Unmanned aerial vehicle wetland plant accurate identification system based on multi-dimensional monitoring fusion technology

Abstract

The invention relates to an unmanned aerial vehicle wetland plant accurate identification system based on a multi-dimensional monitoring fusion technology, and relates to the field of computer vision and image processing, and the method comprises the steps: extracting contour information and color features in a region image; performing point cloud fitting on the regional point cloud set by using a plant contour to construct a plant three-dimensional model; performing color rendering through plant color distribution to generate a standard plant three-dimensional model; and performing multi-dimensional feature similarity comparison on the standard plant three-dimensional model and a plurality of predetermined three-dimensional models, and setting the plant type corresponding to the maximum similarity as an identification result. The method can solve the technical problems that a traditional method often depends on manual labeling or simple feature extraction, is easily affected by the environment, and is low in plant recognition efficiency and insufficient in recognition accuracy. Through the unmanned aerial vehicle platform, in combination with multi-source data of optical imaging and laser radar, accurate identification of plant species can be realized, and the accuracy and efficiency of wetland plant identification are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and image processing, and in particular to an unmanned aerial vehicle wetland plant precision identification system based on multi-dimensional monitoring fusion technology. Background Art

[0002] Traditional plant recognition methods usually rely on image processing technology, which first requires manual annotation of a large amount of sample data, and then uses simple image features (such as color, shape, texture, etc.) for recognition. However, this feature extraction-based algorithm is limited by environmental changes, such as lighting, weather, and background clutter, which can easily lead to recognition errors; in addition, the process of relying on manual annotation is time-consuming and labor-intensive, and it is difficult to meet the needs of large-scale, real-time plant monitoring. Summary of the invention

[0003] In view of the technical problems that traditional plant identification methods often rely on manual labeling or simple feature extraction, are easily affected by the environment, and have low identification efficiency and insufficient identification accuracy, the present invention provides a UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology to solve the problem.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: The present invention provides an unmanned aerial vehicle wetland plant precision identification system based on multi-dimensional monitoring fusion technology, including: a data acquisition module, used to collect data on a wetland area through an optical camera and a laser radar carried by a unmanned aerial vehicle, and obtain a regional image and a regional point cloud set; a feature extraction module, used to extract contour information and color features in the regional image, and obtain plant contours and plant color distribution; a point cloud fitting module, used to perform point cloud fitting on the regional point cloud set using the plant contour, and construct a plant three-dimensional model according to the optimal point cloud fitting result; a color rendering module, used to perform color rendering on the plant three-dimensional model according to the plant color distribution, and generate a standard plant three-dimensional model; a similarity comparison module, used to perform multi-dimensional feature similarity comparison on the standard plant three-dimensional model with multiple predetermined three-dimensional models in a plant database, and set the plant type corresponding to the maximum comprehensive similarity as the identification result.

[0005] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: construct a plant contour recognition plug-in based on a convolutional neural network; input the regional image into the plant contour recognition plug-in to extract contour information, and output the plant contour.

[0006] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: collect a set of sample plant images, and mark the plant contours in different sample plant images to obtain sample plant contours, and construct a set of sample plant contours; using sample plant images as input and sample plant contours as supervision, the sample plant image set and the sample plant contour set are used to perform supervised learning on the convolutional neural network until convergence to obtain the plant contour recognition plug-in.

[0007] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: use the plant contour to crop the area image to obtain a plant image; extract color features from the plant image to generate a plant color distribution.

[0008] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: perform random point cloud fitting on the regional point cloud set according to the plant outline to obtain a first point cloud fitting result, calculate the ratio of the number of point clouds in the first point cloud fitting result to the total number of point clouds in the regional point cloud set, and set it as a first fitting accuracy; perform random point cloud fitting on the regional point cloud set again according to the plant outline to obtain a second point cloud fitting result, and calculate a second fitting accuracy; continue to perform iterative fitting until a predetermined number of fitting times is reached, and output the point cloud fitting result corresponding to the maximum fitting accuracy as the optimal point cloud fitting result.

