A highway greening engineering maintenance method and system based on machine learning

Abstract

The application belongs to the technical field of highway maintenance, and provides a highway greening engineering maintenance method and system based on machine learning, comprising the following steps: S100, acquiring a to-be-maintained greening engineering area map and generating a collection path, and flying a UAV along the collection path to acquire multiple hyperspectral images of the area map; S200, generating a first image according to the acquired multiple hyperspectral images; S300, marking a pest area in the first image and updating the first image into a second image; S400, introducing a decision tree model to train and learn pest image features and environmental parameters by training the second image to the initialized decision tree model obtained by training the second image to the decision tree model, so that key influencing factors can be autonomously extracted, intelligent identification of pest areas and judgment of diffusion trends are realized, and targeted maintenance suggestions are generated according to the training results, the accuracy and reliability of pest detection are improved, and the method is particularly suitable for highway scenes with wide distribution of green belts and rapid environmental changes.

Description

Technical Field

[0001] This invention belongs to the field of highway maintenance technology, specifically relating to a method and system for highway greening engineering maintenance based on machine learning. Background Technology

[0002] Highway green belts face numerous challenges in maintenance due to their wide geographical distribution, diverse plant species, and significant climate differences. In particular, the frequent occurrence and difficulty in controlling pests and diseases pose significant challenges. Traditional maintenance methods rely mainly on manual inspections and experience-based judgments, which suffer from problems such as delayed response, missed inspections, misjudgments, and untimely treatment. This can easily lead to the spread of pests and the death of vegetation, increasing maintenance costs and affecting road safety and the overall landscape effect. Therefore, there is an urgent need for a more efficient and intelligent pest and disease management method to achieve accurate identification and dynamic management of the health status of green plants, thereby improving the sustainable maintenance level of highway greening projects. Summary of the Invention

[0003] The purpose of this invention is to propose a method and system for highway greening engineering maintenance based on machine learning, so as to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0004] To achieve the above objectives, according to one aspect of the present invention, a method for highway greening engineering maintenance based on machine learning is provided, the method comprising the following steps:

[0005] S100: Obtain a map of the greening project area to be maintained and generate a collection path. The drone will fly along the collection path to obtain multiple hyperspectral images of the area map.

[0006] S200, Generate a first image based on the acquired multiple hyperspectral images;

[0007] S300, mark the pest-infested areas in the first image and update it to the second image;

[0008] S400, The initial decision tree model is obtained by training the decision tree model with the second image.

[0009] Furthermore, it also includes step S500, which identifies the maintenance location from the second image by initializing the decision tree model.

[0010] Furthermore, the UAV is a fixed-wing UAV or a rotary-wing UAV including a hyperspectral imaging system, and the initial parameters of the UAV are set as follows: field of view of the hyperspectral imaging system (10-30°), flight path spacing (10-50m), flight altitude (15-60m), flight speed (8-30m / s), boundary safety distance (3-7m), and obstacle safety distance (3-7m).

[0011] Furthermore, in S100, the specific method for acquiring a map of the greening project area to be maintained and generating a collection path, and then having the UAV fly along the collection path to acquire multiple hyperspectral images of the area map, is as follows: acquire a map of the greening project area to be maintained, set the start point, end point, and obstacle location information of the UAV's flight in the greening project area to be maintained, generate the flight path of the UAV from the start point to the end point, generate collection points at fixed distances, and have the UAV fly along the path to acquire spectral images at the collection points using the onboard spectrometer.

[0012] Furthermore, the drone is a coaxial dual-rotor drone, a micro-rotor drone, or a multi-rotor drone equipped with a spectrometer.

[0013] Furthermore, the spectrometer is a GaiaField-mini spectrometer, a Specim AFX series hyperspectral camera, or an ATH9020 hyperspectral imager.

