Desert shrub identification method based on YOLOv8 model

Abstract

The invention discloses a desert shrub identification method based on a YOLOv8 model, and the method comprises the steps: collecting an unmanned plane remote sensing image of a ground shrub plant for a sample plot selected in a desert gobi desert; splitting the unmanned aerial vehicle remote sensing image into a plurality of sub-images and numbering; on the basis of a YOLOv8 model, a sample data set constructed by desert forest remote sensing images shot by an unmanned aerial vehicle and ground measurement data is used for training, and a target detection model is obtained; detecting a group of sub-graphs through the target detection model to obtain a detection result, and recording the plant species and the crown breadth; combining the group of sub-images into a fused image according to the numbers corresponding to the group of sub-images and the position information of the detection frame; and determining the plant coverage of the sample plot on the fused image based on the plant species and the crown breadth size. The overall precision of the current model is higher than the expected precision of other target detection models, and the method is suitable for desert forest data sets.

Description

Technical Field

[0001] The present invention relates to the field of desert forest monitoring, and in particular to, but not limited to, a desert shrub identification method based on a YOLOv8 model. Background Art

[0002] Unmanned aerial vehicle (UAV) remote sensing technology has been widely used in ecological monitoring, agricultural management and environmental assessment due to its advantages of high resolution, high timeliness and low cost. The high-resolution camera carried by UAVs can capture large-area surface images, providing a new data source for vegetation monitoring.

[0003] In terms of desert shrub identification, traditional remote sensing classification methods and low spatial resolution remote sensing images are difficult to meet the needs of shrub classification and identification in desert areas, and the accuracy of extracted patches is low. In order to improve the recognition accuracy, researchers began to explore high-resolution remote sensing images based on drones, combined with field ground survey data, and used object-oriented manually defined rules, Bayesian, K nearest neighbors, decision trees, random forests and other methods to identify and extract shrub species in desert areas. These methods usually have low recognition accuracy and high computational costs. Therefore, it is urgent to provide a new method for desert shrub identification and provide technical support for smart agriculture and environmental management. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a desert shrub identification method based on the YOLOv8 model to realize automatic plant species identification and coverage calculation, reduce manual intervention, and improve the efficiency and accuracy of monitoring work.

[0005] The technical solutions of the embodiments of the present invention are as follows:

[0006] An embodiment of the present invention provides a method for identifying desert shrubs based on a YOLOv8 model, the method comprising:

[0007] For the sample plots selected in the desert Gobi, drone remote sensing images of ground shrubs are collected; the drone remote sensing images are split into multiple sub-images according to a preset overlap rate and numbered in row and column order; based on the YOLOv8 model, a target detection model is obtained by training using a sample data set constructed using remote sensing images of desert forests taken by drones and ground measurement data; wherein the ground measurement data includes the latitude and longitude coordinates, height, crown width and ground diameter of the plants; the group of sub-images are detected by the target detection model to obtain the detection results and record the plant species and crown width size; the group of sub-images are merged into a fused image according to the corresponding numbers of the group of sub-images and the position information of the detection frame; based on the plant species and the crown width size, the total plant area and the sample surface area are respectively calculated on the fused image, and the plant coverage of the sample plot is determined.

[0008] Furthermore, the sample data set is constructed through the following process: obtaining a set of visible light images of desert forest plants in the target sample plot when the drone is flying at an altitude of 20 meters; using pix4d to stitch the set of visible light images according to the drone flight route and then crop them to a uniform size, eliminating images that do not contain target plants, and obtaining a desert forest remote sensing image of the entire sample plot; using ArcGIS to correspond the plant longitude and latitude information collected by the RTK positioning device with the longitude and latitude information contained in the collected image, and combining the recorded plant information to mark the plant species in the desert forest remote sensing image; using Labelme software to mark the plant contours in the desert forest remote sensing image, constructing the sample data set, and dividing it into training set, test set, and validation set in a ratio of 6:2:2.

[0009] Furthermore, the method also includes: during the labeling process, plants with a crown width smaller than a fixed threshold are not labeled and are treated as background.