[0009] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: randomly select a first predetermined three-dimensional model from the multiple predetermined three-dimensional models; based on a multi-dimensional feature comparison index, perform a multi-dimensional feature similarity comparison between the standard plant three-dimensional model and the first predetermined three-dimensional model to obtain multiple first comparison similarities; perform weighted fusion on the multiple first comparison similarities to obtain a first comprehensive similarity; analyze in turn to obtain multiple comprehensive similarities between the standard plant three-dimensional model and the multiple predetermined three-dimensional models, and select the plant type corresponding to the maximum comprehensive similarity as the identification result.

[0010] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used for: the multi-dimensional feature comparison indicators at least include plant size characteristics, plant structure characteristics and plant color distribution characteristics.

[0011] Preferably, the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology is also used to: use the coefficient of variation method to respectively weight the plant size characteristics, plant structure characteristics and plant color distribution characteristics to obtain size weight, structure weight and color weight, wherein the sum of the size weight, structure weight and color weight is 1; based on the size weight, structure weight and color weight, perform weighted calculation on the multiple first comparison similarities to obtain the first comprehensive similarity.

[0012] The beneficial effects of the present invention are as follows: data is collected from the wetland area through the optical camera and laser radar carried by the drone to obtain the regional image and the regional point cloud set; then the contour information and color features in the regional image are extracted to obtain the plant contour and plant color distribution; then the plant contour is used to perform point cloud fitting on the regional point cloud set, and a plant three-dimensional model is constructed according to the optimal point cloud fitting result; the plant three-dimensional model is further rendered by the plant color distribution to generate a standard plant three-dimensional model; finally, the standard plant three-dimensional model is compared with multiple predetermined three-dimensional models in the plant database for multi-dimensional feature similarity, and the plant type corresponding to the maximum comprehensive similarity is set as the identification result. In other words, through the drone platform, combined with the multi-source data of optical imaging and laser radar, accurate identification of plant species can be achieved, significantly improving the accuracy and efficiency of wetland plant identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the structure of the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology provided by the present invention; Figure 2 A schematic diagram of the process of point cloud fitting in the UAV wetland plant precision identification system based on multi-dimensional monitoring fusion technology provided by the present invention.

[0014] In the accompanying drawings, the components represented by the reference numerals are described as follows: Data acquisition module 11, feature extraction module 12, point cloud fitting module 13, color rendering module 14, similarity comparison module 15. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0016] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0017] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0018] Examples, such as Figure 1 As shown, the embodiment of the present invention provides a UAV wetland plant accurate identification system based on multi-dimensional monitoring fusion technology, including: The data acquisition module 11 is used to collect data on the wetland area through the optical camera and laser radar carried by the drone to obtain regional images and regional point cloud sets.

[0019] Specifically, as one of the modern remote sensing platforms, drones have the advantages of flexibility, high efficiency and relatively low cost. Drones can cover large areas and collect data at lower altitudes, which is of great significance for the detailed monitoring of wetland areas; compared with traditional satellite remote sensing, drones have greater advantages in spatial resolution and flexibility.

[0020] First, data on the wetland area is collected through the optical camera and lidar carried by the drone. The optical camera can clearly obtain two-dimensional images of the wetland area by capturing information in the visible light and near-infrared light bands. These images provide surface features such as the appearance, color, and texture of wetland plants, helping to identify the type and health of the plants. The lidar emits laser pulses to the ground and receives the returned laser reflection signals, thereby measuring the precise three-dimensional position of the ground and objects, and recording in detail the spatial distribution of plants, terrain and other objects in the wetland area. Regional images and regional point clouds are obtained.

[0021] Data collection through the optical camera and lidar carried by the drone can effectively obtain images and point cloud data of the wetland area, providing rich multi-dimensional information. Among them, the optical camera can provide the appearance characteristics of the plants, while the lidar can accurately record the spatial structure of the plants. Combined with these data, subsequent plant identification and three-dimensional modeling will be more accurate and efficient.