[0014] Furthermore, the flight path of the UAV from the starting point to the destination is generated using any one of the following algorithms: A* algorithm, K-shortest path algorithm, Dijkstra algorithm, APF algorithm (Artificial Potential Field), and SAA algorithm (Simulated Annealing Algorithm).

[0015] Furthermore, the starting point is the location where the drone takes off in the area to be monitored (generally the current location), the ending point is the destination location of the drone's flight in the area to be monitored, the obstacle location information is the location of the obstacles in the greening project area to be maintained, and the drone needs to maintain a preset obstacle safety distance (2-5m) from the obstacle location.

[0016] Furthermore, in S200, the specific method for generating the first image based on the acquired multiple hyperspectral images is as follows: all the acquired multispectral images are stitched together in the order of acquisition time to generate the first image.

[0017] Furthermore, the stitching process includes the following steps: geometric correction, image preprocessing, image registration, and image fusion.

[0018] Current common methods for monitoring pests and diseases mainly rely on manual inspections and fixed-point image acquisition, supplemented by periodic spraying or pruning. However, in practical applications, these methods have many limitations: Firstly, because image acquisition usually relies on fixed cameras or manual collection, limitations in the number of devices, network transmission, and image processing capabilities lead to problems such as data update delays and information feedback lags, making it difficult to achieve real-time monitoring and rapid response to pests. Secondly, the spread of pests in green belts is affected by factors such as traffic disturbances, wind changes, and differences in vegetation structure, exhibiting complex characteristics such as unstable propagation paths and uneven spread speeds, resulting in inaccurate identification of high-incidence areas. Furthermore, current systems often lack the ability to predict and dynamically analyze pest development trends, leading to untimely, mismatched, or insufficient coverage of pesticide application or intervention measures, ultimately affecting maintenance effectiveness and increasing operation and maintenance costs. To solve these problems, this invention proposes the following method: by identifying pest-infested areas and calculating the diffusion coefficient of all pest-infested areas, it achieves accurate prediction of the specific geographical location of pests, improving the intelligent maintenance level of highway greening projects.

[0019] Furthermore, in S300, the specific method for marking the pest-infested areas in the first image and updating it to the second image is as follows: Multiple pest-infested areas PA are marked in the first image using the RX algorithm, with i as the index of the pest-infested area, PA... i Let i represent the i-th pest-infested area, and let INP be a set of all pest-infested areas. Use an edge detection algorithm to divide the first image into multiple sub-regions RAj, with j as the index of the sub-region, and RAj representing the j-th sub-region.

[0020] Let P(i) be the geometric center of all pest-infested areas and R(i) be the geometric center of all sub-regions. Traverse all pest-infested areas in the set INP and calculate PR(j) the distance between the geometric center P(i) of all pest-infested areas and the geometric center R(i) of all projections. Add all projections whose PR(j) is less than PR_mean to the list Listi, where PR_mean is the average distance between point P(i) and all points R(i).

[0021] Furthermore, the pest-affected area is the region composed of pest-affected pixels, or the internal region of the edge composed of pest-affected pixels; wherein, pest-affected pixels are pixels in a spectral image whose visible light reflectance is higher than the average visible light reflectance of each pixel in the spectral image, or, pest-affected pixels are pixels in a spectral image whose near-infrared spectral reflectance is lower than the average near-infrared spectral reflectance of each pixel in the spectral image.

[0022] Within the range of values ​​for k, all sub-regions of Listi are traversed sequentially, and the pest-affected region PA is calculated using a formula. iThe diffusion coefficient Diff in each sub-region of Listi is calculated as follows: The average pixel value in each sub-region is GreenK_Mean. The maximum pixel value in the k-th sub-region is GreenK_Max, and the minimum pixel value is GreenK_Min. The pixel value of the edge point in the sub-region is Green_edg(k,r), where K represents the sub-region index in Listi, r represents the edge point index, and Green_edg(k,r) represents the pixel value of the r-th edge point in the k-th sub-region. This is achieved using the formula: Calculate the diffusion coefficient Diff, where q(k) is the distance between the pixel with the maximum value and the pixel with the minimum value in the k-th sub-region, and PR(k) represents the pest-infested region PA. i The distance between the geometric center of the region and the geometric center of the subregion;

[0023] If Diff is greater than Green_Mean, where Green_Mean is the average value of pixels in all sub-regions contained in List1, then it is considered that the current pest area has spread and needs to be corrected; otherwise, the current pest area has not spread and does not need to be corrected.