[0010] Furthermore, merging the group of sub-images into a fused image according to the numbers corresponding to each of the group of sub-images and the position information of the detection frames includes: positioning all detection frames in the group of sub-images on the UAV remote sensing image according to the position information of the detection frames, and determining the intersection-and-union ratio between two overlapping detection frames; when the two detection frames are of the same type and the intersection-and-union ratio is greater than 0.5, fusing the two detection frames to obtain a new detection frame, the size of which covers the two detection frames; merging the group of sub-images according to the numbers corresponding to each of the group of sub-images.

[0011] Furthermore, the total plant area and the sample plot area are calculated based on the fused image, and the plant coverage of the sample plot is determined, including: taking the length and width of the detection frame as the major axis and minor axis of the corresponding plant, respectively, and calculating the projection area of ​​the plant according to the ellipse area calculation formula; dividing the cumulative sum of the projection areas of all plants by the sample plot area to obtain the plant coverage of the sample plot.

[0012] Furthermore, the method also includes: using average accuracy as an evaluation indicator for each shrub species and setting the intersection-over-union ratio to at least 40% to be considered a successful detection; taking a weighted average of the average accuracy of all species to obtain the overall accuracy of the model, wherein the weight coefficient is the proportion of the number of corresponding tree species in the sample data set; and using the overall accuracy to evaluate the performance of the target detection model.

[0013] Furthermore, the plant species include at least the following desert shrubs: Haloxylon ammodendron, Tamarix chinensis, Ephedra sinica, Aralia dasyphylla, and Calligonum mongolica.

[0014] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0015] In an embodiment of the present invention, by improving the YOLOv8 model and combining it with UAV remote sensing imaging technology, a new desert shrub identification method is proposed, aiming to improve the recognition accuracy, reduce the computing cost, and provide technical support for smart agriculture and environmental management. The high accuracy of the YOLOv8 model and the improved image processing technology are used to improve the accuracy of plant species identification and crown width measurement. At the same time, through accurate plant coverage and species distribution information, effective monitoring and evaluation of vegetation in desert Gobi sample plots can be achieved, providing a scientific basis for ecological protection and environmental management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0017] Figure 1 A schematic diagram of a process for identifying desert shrubs based on a YOLOv8 model provided in an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of the YOLOv8 model structure provided in an embodiment of the present invention;

[0019] Figure 3 Visible light images collected at different flight altitudes provided by an embodiment of the present invention;

[0020] Figure 4 The multispectral images collected at different flight altitudes provided by the embodiment of the present invention;

[0021] Figure 5 A simulated diagram of the flight route of a drone provided in an embodiment of the present invention;

[0022] Figure 6 A schematic diagram of the stitching results of UAV remote sensing images provided by an embodiment of the present invention;

[0023] Figure 7 A schematic diagram of plant numbers marked with ArcGIS provided in an embodiment of the present invention;

[0024] Figure 8 A schematic diagram of a plant outline annotated with Labelme provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0027] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0028] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the field to which the embodiments of the present invention belong. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0029] Figure 1 A schematic diagram of a process for identifying desert shrubs based on a YOLOv8 model provided in an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method comprises at least the following steps:

[0030] Step S110, collecting drone remote sensing images of ground shrubs from sample plots selected in the desert Gobi.

[0031] Here, the selection of sample plots requires that the number of shrub plants be between 500 and 5,000. For example, a 100m (meter) × 100m (meter) area is selected in the desert Gobi as the sample plot of the desert. Use the camera carried by the drone to take aerial photos in the desert forest area to obtain high-resolution remote sensing images.

[0032] In order to explore the influence of shooting height on image detection, 20m, 30m and 50m were used as the UAV flight heights to shoot the ground plants. When the flight height is low, shooting is more time-consuming. Therefore, when the height is 20m, the shooting range is set to 40m×40m, and when the height is 30m and 50m, the shooting range is set to 100m×100m.

[0033] The embodiment of the present invention does not limit the types of drones and the sensors carried by them. Preferably, the drone remote sensing images of the desert shrub forest are collected using the DJI Phantom Multispectral Edition, the DJI Phantom Visible Light Edition, and the DJI M4 drone.

[0034] Step S120: split the UAV remote sensing image into multiple sub-images according to a preset overlap ratio and number them in order of rows and columns.