[0022] The feature extraction module 12 is used to extract the contour information and color features in the regional image, and obtain the plant contour and plant color distribution.

[0023] Furthermore, the feature extraction module 12 is also used for: Build a plant contour recognition plug-in based on convolutional neural network.

[0024] Furthermore, the present invention further comprises the following steps: A sample plant image set is collected, and plant contours in different sample plant images are annotated to obtain sample plant contours and construct a sample plant contour set; with sample plant images as input and sample plant contours as supervision, the sample plant image set and the sample plant contour set are used to perform supervised learning on a convolutional neural network until convergence to obtain the plant contour recognition plug-in.

[0025] Specifically, collecting a sample plant image set is a key step in the plant recognition system. This dataset should contain a variety of wetland plant images, preferably covering different plant species, different growth stages, and different environmental conditions. Each image should clearly show the plant's contour information and ensure good image quality to facilitate subsequent feature extraction and obtain a sample plant image set. Then, the plant contours in different sample plant images are annotated, such as using an open source annotation tool (such as LabelMe) to annotate each image. When annotating, it is necessary to circle the plant's contours, accurately outline the plant's edges, obtain the sample plant contours, and annotate them in sequence to construct a sample plant contour set.

[0026] In plant contour recognition, convolutional neural networks (CNNs) are widely used for feature extraction and classification of images. By inputting annotated plant images and corresponding contour annotations, supervised learning training is performed to optimize model parameters. Among them, CNN usually includes an input layer, multiple convolutional layers, pooling layers, fully connected layers, and output layers. Here, the convolutional layer and the pooling layer are responsible for extracting local features in the image, and the output layer will output the predicted contour. Then supervised learning is performed using a set of sample plant images and a set of sample plant contours. The loss function usually uses pixel-level cross entropy or other loss functions suitable for contour recognition, such as the Dice coefficient, to measure the similarity between the predicted contour and the actual annotated contour. The weights of the CNN model are adjusted by the back propagation algorithm until the training error converges. After the supervised learning is completed and converged, a CNN model that can accurately extract plant contours from the input image is obtained as a plant contour recognition plug-in.

[0027] The area image is input into the plant outline recognition plug-in to extract outline information, and the plant outline is output.

[0028] Specifically, the regional image is finally input into the plant contour recognition plug-in to extract contour information, and the plant contour is output. By constructing a plant contour recognition plug-in based on a convolutional neural network, contour information in plant images can be automatically extracted, and accurate input can be provided for further plant identification and classification, thereby improving the recognition accuracy and efficiency of wetland plants.

[0029] Furthermore, the feature extraction module 12 is also used for: The regional image is cropped using the plant outline to obtain a plant image; and color features are extracted from the plant image to generate a plant color distribution.

[0030] Specifically, first, the image of the region is cropped using the plant contour, such as using a contour-based image mask method to separate the plant region from the background to obtain a plant image. Through these methods, the cropped plant image will retain the core features of the plant and remove the background part, which is helpful for subsequent color feature extraction and plant identification. Then, the color feature of the plant image is extracted, such as using a clustering algorithm (such as K-means clustering) to cluster the colors in the image and extract the main color features in the image. The extracted color features are then further analyzed to obtain a probability model or feature vector of the plant color distribution, and the plant color distribution is constructed. The color distribution describes the color composition and proportion of different areas on the plant surface.

[0031] By using the plant contour to crop the image and extracting color features from the cropped plant image, effective support can be provided for the accurate identification of plants. The color distribution characteristics can not only improve the recognition accuracy of plant species, but also assist in judging the health status of plants. It is an indispensable part of wetland plant monitoring.

[0032] The point cloud fitting module 13 is used to perform point cloud fitting on the regional point cloud set using the plant contour, and to construct a three-dimensional plant model according to the optimal point cloud fitting result.

