is the overall average radius of all layers.

[0041] Divide the point clouds of the second to fifth layers into eight horizontal sectors; For each sector, count the total number of points N in the layer and the number of sector points ,like , then it is determined that there is a sparse area in the sector, and the count is 1; if , the count is 0; Counting the number to obtain the number of sectors with sparse areas; The horizontal heterogeneity sparsity index is calculated using the number of sectors with sparse areas and the total number of sectors. The expression is as follows:

[0042] in, is the total number of sectors with sparse areas, 32 is the total number of sectors from the second to the fifth layer, The larger the number, the more uneven the horizontal branches are distributed, and the better the light penetration.

[0043] At the top L1 layer, repeat the vertical difference index for its five equal height sublayers The calculation method of , separately obtains the top sparsity index ; This indicator reflects whether the top branches and leaves are concentrated or sparse. The larger the number, the sparser the top; Add the three sub-indicators to construct the overall light transmittance evaluation index , the expression is as follows:

[0044] Use the Sigmoid function for nonlinear transformation to output the transmittance index of a single fruit tree , the expression is as follows:

[0045] in, and The first control function slope and the first center threshold are respectively, in this embodiment, The value is 2.7, The value is 2.2. ,but ;like ,but ;like The final output is the light transmittance of each fruit tree for subsequent fruit tree shape optimization, planting density regulation or orchard spatial structure evaluation.

[0046] In order to comprehensively evaluate the air flow capacity and structural permeability of fruit trees in planting space, a ventilation rate evaluation method combining ground exposure and tree shape is proposed, including the following steps: The crown diameter of each fruit tree is used to construct a spatial buffer zone with the geometric center of the fruit tree as the center. The height threshold is determined according to the difference between the digital surface model (DSM) and the digital elevation model (DEM), which is 0.5 meters in this embodiment. The ground exposure index is calculated by distinguishing ground points and non-ground points, and the expression is as follows:

[0047] wherein, is the number of ground points in the spatial buffer zone, is the number of non-ground points, The larger the value, the more exposed the ground is, the more spacious the space around the fruit tree is, and the stronger the air flow is.

[0048] The ground exposure index is nonlinearly mapped using the Sigmoid function to obtain the ventilation rate index , and the expression is as follows:

[0049] wherein, and are the second control function slope and the second center threshold, respectively, which are 2.7 and -0.11 in this embodiment; if , then ; if , then ; if , then .

[0050] Since the degree of branch and leaf sparseness of the tree itself also affects the ventilation capacity, the calculated light transmittance is introduced as an adjustment factor, and the ventilation rate is corrected using the light transmittance index to obtain the ventilation rate index of a single fruit tree.

[0051] More specifically, the method for correcting the ventilation rate using the light transmittance index is as follows: If the light transmittance of the fruit tree , the ventilation rate is corrected and amplified to obtain the amplified ventilation rate , and the expression is as follows: ​

[0052] wherein b is the amplification factor, which is taken as 1.2 in this embodiment; If the light transmittance of the fruit tree is , the original value is maintained.

[0053] Finally, the ventilation rate of the fruit tree is output as a comprehensive evaluation index.

[0054] The method fuses the comprehensive features of the surrounding space shielding condition of the fruit tree and the structure of the tree body itself, and can effectively reflect the actual influence of the fruit tree spacing design on the ventilation environment of the orchard.

[0055] Geographic Information System (Geographic Information System or Geo-Information system, GIS) is sometimes also called "geographical information system". It is a specific and very important spatial information system. It is a technical system that collects, stores, manages, calculates, analyzes, displays and describes the geographical distribution data in the whole or part of the earth's surface (including the atmosphere) space under the support of computer hardware and software system.

[0056] The core ecological index is input into the geographic information system, and an ecological index heat map of the orchard scale is output. For the fruit trees with low ecological index, pruning and other agricultural suggestions are proposed to realize quantitative evaluation of the ecological condition of the orchard, such as Figure 6 Fig. 4 shows a schematic diagram of a fruit tree ventilation rate heat map.