[0035] Here, since the images taken by drones are too large and have extremely high resolution, generally hundreds of millions or billions of pixels, and the model running time increases exponentially with the number of pixels, if the images are directly entered into the model, the detection time will be greatly increased. Therefore, it is necessary to split the original drone remote sensing images into 1024×1024 sub-images, i.e. small images, and number them in row and column order for easy merging. In this way, the orderliness and systematicness of subsequent processing are ensured.

[0036] The setting of the preset overlap rate needs to ensure that the plant will not be split, resulting in a decrease in model accuracy, which is not limited in the embodiment of the present invention. Preferably, the preset overlap rate between each small image is set to 20% to avoid the plant being split.

[0037] Step S130, based on the YOLOv8 model, a sample data set constructed by using remote sensing images of desert forests taken by drones and ground measurement data is trained to obtain a target detection model.

[0038] Here, the ground measurement data includes the latitude and longitude coordinates, height, crown width and ground diameter of the plant, which can provide the accurate position and size information of the shrub plant in the desert forest remote sensing image. The ground measurement data in the embodiment of the present invention uses Beidou RTK equipment to obtain the longitude and latitude information of the shrub plant, and uses a tape measure to measure the height, crown width and ground diameter of the plant.

[0039] In the embodiment of the present invention, an improved model based on the YOLOv8 model is used to detect targets in remote sensing images of desert forests taken by drones. YOLO (You Only Look Once) is a one-stage target detection algorithm, which means that the category and bounding box of objects in an image can be identified with only one scan. YOLOv8 is the latest target detection algorithm of the Yolo series launched by Ultralytics, which can be used for tasks such as image classification, object detection, and instance segmentation. The model structure is as follows Figure 2 In the implementation, you can choose a suitable YOLOv8 model architecture, such as YOLOv8n, YOLOv8s, etc., and choose the appropriate model size according to project requirements and resource constraints.

[0040] Based on the ground measurement data, the collected desert forest remote sensing images are manually or semi-automatically annotated to determine the location and category of shrubs, generate annotation files, and thus construct a sample data set, which is then divided into training set, test set, and validation set. The YOLOv8 model is trained using the training set and test set data, and the model parameters, such as learning rate, batch size, etc., are adjusted using the validation set data, and the model performance is monitored.

[0041] Step S140: Detect the group of sub-images using the target detection model to obtain detection results and record plant species and canopy sizes.

[0042] Here, in the implementation, the trained model is deployed in practical applications, such as being integrated into the automatic navigation system of a drone to achieve real-time target detection and automated vegetation monitoring. The detection result includes the detection frame, confidence and species information of the plants contained in the drone remote sensing image.

[0043] First, the sub-images of the UAV remote sensing images are preprocessed as necessary, including scaling and normalization, to match the input requirements of the model training. Then, the preprocessed sub-images are input into the target detection model, and the model will output the detection box, confidence, and predicted category label of the plant in each sub-image. The detection results are parsed, the detection box coordinates of each plant are extracted, and then the crown size is estimated based on the size of the detection box. The crown size usually refers to the width and height of the plant crown, which can be approximated by the width and height of the detection box.

[0044] In other implementations, the detection results of the UAV remote sensing image can also be verified in combination with ground measured data to ensure the accuracy of the target detection model recognition.

[0045] Step S150: merging the group of sub-images into a fused image according to the serial numbers corresponding to the group of sub-images and the position information of the detection frame.

[0046] Here, the detection results of a set of sub-images are merged back into the original UAV remote sensing image using image alignment and stitching techniques to obtain a fused image corresponding to the entire sample site to ensure the continuity and consistency of the image.

[0047] Step S160, based on the plant species and the crown size, respectively calculate the total plant area and the sample plot area on the fused image, and determine the plant coverage of the sample plot.

[0048] Here, image analysis software can be used to estimate the total area of ​​vegetation coverage by accumulating the crown area of ​​all plants. The ratio of the total plant area to the plot area is taken as the plant coverage of the plot.

[0049] In an embodiment of the present invention, by improving the YOLOv8 model and combining it with UAV remote sensing imaging technology, a new desert shrub identification method is proposed, aiming to improve the recognition accuracy, reduce the computing cost, and provide technical support for smart agriculture and environmental management. The high accuracy of the YOLOv8 model and the improved image processing technology are used to improve the accuracy of plant species identification and crown width measurement. At the same time, through accurate plant coverage and species distribution information, effective monitoring and evaluation of vegetation in desert Gobi sample plots can be achieved, providing a scientific basis for ecological protection and environmental management.