[0033] Further, if Figure 2 As shown, the point cloud fitting module 13 is also used for: The point cloud of the regional point cloud set is randomly fitted according to the plant outline to obtain a first point cloud fitting result, and the ratio of the number of point clouds in the first point cloud fitting result to the total number of point clouds in the regional point cloud set is calculated and set as a first fitting accuracy; the point cloud of the regional point cloud set is randomly fitted again according to the plant outline to obtain a second point cloud fitting result, and a second fitting accuracy is calculated; the iterative fitting is continued until a predetermined number of fitting times is reached, and the point cloud fitting result corresponding to the maximum fitting accuracy is output and set as the optimal point cloud fitting result.

[0034] Specifically, points are randomly selected in the regional point cloud set according to the plant outline to perform the first fitting. The purpose of fitting is to make the fitted three-dimensional model as close as possible to the actual form of the plant by selecting a reasonable point set. The reason for randomly selecting a point set for fitting is that the point cloud data itself may contain noise or irregular data. Therefore, randomly selecting different point sets through multiple iterations helps to filter out noise points, extract the effective point cloud of the plant, and obtain the first point cloud fitting result. Next, the ratio of the number of point clouds in the first point cloud fitting result to the total number of point clouds in the regional point cloud set is calculated, and is set as the first fitting accuracy. The calculation of the fitting accuracy is an important criterion for judging whether the fitting result meets the requirements. The fitting accuracy can be measured by comparing the ratio between the number of point clouds in the fitting result and the total number of point clouds.

[0035] Next, the point cloud of the regional point cloud set is randomly fitted again according to the plant outline to obtain the second point cloud fitting result. The accuracy of the second fitting is calculated through different random point sets. The fitting is continued in the same way for multiple times, and the fitting accuracy is calculated for each fitting, and the fitting process is continuously optimized. Usually, the number of fittings is predetermined, such as 10 or more iterations. After multiple point cloud fittings, a series of different fitting results and corresponding fitting accuracies are obtained, and then the one with the highest fitting accuracy is selected from all iterative results as the final fitting result, and the point cloud fitting result with the highest fitting accuracy is finally output, that is, the optimal point cloud fitting result. This result will be used to construct a three-dimensional model of the plant, and further perform color rendering and similarity comparison with the plant model in the database.

[0036] Finally, the plant 3D model is constructed based on the optimal point cloud fitting result. For example, by performing Delaunay triangulation on the point cloud data, a mesh composed of triangles can be generated. The three vertices of each triangle are the three points in the fitted point cloud. Then, the generated mesh is smoothed to eliminate the surface irregularities caused by the point cloud sampling error. Commonly used smoothing methods include Gaussian smoothing and Laplace smoothing.

[0037] By performing random point cloud fitting on the regional point cloud set and optimizing the fitting accuracy through an iterative process, the three-dimensional structure of the plant can be effectively extracted from the complex point cloud data; this process can significantly improve the accuracy of point cloud fitting and provide support for the three-dimensional modeling and precise identification of plants; at the same time, through multiple iterations and precision calculations, it can ensure that the final output point cloud fitting result is more accurate, thereby improving the accuracy and reliability of plant identification.

[0038] The color rendering module 14 is used to perform color rendering on the plant three-dimensional model according to the plant color distribution to generate a standard plant three-dimensional model.

[0039] Specifically, the plant three-dimensional model is rendered in color by the plant color distribution. The surface of the three-dimensional model usually has different material properties (such as smooth, rough, transparent, etc.). The surface under each material property reflects and absorbs light differently, which affects the color performance of the model. Therefore, it is necessary to first set the material parameters of the plant three-dimensional model surface to ensure that the model can correctly render different color features; each vertex of the three-dimensional model has a texture coordinate, which is responsible for indicating how to map image data (such as color) to the surface of the three-dimensional model. By associating the extracted plant color distribution with the texture coordinate, the color information can be accurately applied to each vertex of the three-dimensional model. Then, the plant three-dimensional model is rendered in color by a modern rendering algorithm using the extracted plant color distribution, and the plant color distribution is applied to the texture coordinates of the plant three-dimensional model, which means that the area on the surface of each three-dimensional model will be filled with the color extracted from the plant image. Accurate color rendering is achieved by mapping the color information of different areas (such as leaves, flowers, stems, etc.) to the corresponding three-dimensional model parts respectively. For different plant areas, there may be some color transition parts. At this time, color interpolation technology (such as bilinear interpolation or trilinear interpolation) can be used to smooth the color of the transition area so that the model looks more natural. After color rendering, the final output 3D plant model will include the color distribution, shape and structure of the plant. At this point, the standard plant 3D model already has accurate color information and presents high-quality visual effects.