[0024] The beneficial effects of the above steps are as follows: By calculating the diffusion coefficient Diff of the pest-affected area in each sub-region of the image, it is possible to accurately determine whether the pest has spread based on the pixel statistical features and edge point deviations within the region, comprehensively considering the gray-level gradient distribution and spatial geometric relationship of the pest-affected area. It no longer relies on a single threshold or visual connectivity to judge the spread, but introduces a diffusion formula that includes the maximum pixel difference, edge offset, and center distance weights. While maintaining computational efficiency, it effectively improves the sensitivity and stability of diffusion recognition. Especially in scenarios such as highway green belts, which are greatly affected by traffic flow disturbances and changes in lighting, this method can significantly improve the accuracy of dynamic pest diffusion recognition, reduce the false judgment rate, and provide a scientific and quantitative decision-making basis for subsequent correction of the diseased area, thereby optimizing greening maintenance strategies and improving the timeliness and accuracy of pest and disease treatment.

[0025] Furthermore, the specific correction method is as follows: within the range of values ​​for k, the diffusion coefficient Diff of the pest-affected area in each sub-region of Listi is calculated sequentially. k Where k = 1, 2, 3...m, m represents the number of sub-regions contained in Listi. All sub-regions contained in Listi are rearranged in ascending order of Diff, and the diffusion coefficients Diff of the sub-regions in Listi are iterated and updated sequentially. kIf satisfied, Diff k -Diff k-1 <Diff k+1 -Diff k And Diff k >Diff mean Diff mean Let Diff represent the average diffusion coefficient (Diff) of the sub-regions in Listi. Then, Diff... k The corresponding sub-region is denoted as the strongly influential sub-region.

[0026] Traverse the infested area PA in sequence i Let PL be the pixel value on the edge. Select the minimum pixel value as the starting point ST and the maximum pixel value as the ending point ED. Starting from ST, search for PA along the minimum path to ED. i The magnitude of the pixel value PL on the edge, with s and q as the index of the pixel value on the edge, if satisfying and At that time, update the pixel corresponding to PLs to the starting point ST; starting from ED, search for PA along the minimum path to ST. i The size of the pixel value PL on the edge, with s as the index of the pixel value on the edge, if satisfying and At that time, update the pixel corresponding to PLs to the starting point ED, where z represents PA. i The number of pixels on the edge;

[0027] Mark the position of the pixel corresponding to ST as point A, and the position of the pixel corresponding to ED as point B. Mark the minimum curve segment between points A and B as the deletion segment and delete it. Then, sequentially strengthen the pixel values ​​on the edges of the affected sub-regions, marking the point corresponding to the maximum pixel value as point C and the point corresponding to the minimum pixel value as point D. Mark the minimum curve segment between points C and D as the replacement segment. Copy the replacement segment and scale it proportionally to the area PA between points A and B to correct the pest infestation. i ;

[0028] Correct all insect-infested areas in the first image using the method described above, and update the first image to the second image.

[0029] The beneficial effects of the above steps are as follows: Combining the quantitative judgment of the diffusion trend, the above steps can effectively make up for the time delay and lag response problems of existing pest monitoring methods. By constructing the path retrieval logic of pest edge pixels, the extreme points of boundary pixels in the pest area are accurately extracted and the statistical difference between the minimum path and the edge pixel sequence is used to dynamically determine the true diffusion boundary of the pest area. In addition, by implementing boundary segment replacement and interpolation repair strategies in the strong influence sub-region, the automatic correction and structured reconstruction of the pest area boundary are realized, thereby generating a corrected area map that is more consistent with the actual pest propagation characteristics. The above methods not only improve the timeliness and accuracy of pest area identification, but also enhance the system's adaptability to the dynamic diffusion process of pests, providing a reliable basis for subsequent precise greening maintenance and decision-making, and significantly improving the intelligence and automation level of greening pest control.