[0050] In some embodiments, the sample data set is constructed by the following process: obtaining a set of visible light images of desert forest plants in the target sample plot when the drone is flying at an altitude of 20 meters; using pix4d to stitch the set of visible light images according to the drone flight route and then crop them to a uniform size, eliminating images that do not contain target plants, and obtaining a desert forest remote sensing image of the entire sample plot; using ArcGIS to correspond the plant longitude and latitude information collected by the RTK positioning device with the longitude and latitude information contained in the collected image, and combining the recorded plant information to mark the plant species in the desert forest remote sensing image; using Labelme software to mark the plant contours in the desert forest remote sensing image to construct the sample data set, and divide it into training set, test set, and validation set in a ratio of 6:2:2.

[0051] Here, observe the remote sensing images of drones at different flight altitudes, such as Figure 3 and Figure 4As shown in the figure, the visible light images and multispectral images collected at different flight altitudes are respectively based on the shrub plant "Haloxylon ammodendron" as an example. It can be seen that when the flight altitude is 20 meters, the desert forest plants are clearly visible and easy to distinguish; while at other flight altitudes, the resolution of desert forest plants is low and difficult to see clearly. At the same time, when comparing the visible light image and multispectral image at the same altitude, the plant outline is clearer in the visible light image. Therefore, the visible light image of the desert forest with a flight altitude of 20 meters was selected as the research object.

[0052] The flight path of the drone is as follows Figure 5 As shown, first use pix4d to stitch the images taken by the drone to generate Figure 6 The high-resolution image of the entire plot is shown in Figure 1. Then ArcGIS is used to match the longitude and latitude information of the plants collected by RTK with the longitude and latitude information contained in the image, as shown in Figure 1. Figure 7 As shown in the figure, combined with the recorded plant information, the plant species in the image are identified. Finally, Labelme software is used to mark the plant outlines in the image. The marking results are shown in the figure. Figure 8 As shown, construct a sample data set.

[0053] In some embodiments, the plant species include at least the following desert shrubs: Haloxylon ammodendron, Tamarix chinensis, Ephedra sinica, Aralia ovata, and Calligonum mongolica.

[0054] Here, desert shrub species such as Haloxylon ammodendron, Tamarix chinensis, Ephedra sinica, Aquilegia ovata, and Calligonum mongolica are of great significance to ecological protection. Therefore, it is particularly important to master the distribution characteristics of these types of shrubs.

[0055] In some embodiments, the method further includes: during the labeling process, plants with a crown width smaller than a fixed threshold are not labeled and are treated as background.

[0056] Here, during the labeling process, there are plants with too small crown width (such as less than 0.1m), and their resolution is low, resulting in the inability to see the texture details clearly. Therefore, such plants are not labeled and are treated as background.

[0057] In some embodiments, the above-mentioned step S150 "merging the group of sub-images into a fused image according to the numbers corresponding to each of the group of sub-images and the position information of the detection frames" is further implemented through the following process: according to the position information of the detection frames, all detection frames in the group of sub-images are positioned on the UAV remote sensing image, and the intersection-and-union ratio between two overlapping detection frames is determined; when the two detection frames are of the same type and the intersection-and-union ratio is greater than 0.5, the two detection frames are fused to obtain a new detection frame, and the size of the new detection frame covers the two detection frames; the group of sub-images are merged according to the numbers corresponding to each of the group of sub-images.

[0058] Here, the sub-image and its detection frame are located back to the original UAV remote sensing image according to the sub-image number and the position of the detection frame detected by the sub-image. For the overlapping area, the intersection over union (IoU) is used to determine whether it is the same plant. The detection frames of the same plant are fused to obtain a new detection frame, and the type of the new detection frame is the type of the first two detection frames.

[0059] In some embodiments, the above step S160 "calculating the total plant area and the sample plot area based on the fused image, and determining the plant coverage of the sample plot" is further implemented through the following process: taking the length and width of the detection frame as the major axis and minor axis of the corresponding plant, respectively, and calculating the projection area of ​​the plant according to the ellipse area calculation formula; dividing the cumulative sum of the projection areas of all plants by the sample plot area to obtain the plant coverage of the sample plot.