[0040] By extracting the color distribution of plants and mapping it to the surface of the three-dimensional model, the plant three-dimensional model can be effectively given a realistic visual effect. The process involves steps such as color feature extraction, texture mapping, lighting rendering, and surface detail adjustment, and ultimately generates a standard plant three-dimensional model with accurate color representation and realism.

[0041] The similarity comparison module 15 is used to perform multi-dimensional feature similarity comparison between the standard plant three-dimensional model and a plurality of predetermined three-dimensional models in the plant database, and set the plant type corresponding to the maximum comprehensive similarity as the recognition result.

[0042] Furthermore, the similarity comparison module 15 is also used for: A first predetermined three-dimensional model is randomly selected from the plurality of predetermined three-dimensional models.

[0043] Based on the multi-dimensional feature comparison index, the standard plant three-dimensional model and the first predetermined three-dimensional model are compared for multi-dimensional feature similarity to obtain a plurality of first comparison similarities.

[0044] Furthermore, the present invention further comprises the following steps: The multi-dimensional feature comparison index includes at least plant size features, plant structure features and plant color distribution features.

[0045] Specifically, first, any one predetermined three-dimensional model is randomly selected from the multiple predetermined three-dimensional models and set as the first predetermined three-dimensional model; then a multidimensional feature comparison index is obtained, wherein the multidimensional feature comparison index at least includes plant size features, plant structure features and plant color distribution features, and the size features are extracted by measuring various dimensions (such as height, width, depth, etc.) of the plant three-dimensional model, and the size similarity of the standard plant and the predetermined model is compared to determine whether the sizes of the two are close; the plant structure features include the overall shape of the plant, leaf distribution, branch structure, etc., and the topological structure information of the model (such as the number of leaves, arrangement, etc.) can be extracted through geometric morphological analysis and compared; the plant color distribution features are obtained by extracting the color distribution of each three-dimensional model, including the colors of different parts such as leaves, stems and flowers; the color distribution features of each model are obtained by using feature extraction technology in a color histogram or a color space (such as RGB, HSV), and these features are used for subsequent similarity evaluation to ensure that information in multiple dimensions is comprehensively considered to improve the accuracy of the comparison results.

[0046] Then, based on the multi-dimensional feature comparison index, the standard plant three-dimensional model and the first predetermined three-dimensional model are compared for multi-dimensional feature similarity, and different similarity calculation methods are used to evaluate the similarity of the two three-dimensional models in each feature, such as measuring the difference between the two in size, structure and color space by Euclidean distance; using cosine similarity to evaluate the angle difference between the feature vectors of the two, which is particularly suitable for comparing color distribution; based on the above method, the similarity in multiple feature dimensions is calculated (for example, plant size similarity, structural similarity, color similarity), and the comparison similarity of each feature is obtained, and multiple first comparison similarities are obtained. Each similarity value reflects the degree of match between the standard plant and the first predetermined three-dimensional model in a specific feature, and these comparison similarities can be used for subsequent comprehensive evaluation to ensure that the predetermined three-dimensional model that best matches the standard plant three-dimensional model is selected as the final recognition result.

[0047] Through the above-mentioned multidimensional feature comparison method, the similarity between the standard plant three-dimensional model and multiple predetermined three-dimensional models can be effectively compared to ensure the accuracy of the final recognition result. This method can not only process multi-dimensional information such as size, structure and color, but also optimize the recognition process and improve the efficiency and accuracy of accurate identification of wetland plants.

[0048] The multiple first comparison similarities are weightedly merged to obtain a first comprehensive similarity.