[0030] Furthermore, in S400, the specific method for initializing the decision tree model by training the decision tree model with the second image is as follows:

[0031] The pest-infested regions in the second image are divided into training and validation sets. Specifically, each pest-infested region is labeled as a positive sample (1), and gray pest-infested regions with a diffusion coefficient greater than the pixel average of the second image are labeled as negative samples (0), thus completing the sample labeling. The labeled positive and negative samples constitute the training sample set. The training sample set is then divided into a training set and a validation set in a 4:1 ratio. The training set is used to train the decision tree model, and the validation set is used to verify the prediction performance of the trained decision tree model.

[0032] Furthermore, the decision tree model is the GBDT model.

[0033] The decision tree model is trained using a training set and a validation set to obtain a pre-trained decision tree model. The method for training the decision tree model using the training set and validation set is as follows: extract the features of the training set and input them into the decision tree model, optimize the hyperparameters in the decision tree model using a grid search method, and retrain the decision tree model according to the optimized hyperparameters to obtain a trained initial decision tree model.

[0034] Furthermore, the specific method for identifying maintenance locations in the greening project area map by initializing the decision tree model in S500 is as follows:

[0035] The hyperspectral image of the greening project area to be maintained was obtained by a drone equipped with a spectrometer. Multiple pest areas were identified by the RX algorithm. The positive sample areas among the multiple pest areas were screened using an initial decision tree model. If a positive sample area could be screened, it was determined that there were spread pest areas in the greening project area to be maintained. The geometric center of all spread pest areas was taken as the maintenance location, and the application path was generated for all maintenance locations by curve fitting.

[0036] Beneficial Effects: This invention, by integrating hyperspectral image analysis and environmental factor modeling, enables precise location and dynamic spread prediction of pest and disease areas. It effectively overcomes the limitations of existing highway greening pest and disease monitoring methods, which rely on manual experience, have slow response times, and limited coverage. By introducing a decision tree model to train and learn from pest image features and environmental parameters, the system can autonomously extract key influencing factors, achieve intelligent identification of pest and disease areas and judgment of spread trends, and generate targeted maintenance suggestions based on the training results. Compared with traditional static image recognition methods, this invention uses a decision tree classification mechanism with high interpretability and strong generalization ability, improving the accuracy and reliability of pest and disease detection. It is particularly suitable for highway scenarios with wide distribution of green belts and rapid environmental changes, significantly enhancing the intelligence level and practical value of pest and disease monitoring systems.

[0037] This invention also provides a highway greening engineering maintenance system based on machine learning. The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program within the following system units:

[0038] The path generation unit is used to acquire a map of the greening project area to be maintained and generate a data acquisition path.

[0039] The image acquisition unit is used to control the UAV to fly along the acquisition path and acquire multiple hyperspectral images of the region.

[0040] An image stitching unit is used to generate a first image based on multiple acquired hyperspectral images;

[0041] Pest marking unit, used to mark the pest-infested area in the first image;

[0042] The coefficient calculation unit is used to calculate the diffusion coefficient of all marked pest areas;

[0043] An image correction unit is used to correct the first image according to the calculated diffusion coefficient to obtain the second image;

[0044] The model training unit is used to train the decision tree model with the second image to obtain the initial decision tree model.