[0060] Here, since the detection box of the target detection is a rectangle and the shape of the plant is irregular, if the rectangular area is directly used as the plant area, the coverage will be too large. In actual surveys, it is found that the vertical projection of plants is generally circular or elliptical, so the length and width of the rectangular detection box are used as the major and minor axes of the plant, and the calculation is performed according to the ellipse area calculation formula. Assuming the length and width of the detection box are h and w respectively, the plant projection area is

[0061] In some embodiments, the method further includes: using average accuracy as an evaluation indicator for each shrub species and setting the intersection-over-union ratio to at least 40% to be considered a successful detection; taking a weighted average of the average accuracy of all species to obtain the overall accuracy of the model, wherein the weight coefficient is the proportion of the number of corresponding tree species in the sample data set; and using the overall accuracy to evaluate the performance of the target detection model.

[0062] Here, the average precision AP (Average Precision) is used as the evaluation index for each shrub species. Since the detection accuracy requirement for the position is not high, the IoU is set to 40%, that is, the detection box and the annotation box overlap by more than 40%, which is considered a successful detection, named AP40. For the overall model evaluation index, due to the uneven distribution of shrubs, the number of dominant tree species accounts for more than 90% of the data set, and the impact of different numbers on the model results needs to be considered. Therefore, the AP40 of each type is weighted averaged, and the weight coefficient is the proportion of the target number in the data set. The calculation formula is:

[0063]

[0064] Where N single class is the number of shrubs of a single category, N all class Number of shrubs of all types, AP40 single classis the average precision of the single category, mAP40 weight is the overall accuracy of the model.

[0065] It should be noted that the number distribution of various types of desert forest shrubs is uneven. Among them, Haloxylon ammodendron, Ephedra sinica and Tamarix chinensis are the largest in number, accounting for more than 75% of the total number, and are the dominant tree species in desert shrub forests; while plants such as Lycium barbarum and Polygonum aviculare have a total number of no more than 100 plants, and are inferior tree species. The impact caused by the proportion of tree species should be considered in the detection.

[0066] The embodiment of the present invention enhances the generalization ability of the model in different environments by performing weighted average evaluation on plants of different types and distribution in different quantities.

[0067] The above-mentioned desert shrub recognition method based on the YOLOv8 model is described below in conjunction with a specific embodiment. However, it is worth noting that this specific embodiment is only for better illustrating the present invention and does not constitute an improper limitation of the present invention.

[0068] In this specific embodiment, a set of visible light images collected by a drone at a flight altitude of 20m is used to form a data set. The images are cropped to a size of 1024×1024, and images that do not contain target plants are removed, resulting in 361 training sets, 121 test sets, and 121 validation sets.

[0069] Other target detection models are selected for comparison with the pre-trained YOLOv8 model and evaluated on the sample dataset constructed by the present invention. The evaluation results are shown in Table 1.

[0070] Table 1 Horizontal comparison of YOLOv8 model and other models

[0071] Model mAP50 mAP50-95 YOLOv8 0.503 0.341 DINO(ICLR2023) 0.196 0.109 DDQ(CVPR2023) 0.288 0.155 DiffusionDet (ArXiv2023) 0.408 0.236

[0072] As can be seen from Table 2, the accuracy of the YOLOv8 model is higher than that of other target detection models. Therefore, the YOLOv8 model is more suitable for the desert forest dataset.

[0073] The accuracy of the target detection model obtained by training the YOLOv8 model using the constructed sample data set is shown in Table 2. In the current sample data set, Ephedra, Haloxylon ammodendron, Calligonum elata, Aquilegia serrata, and Tamarix chinensis are relatively large in number and are dominant tree species, accounting for more than 80% of the total number. Other plants are relatively small in number and it is difficult to obtain high accuracy if they are classified separately, so they are merged into one category. The overall accuracy of the current model reaches 80.1%, which meets the expected accuracy.