[0049] Furthermore, the present invention further comprises the following steps: Using the coefficient of variation method, the plant size characteristics, plant structure characteristics and plant color distribution characteristics are weighted respectively to obtain size weight, structure weight and color weight, wherein the sum of the size weight, structure weight and color weight is 1; based on the size weight, structure weight and color weight, the multiple first comparison similarities are weightedly calculated to obtain the first comprehensive similarity.

[0050] Specifically, the coefficient of variation method is used to configure the weights of the plant size characteristics, plant structure characteristics and plant color distribution characteristics. First, the mean and standard deviation of the plant size characteristics (such as height, width, depth) are calculated, and the coefficient of variation of the size characteristics is obtained; the mean and standard deviation of the plant structure characteristics (such as the number of leaves, leaf shape, etc.) are calculated to obtain the coefficient of variation of the structure characteristics; the mean and standard deviation of the plant color distribution characteristics (such as the distribution of RGB or HSV values) are calculated to obtain the coefficient of variation of the color characteristics. Then, the coefficient of variation method is used to configure the weight of each feature. The larger the coefficient of variation, the greater the volatility of the feature and the more significant the impact. Therefore, the weight of the feature should be relatively high, and the size weight, structure weight and color weight are obtained, where the sum of the size weight, structure weight and color weight is 1.

[0051] Then, based on the size weight, structure weight and color weight, the multiple first comparison similarities are weighted and calculated. Through this weighted calculation, the similarities of plant size, structure and color characteristics are comprehensively considered to obtain a unified first comprehensive similarity, which represents the comprehensive similarity between the standard plant and a predetermined plant model. By using the coefficient of variation method to calculate the weight of each feature and performing weighted similarity calculation, the multi-dimensional information of size, structure and color characteristics can be effectively integrated to ensure the accuracy and stability of the wetland plant identification process; this method significantly improves the adaptability of the identification system in complex environments, especially in wetland areas with diverse plant species and large environmental changes.

[0052] A plurality of comprehensive similarities between the standard plant three-dimensional model and the plurality of predetermined three-dimensional models are analyzed in sequence, and the plant type corresponding to the maximum comprehensive similarity is selected as the recognition result.

[0053] Specifically, the same method for calculating the first comprehensive similarity is used to sequentially analyze and obtain multiple comprehensive similarities between the standard plant three-dimensional model and the multiple predetermined three-dimensional models, and the plant type corresponding to the maximum comprehensive similarity is selected as the recognition result. By calculating the comprehensive similarities between the standard plant three-dimensional model and the multiple predetermined three-dimensional models one by one, and selecting the plant type corresponding to the maximum value, high-precision plant species recognition can be achieved. This method combines multi-dimensional features such as size, structure, and color, and can effectively respond to the changes and environmental impacts of different plants in wetland areas, thereby ensuring the accuracy and stability of plant recognition.

[0054] The UAV wetland plant accurate identification system based on multi-dimensional monitoring fusion technology provided by the embodiment of the present invention has at least the following technical effects: The optical camera and laser radar carried by the drone are used to collect data from the wetland area to obtain regional images and regional point cloud sets; then the contour information and color features in the regional images are extracted to obtain the plant contours and plant color distribution; then the plant contours are used to perform point cloud fitting on the regional point cloud set, and a plant three-dimensional model is constructed based on the optimal point cloud fitting result; the plant three-dimensional model is further rendered using the plant color distribution to generate a standard plant three-dimensional model; finally, the standard plant three-dimensional model is compared with multiple predetermined three-dimensional models in the plant database for multi-dimensional feature similarity, and the plant type corresponding to the maximum comprehensive similarity is set as the identification result. In other words, through the drone platform, combined with multi-source data from optical imaging and laser radar, accurate identification of plant species can be achieved, significantly improving the accuracy and efficiency of wetland plant identification.

[0055] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art understand the basic inventive concepts.