[0045] The beneficial effects of this invention are as follows: By integrating hyperspectral image analysis and environmental factor modeling, this invention can achieve precise location and dynamic spread prediction of pest and disease areas, effectively overcoming the limitations of existing highway greening pest and disease monitoring methods that rely on manual experience, have slow response times, and limited coverage. By introducing a decision tree model to train and learn pest image features and environmental parameters, the system can autonomously extract key influencing factors, achieve intelligent identification of pest and disease areas and judgment of spread trends, and generate targeted maintenance suggestions based on the training results. Compared with traditional static image recognition methods, this invention uses a decision tree classification mechanism with high interpretability and strong generalization ability, improving the accuracy and reliability of pest and disease detection. It is particularly suitable for highway scenarios with wide distribution of green belts and rapid environmental changes, significantly enhancing the intelligence level and practical value of the pest and disease monitoring system. Attached Figure Description

[0046] Figure 1 The diagram shows a flowchart of a highway greening engineering maintenance method based on machine learning.

[0047] Figure 2 The diagram shows a structure of a highway greening maintenance system based on machine learning. Detailed Implementation

[0048] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0049] Example 1:

[0050] Figure 1 The diagram shows a flowchart of a highway greening engineering maintenance method based on machine learning.

[0051] Reference Figure 1 This invention proposes a method for highway greening engineering maintenance based on machine learning, the method comprising the following steps:

[0052] S100: Obtain a map of the greening project area to be maintained and generate a collection path. The drone will fly along the collection path to obtain multiple hyperspectral images of the area map.

[0053] S200, Generate a first image based on the acquired multiple hyperspectral images;

[0054] S300, mark the pest-infested areas in the first image and update it to the second image;

[0055] S400, The initial decision tree model is obtained by training the decision tree model with the second image.

[0056] Furthermore, it also includes step S500, which identifies the maintenance location from the second image by initializing the decision tree model.

[0057] Furthermore, the UAV is a fixed-wing UAV or a rotary-wing UAV that includes a hyperspectral imaging system. The initial parameters of the UAV are set as follows: the field of view of the hyperspectral imaging system is 16°, the flight altitude is 30m, the flight speed is 15m, the flight path spacing is 30m, the boundary safety distance is 5m, and the obstacle safety distance is 5m.

[0058] Furthermore, in S100, the specific method for acquiring a map of the greening project area to be maintained and generating a collection path, and then having the UAV fly along the collection path to acquire multiple hyperspectral images of the area map, is as follows: acquire a map of the greening project area to be maintained, set the start point, end point, and obstacle location information of the UAV's flight in the greening project area to be maintained, generate the flight path of the UAV from the start point to the end point, generate collection points at fixed distances, and have the UAV fly along the path to acquire spectral images at the collection points using the onboard spectrometer.

[0059] Furthermore, the drone is a coaxial dual-rotor drone equipped with a spectrometer.

[0060] Furthermore, the spectrometer is a GaiaField-mini spectrometer.

[0061] Furthermore, the flight path of the drone from the starting point to the destination is generated using the A* algorithm to generate the drone data collection path from the starting point to the destination.

[0062] Furthermore, the starting point is the location where the drone takes off in the area to be monitored, and the ending point is the destination location of the drone's flight in the area to be monitored. The obstacle location information is the location of the obstacles in the greening project area to be maintained. The drone must maintain a preset obstacle safety distance (5m) from the obstacle location.

[0063] Furthermore, in S200, the specific method for generating the first image based on the acquired multiple hyperspectral images is as follows: all the acquired multispectral images are stitched together in the order of acquisition time to generate the first image.

[0064] Furthermore, the stitching process includes the following steps: geometric correction, image preprocessing, image registration, and image fusion.

[0065] Furthermore, in S300, the specific method for marking the pest-infested areas in the first image and updating it to the second image is as follows: Multiple pest-infested areas PA are marked in the first image using the RX algorithm, with i as the index of the pest-infested area, PA... iRepresenting the i-th infested region, all infested regions are grouped into a set INP; the first image is divided into multiple sub-regions RA using an edge detection algorithm. j Let j be the index of the subregion, RA j This represents the j-th subregion;

[0066] Let P(i) be the geometric center of all pest-infested areas and R(i) be the geometric center of all sub-regions. Traverse all pest-infested areas in the set INP and calculate PR(j) the distance between the geometric center P(i) of all pest-infested areas and the geometric center R(i) of all projections. Add all projections whose PR(j) is less than PR_mean to the list Listi, where PR_mean is the average distance between point P(i) and all points R(i).