[0057] An independent report is generated for each fruit tree, including its position, height, crown width, structural features, ventilation rate, light transmittance and hierarchical structure parameters. Among them, the position, height and crown width can be obtained through the single tree segmentation model, and the structural features and hierarchical structure can be obtained through the hierarchical method of point cloud.

[0058] By fusing the unmanned aerial vehicle orthographic image and the three-dimensional point cloud data, the method can fully utilize the two-dimensional and three-dimensional information, accurately depict the structural features of the fruit tree crown layer, innovatively propose to calculate the core ecological index of the fruit tree by using independent hierarchical three-dimensional point cloud, make the ecological evaluation of the fruit tree more comprehensive and quantitative, and based on the fully automatic data processing process, is suitable for efficient monitoring and evaluation of large-scale orchards, significantly improves the evaluation accuracy and efficiency, and overcomes the shortcomings of qualitative evaluation and limited coverage of traditional methods.

[0059] The method described in this embodiment has the characteristics of high automation and high precision, is suitable for fine monitoring of large-scale orchards, and can effectively improve the scientificity and efficiency of fruit tree health management, pest monitoring, pruning optimization and other operations.

[0060] It should be noted that the above step sequence can be appropriately adjusted according to application scenarios; the algorithm can be flexibly replaced or enhanced; the ventilation rate and the light transmittance can also be fused with a spectral vegetation index, a terrain factor, etc., to construct a multi-dimensional ecological evaluation model.

[0061] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the implementation modes of the present application. Based on the above description, other different forms of changes or variations can also be made by those skilled in the art. Here, all the implementation modes are not required or can not be exhausted. Any modification, equivalent replacement and improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A fruit tree ecological index evaluation method based on orthophoto and 3D point cloud data, characterized in that: The following steps are involved: Obtaining an orthophoto map and three-dimensional point cloud data of the fruit tree area, performing single-tree segmentation on the orthophoto map, and determining the single-tree area of ​​each fruit tree; Constructing a three-dimensional point cloud dataset on the single tree area, performing point cloud denoising and layering, and outputting an independent layered three-dimensional point cloud of the single fruit tree; The core ecological indicators of fruit trees are calculated using the independent layered three-dimensional point cloud.

2. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 1 is characterized in that: Performing tree segmentation on the orthophoto map to determine the single tree area of ​​each fruit tree includes the following steps: A segmentation model was used to perform preliminary tree segmentation on the orthophoto image, and vector boundary correction was performed based on manual visual interpretation to generate a fruit tree classification map. The fruit tree classification map was then cropped into several image blocks of preset sizes to obtain a single tree segmentation remote sensing image dataset. Using the remote sensing image dataset to train a dense instance segmentation network to obtain a trained dense instance segmentation network; Input the orthophoto of the area to be detected into the trained dense instance segmentation network and output the instance segmentation results of the fruit trees; According to the segmentation results of the example, the pixel coordinates of the center point and the canopy pixel area of ​​each fruit tree are extracted. Combined with the coordinates of the image control points, the geographic coordinates and actual canopy area of ​​each fruit tree are calculated and output using the affine transformation model to determine the single tree area of ​​each fruit tree.

3. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 2 is characterized in that: The dense instance segmentation network is a YOLO model, which consists of the following modules: Feature extraction module: uses the ResNet101 residual network to extract low-level spatial features and high-level semantic features, and uses a dilated convolutional pyramid pooling layer to achieve more accurate feature localization; ROI encoder module: includes a convolutional neural network and a residual connection structure to enhance the feature expression capability of the feature extraction module; Candidate region generator: extracts regions that may contain targets from the ROI encoder and selects multiple final candidate boxes using the non-maximum suppression algorithm; Dense instance segmentation module: performs instance prediction for each final candidate box, outputs category label and location information, and optimizes the matching quality of the instance set through the earth moving distance loss function.

4. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 1, characterized in that: Construct a 3D point cloud dataset on the single tree area, perform point cloud denoising and stratification, and output an independent layered 3D point cloud of a single fruit tree. The following steps are included: Use the geographic boundaries of the single tree area to crop the corresponding single fruit tree point cloud area from the original 3D point cloud data, and output the original point cloud of the single fruit tree; A statistical outlier removal algorithm based on mean and variance is used to identify and remove isolated outliers. A height threshold removal method is used to remove abnormally high or low weeds and ground noise points, and the denoised single tree point cloud is output. Use surface fitting or voxel interpolation methods to fill the void areas in the denoised point cloud and restore the integrity of the tree crown structure; Using the denoised point cloud, determine the highest and lowest points of the point cloud of a single fruit tree, calculate the total height difference of the tree crown, and vertically divide the tree crown into n height layers; Repeat the above layering process for each fruit tree until all individual fruit trees generate independent layered 3D point cloud data.

5. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 4 is characterized in that: The tree crown is divided into 6 height layers with equal vertical spacing. The first layer is the top canopy layer, which is further divided into five equal-height sub-layers. The second to fifth layers are horizontally projected according to eight sectors to calculate the average radius of the point cloud. The sixth layer is the ground layer.

6. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 5 is characterized in that: There are 6 levels with each level being 0.5-0.9m high.

7. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 5, characterized in that: The core ecological indicators of fruit trees calculated using independent layered 3D point cloud computing include the light transmittance index and ventilation rate index of the fruit trees. The calculation of the light transmittance index includes the following steps: Calculate the average radius of each independent layer except the ground layer and the overall average radius, and use the average radius of each layer and the overall average radius to calculate the vertical difference index; Using the eight direction sectors of the second to fifth layered point clouds, the number of points in each sector is counted to determine the number of sparse directions in each sector and obtain the horizontal heterogeneity sparsity index; For the average radius difference of the five sub-layers of the first top canopy layer, the top sparseness index is calculated according to the calculation method of the vertical difference index; The vertical difference index, horizontal heterogeneity sparse index and top sparsity index are summed to obtain the sum result T, and the Sigmoid function is used for nonlinear transformation to output the transmittance index of a single fruit tree. , the expression is as follows: in, and the first control function slope and the first center threshold, respectively; The calculation of the ventilation rate index includes the following steps: Using the crown diameter of each fruit tree, a spatial buffer zone is constructed with the geometric center of the fruit tree as the center of the circle. The height threshold is determined based on the difference between the digital surface model and the digital elevation model to distinguish between ground points and non-ground points, and the ground exposure index is calculated. , the expression is as follows: in, is the number of ground points in the spatial buffer, is the number of non-ground points; The Sigmoid function is used to perform nonlinear mapping on the ground exposure index to obtain the ventilation rate index. , the expression is as follows: in, and the second control function slope and the second center threshold, respectively; The ventilation rate was corrected using the light transmittance index to obtain the ventilation rate index of a single fruit tree.

8. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 7 is characterized in that: Vertical Difference Indicator The expression is as follows: Where n is the total number of layers, is the weight corresponding to the i-th layer, is the average radius of the i-th layer, is the overall average radius.

9. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 7, characterized in that: The calculation method of the horizontal heterogeneity sparseness index includes the following steps: Divide the point clouds of the second to fifth layers into eight horizontal sectors; For each sector, count the total number of points N in the layer and the number of sector points ,like , then it is determined that there is a sparse area in the sector, and the count is 1; if , the count is 0; Counting the number to obtain the number of sectors with sparse areas; The horizontal heterogeneity sparsity index is calculated using the number of sectors with sparse areas and the total number of sectors. The expression is as follows: in, is the sum of sectors with sparse areas, and 32 is the total number of sectors from the second to the fifth layers.

10. The fruit tree ecological index evaluation method based on orthophoto and three-dimensional point cloud data according to claim 7, characterized in that: The method of using the transmittance index to correct the ventilation rate is: If the light transmittance of the fruit tree , ventilation rate Perform correction and amplification to obtain the amplified ventilation rate The expression is as follows: Where b is the magnification; If the light transmittance of the fruit tree , the original value is maintained.

Images