[0056] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. The UAV wetland plant accurate identification system based on multi-dimensional monitoring fusion technology is characterized by: The UAV wetland plant precise identification system comprises: The data acquisition module is used to collect data of the wetland area through the optical camera and laser radar carried by the drone to obtain regional images and regional point cloud sets; A feature extraction module, used to extract contour information and color features in the regional image, and obtain plant contours and plant color distribution; A point cloud fitting module, used to perform point cloud fitting on the regional point cloud set using the plant outline, and to construct a three-dimensional plant model according to the optimal point cloud fitting result; A color rendering module, used for performing color rendering on the plant three-dimensional model according to the plant color distribution to generate a standard plant three-dimensional model; The similarity comparison module is used to perform multi-dimensional feature similarity comparison between the standard plant three-dimensional model and multiple predetermined three-dimensional models in the plant database, and set the plant type corresponding to the maximum comprehensive similarity as the recognition result.

2. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 1 is characterized in that: Extracting contour information from the region image includes: Build a plant contour recognition plug-in based on convolutional neural network; The area image is input into the plant outline recognition plug-in to extract outline information, and the plant outline is output.

3. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 2 is characterized in that: A plant contour recognition plug-in is built based on a convolutional neural network, including: Collecting a set of sample plant images, and marking plant contours in different sample plant images to obtain sample plant contours, and constructing a set of sample plant contours; The sample plant images are used as input and the sample plant contours are used as supervision. The sample plant image set and the sample plant contour set are used to perform supervised learning on the convolutional neural network until convergence, thereby obtaining the plant contour recognition plug-in.

4. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 2 is characterized in that: Extracting color features in the regional image includes: Using the plant outline to perform image cropping on the regional image to obtain a plant image; Color features are extracted from the plant image to generate plant color distribution.

5. The UAV wetland plant accurate identification system based on multi-dimensional monitoring fusion technology according to claim 1 is characterized in that: Performing point cloud fitting on the regional point cloud set using the plant outline includes: Performing point cloud random fitting on the regional point cloud set according to the plant outline to obtain a first point cloud fitting result, and calculating a ratio of the number of point clouds in the first point cloud fitting result to the total number of point clouds in the regional point cloud set, setting the ratio as a first fitting accuracy; Performing random point cloud fitting on the regional point cloud set again according to the plant outline to obtain a second point cloud fitting result, and calculating a second fitting accuracy; Continue iterating the fitting until the predetermined number of fitting times is reached, and output the point cloud fitting result corresponding to the maximum fitting accuracy as the optimal point cloud fitting result.

6. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 1 is characterized in that: Performing multi-dimensional feature similarity comparison between the standard plant three-dimensional model and a plurality of predetermined three-dimensional models in the plant database, and setting the plant type corresponding to the maximum comprehensive similarity as the recognition result, including: Randomly selecting a first predetermined three-dimensional model from the plurality of predetermined three-dimensional models; Based on the multidimensional feature comparison index, the standard plant three-dimensional model and the first predetermined three-dimensional model are compared for multidimensional feature similarity to obtain a plurality of first comparison similarities; Performing weighted fusion on the multiple first comparison similarities to obtain a first comprehensive similarity; A plurality of comprehensive similarities between the standard plant three-dimensional model and the plurality of predetermined three-dimensional models are analyzed in sequence, and the plant type corresponding to the maximum comprehensive similarity is selected as the recognition result.

7. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 6 is characterized in that: The multi-dimensional feature comparison index includes at least plant size features, plant structure features and plant color distribution features.

8. The UAV wetland plant precise identification system based on multi-dimensional monitoring fusion technology according to claim 7 is characterized in that: The weighted fusion of the multiple first comparison similarities to obtain a first comprehensive similarity includes: Using the coefficient of variation method, weights are assigned to the plant size characteristics, plant structure characteristics, and plant color distribution characteristics to obtain size weights, structure weights, and color weights, wherein the sum of the size weights, structure weights, and color weights is 1; Based on the size weight, structure weight and color weight, weighted calculation is performed on the multiple first comparison similarities to obtain the first comprehensive similarity.

Images