[0067] Furthermore, the pest-affected area is the region composed of pest-affected pixels, or the internal region of the edge composed of pest-affected pixels; wherein, pest-affected pixels are pixels in a spectral image whose visible light reflectance is higher than the average visible light reflectance of each pixel in the spectral image, or, pest-affected pixels are pixels in a spectral image whose near-infrared spectral reflectance is lower than the average near-infrared spectral reflectance of each pixel in the spectral image.

[0068] Within the range of values ​​for k, all sub-regions of Listi are traversed sequentially, and the pest-affected region PA is calculated using a formula. i The diffusion coefficient Diff in each sub-region of Listi is calculated as follows: The average pixel value in each sub-region is GreenK_Mean. The maximum pixel value in the k-th sub-region is GreenK_Max, and the minimum pixel value is GreenK_Min. The pixel value of the edge point in the sub-region is Green_edg(k,r), where K represents the sub-region index in Listi, r represents the edge point index, and Green_edg(k,r) represents the pixel value of the r-th edge point in the k-th sub-region. This is achieved using the formula: Calculate the diffusion coefficient Diff, where q(k) is the distance between the pixel with the maximum value and the pixel with the minimum value in the k-th sub-region, and PR(k) represents the pest-infested region PA. i The distance between the geometric center of the region and the geometric center of the subregion;

[0069] If Diff is greater than Green_Mean, where Green_Mean is the average value of pixels in all sub-regions contained in List1, then it is considered that the current pest area has spread and needs to be corrected; otherwise, the current pest area has not spread and does not need to be corrected.

[0070] Furthermore, the specific correction method is as follows: within the range of values ​​for k, the diffusion coefficient Diff of the pest-affected area in each sub-region of Listi is calculated sequentially. k Where k = 1, 2, 3...m, m represents the number of sub-regions contained in Listi. All sub-regions contained in Listi are rearranged in ascending order of Diff, and the diffusion coefficients Diff of the sub-regions in Listi are iterated and updated sequentially. k If satisfied, Diff k -Diff k-1 <Diff k+1 -Diff k And Diff k >Diff mean Diff mean Let Diff represent the average diffusion coefficient (Diff) of the sub-regions in Listi. Then, Diff... k The corresponding sub-region is denoted as the strongly influential sub-region.

[0071] Traverse the infested area PA in sequence i Let PL be the pixel value on the edge. Select the minimum pixel value as the starting point ST and the maximum pixel value as the ending point ED. Starting from ST, search for PA along the minimum path to ED. i The magnitude of the pixel value PL on the edge, with s and q as the index of the pixel value on the edge, if satisfying and At that time, update the pixel corresponding to PLs to the starting point ST; starting from ED, search for PA along the minimum path to ST. i The size of the pixel value PL on the edge, with s as the index of the pixel value on the edge, if satisfying and At that time, update the pixel corresponding to PLs to the starting point ED, where z represents PA. i The number of pixels on the edge;

[0072] Mark the position of the pixel corresponding to ST as point A, and the position of the pixel corresponding to ED as point B. Mark the minimum curve segment between points A and B as the deletion segment and delete it. Then, sequentially strengthen the pixel values ​​on the edges of the affected sub-regions, marking the point corresponding to the maximum pixel value as point C and the point corresponding to the minimum pixel value as point D. Mark the minimum curve segment between points C and D as the replacement segment. Copy the replacement segment and scale it proportionally to the area PA between points A and B to correct the pest infestation. i ;

[0073] Correct all insect-infested areas in the first image using the method described above, and update the first image to the second image.