[0074] Table 2 Model accuracy

[0075] Plant species Number of plants (plants) Accuracy Ephedra 2557 0.903 Pipa Chai 439 0.774 Tamarix 148 0.378 Haloxylon ammodendron 1708 0.7 Calligonum edulis 204 0.683 other 378 0.654 total 5434 0.801

[0076] Based on the existing technology, this invention proposes a new desert shrub identification method by improving the YOLOv8 model and combining it with UAV remote sensing imaging technology, aiming to improve recognition accuracy, reduce computing costs, and provide technical support for smart agriculture and environmental management. First, by integrating the process of large image splitting and model detection, the efficiency of processing large-scale remote sensing images is improved and the consumption of computing resources is reduced. Then, the high accuracy of the YOLOv8 model and the improved image processing technology are used to improve the accuracy of plant species identification and crown width measurement. Automated plant species identification and coverage calculation reduce manual intervention and improve the efficiency and accuracy of monitoring work. Finally, through accurate plant coverage and species distribution information, agricultural managers can develop more effective planting and management strategies to improve agricultural production efficiency and crop yields.

[0077] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0078] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0079] In the several embodiments provided by the present invention, it should be understood that the disclosed methods can be implemented in other ways. The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments. The features disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0080] The above is only an embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A desert shrub identification method based on the YOLOv8 model, characterized in that: include: For the sample plots selected in the desert Gobi Desert, drone remote sensing images of ground shrubs were collected; Splitting the UAV remote sensing image into multiple sub-images according to a preset overlap ratio and numbering them in row and column order; Based on the YOLOv8 model, a target detection model was obtained by training a sample data set constructed using remote sensing images of desert forests taken by drones and ground measurement data; wherein the ground measurement data includes the latitude and longitude coordinates, height, crown width and ground diameter of the plants; Detecting the group of sub-images by using the target detection model, obtaining detection results and recording plant species and canopy sizes; According to the numbers corresponding to the group of sub-images and the position information of the detection frame, the group of sub-images are merged into a fused image; Based on the plant species and the canopy size, the total plant area and the sample plot area are calculated on the fused image, and the plant coverage of the sample plot is determined.

2. The method according to claim 1, characterized in that The sample dataset is constructed through the following process: Obtain a set of visible light images of desert forest plants in the target sample plot when the drone is flying at an altitude of 20 meters; Using pix4d to stitch the set of visible light images according to the flight route of the drone and then crop them to a uniform size, removing images that do not contain target plants, and obtaining a remote sensing image of the desert forest of the entire sample site; Use ArcGIS to match the longitude and latitude information of plants collected by the RTK positioning device with the longitude and latitude information contained in the collected image, and combine the recorded plant information to mark the plant species in the desert forest remote sensing image; Labelme software was used to annotate the plant contours in the desert forest remote sensing image to construct the sample data set, which was divided into a training set, a test set, and a validation set in a ratio of 6:2:

2.

3. The method according to claim 2, characterized in that The method further comprises: During the labeling process, plants with a crown width smaller than a fixed threshold are not labeled and are treated as background.

4. The method according to claim 1, characterized in that: The step of merging the group of sub-images into a fused image according to the serial numbers corresponding to the group of sub-images and the position information of the detection frame includes: According to the position information of the detection frames, all the detection frames in the group of sub-images are positioned on the UAV remote sensing image, and an intersection-over-union ratio between two overlapping detection frames is determined; When the two detection frames are of the same type and the intersection-over-union ratio is greater than 0.5, the two detection frames are merged to obtain a new detection frame, the size of which covers the two detection frames; The group of sub-graphs are merged according to the numbers corresponding to the sub-graphs.

5. The method according to claim 1, characterized in that The calculating the total plant area and the sample plot area respectively based on the fused image and determining the plant coverage of the sample plot comprises: The length and width of the detection frame are respectively used as the major axis and minor axis of the corresponding plant, and the projection area of ​​the plant is calculated according to the ellipse area calculation formula; The cumulative sum of the projected areas of all plants is divided by the area of ​​the sample plot to obtain the plant coverage of the sample plot.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: The average precision was used as the evaluation metric for each shrub species and the intersection-over-union ratio was set to at least 40% to be considered a successful detection; The overall accuracy of the model is obtained by taking a weighted average of the average accuracy of all species, where the weight coefficient is the proportion of the number of the corresponding tree species in the sample data set; The overall accuracy is used to evaluate the performance of the object detection model.

7. The method according to any one of claims 1 to 5, characterized in that: The plant species include at least the following desert shrubs: Haloxylon ammodendron, Tamarix chinensis, Ephedra sinica, Aralia ovata, and Calligonum mongolica.

Images