[0074] The pest-infested regions in the second image are divided into training and validation sets. Specifically, each pest-infested region is labeled as a positive sample (1), and gray pest-infested regions with a diffusion coefficient greater than the pixel average of the second image are labeled as negative samples (0), thus completing the sample labeling. The labeled positive and negative samples constitute the training sample set. The training sample set is then divided into a training set and a validation set in a 4:1 ratio. The training set is used to train the decision tree model, and the validation set is used to verify the prediction performance of the trained decision tree model.

[0075] Furthermore, the decision tree model is the GBDT model.

[0076] The decision tree model is trained using a training set and a validation set to obtain a pre-trained decision tree model. The method for training the decision tree model using the training set and validation set is as follows: extract the features of the training set and input them into the decision tree model, optimize the hyperparameters in the decision tree model using a grid search method, and retrain the decision tree model according to the optimized hyperparameters to obtain a trained initial decision tree model.

[0077] Furthermore, the specific method for identifying maintenance locations in the greening project area map by initializing the decision tree model in S500 is as follows:

[0078] The hyperspectral image of the greening project area to be maintained was obtained by a drone equipped with a spectrometer. Multiple pest areas were identified by the RX algorithm. The positive sample areas among the multiple pest areas were screened using an initial decision tree model. If a positive sample area could be screened, it was determined that there were spread pest areas in the greening project area to be maintained. The geometric center of all spread pest areas was taken as the maintenance location, and the application path was generated for all maintenance locations by curve fitting.

[0079] Furthermore, this invention also provides an embodiment of a highway greening engineering maintenance system based on machine learning, such as... Figure 2The diagram shows a structure of a highway greening maintenance system based on machine learning according to the present invention. This embodiment of the highway greening maintenance system based on machine learning includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above embodiment of the highway greening maintenance system based on machine learning.

[0080] The path generation unit is used to acquire a map of the greening project area to be maintained and generate a data acquisition path.

[0081] The image acquisition unit is used to control the UAV to fly along the acquisition path and acquire multiple hyperspectral images of the region.

[0082] An image stitching unit is used to generate a first image based on multiple acquired hyperspectral images;

[0083] Pest marking unit, used to mark the pest-infested area in the first image;

[0084] The coefficient calculation unit is used to calculate the diffusion coefficient of all marked pest areas;

[0085] An image correction unit is used to correct the first image according to the calculated diffusion coefficient to obtain the second image;

[0086] The model training unit is used to train the decision tree model with the second image to obtain the initial decision tree model.

[0087] The aforementioned machine learning-based highway greening maintenance system can run on computing devices such as desktop minicomputers, laptops, handheld computers, and cloud servers. The system that can run on this machine learning-based highway greening maintenance system may include, but is not limited to, processors and memory. Those skilled in the art will understand that the example described is merely an illustration of a machine learning-based highway greening maintenance system and does not constitute a limitation on such a system. It may include more or fewer components, or a combination of certain components, or different components. For example, the machine learning-based highway greening maintenance system may also include input / output devices, network access devices, buses, etc.

[0088] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the machine learning-based highway greening engineering maintenance system, connecting various parts of the system through various interfaces and lines.

[0089] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the machine learning-based highway greening engineering maintenance system by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0090] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A method for maintaining a highway greening project based on machine learning, characterized in that, The method comprises the following steps: S100, acquiring a green engineering area to be maintained and generating a collection path, and flying a UAV along the collection path to acquire a plurality of hyperspectral images of the green engineering area to be maintained; S200, generating a first image from the plurality of acquired hyperspectral images; S300, marking a pest area in the first image and updating it as a second image; S400, training a decision tree model using the second image to obtain an initialized decision tree model; S500, identifying a maintenance site in the second image through the initialized decision tree model; In S300, the method for marking the pest area in the first image and updating it as the second image is: All sub-regions in List i are traversed in sequence, and the i-th pest region PA i is calculated in each sub-region in List i , and the diffusion coefficient Diff k is calculated, wherein k=1, 2, 3…m, m represents the number of sub-regions contained in List i , if the diffusion coefficient Diff k is greater than the average pixel value contained in each sub-region, it is considered that the current pest region has a diffusion phenomenon, and the current pest region needs to be corrected, otherwise the current pest region does not have a diffusion phenomenon, and the current pest region does not need to be corrected. The specific correction method is: According to the diffusion coefficient Diff k The average value of the diffusion coefficient of the sub-region in List i Diff k The corresponding sub-region is recorded as a strong influence sub-region; The i-th pest-affected area PA will be traversed sequentially. i Let PL be the pixel value on the edge. Select the minimum pixel value as the starting point ST and the maximum pixel value as the ending point ED. Starting from ST, search for PA along the minimum path to ED. i The magnitude of the pixel value PL on the edge, with s and q as the index of the pixel value on the edge, if satisfying ,and At that time, PL s The corresponding pixel is updated to the starting point ST; PA is retrieved from ED along the minimum path to ST. i The size of the pixel value PL on the edge, with s as the index of the pixel value on the edge, if satisfying ,and At that time, PL s The corresponding pixel is updated to the starting point ED, where z represents PA. i The number of pixels on the edge; The position of the pixel point corresponding to ST is recorded as point A, the position of the pixel point corresponding to ED is recorded as point B, the minimum curve segment between point A and point B is recorded as a deletion segment and is deleted, the pixel values on the edge of the strong influence sub-region are traversed in turn, the point corresponding to the maximum pixel value is recorded as point C, the point corresponding to the minimum pixel value is recorded as point D, the minimum curve segment between point C and point D is recorded as a replacement segment, the replacement segment is copied and pasted between point A and point B by equal ratio scaling to correct the pest damage area PA i ; so as to correct all pest damage areas in the first image to obtain the second image.

2. The method of claim 1, wherein the method is characterized by, In S100, the method for acquiring the green engineering area to be maintained and generating the collection path, and flying the UAV along the collection path to acquire the plurality of hyperspectral images of the green engineering area to be maintained is: acquiring the green engineering area to be maintained, setting the starting point, the ending point and the obstacle position information of the UAV flying in the green engineering area to be maintained, generating a flight path of the UAV from the starting point to the ending point, and generating collection points at fixed distances, and the UAV flying according to the path and collecting a plurality of hyperspectral images at the collection points through the carried spectrometer. 3.The highway greening engineering maintenance method based on machine learning according to claim 2, characterized in that, In S200, the method for generating the first image from the plurality of acquired hyperspectral images is: splicing all the plurality of acquired hyperspectral images in the order of collection time to generate the first image.

4. The method of claim 1, wherein, In S400, the method for training the decision tree model using the second image to obtain the initialized decision tree model is: dividing all the pest areas contained in the second image into a training set and a validation set, specifically: taking all the pest areas as positive samples and labeling all the positive samples as 1, and screening out pest areas with a diffusion coefficient greater than the average value of the pixels of the second image as negative samples and labeling all the negative samples as 0; The labeled positive samples and negative samples constitute a training sample set; the training sample set is divided into a training set and a validation set according to a ratio of 4:1, the training set is used to train the decision tree model, and the validation set is used to verify the prediction performance of the trained decision tree model; The training set and the validation set are used to train the decision tree model to obtain a pre-trained decision tree model, wherein the method for training the decision tree model using the training set and the validation set is: extracting features of the training set and inputting the decision tree model, optimizing hyperparameters in the decision tree model using a grid search method, and retraining the decision tree model according to the optimized hyperparameters to obtain a trained initialized decision tree model.

5. A highway green engineering maintenance system based on machine learning, characterized in that, The highway green engineering maintenance system based on machine learning comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the highway green engineering maintenance method based on machine learning according to any one of claims 1-4 when executing the computer program.

Images