Garden vegetation accurate recovery method and device based on unmanned aerial vehicle intelligent monitoring

Abstract

The invention relates to a garden vegetation accurate recovery method and device based on unmanned aerial vehicle intelligent monitoring, and the method comprises the following steps: 110, obtaining a multispectral image of a target monitoring garden through an unmanned aerial vehicle, the multispectral image comprising a plurality of NIR wave band images and a plurality of Red wave band images which are continuously shot at the same position at extremely short intervals; 120, processing the plurality of NIR band images and the plurality of Red band images in an image fusion mode to obtain an NIR fusion image and a Red fusion image, and then preprocessing the NIR fusion image and the Red fusion image; step 130, constructing an NDVI model based on the preprocessed NIR fusion image and the Red fusion image, and determining a vegetation to-be-recovered region of the target detection garden according to the NDVI model; and step 140, carrying out vegetation recovery on the vegetation to-be-recovered region. According to the method, the to-be-recovered vegetation area in the target detection garden is precisely recovered through the unmanned aerial vehicle, the recovery accuracy and efficiency are improved, and the influence of the environment can be overcome to a great extent.

Description

Technical Field

[0001] The present invention relates to the technical field related to image processing and vegetation restoration, and in particular to a method and device for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles. Background Art

[0002] Due to the constraints of many environmental factors and insufficient technical support, some aspects of traditional forestry are relatively extensive and inefficient. For example, traditional forestry vegetation resource surveys are labor-intensive, costly, and time-consuming, and are usually limited to sample plots, making it difficult to obtain spatial distribution information of forestry vegetation parameters on a larger scale. In areas where afforestation is difficult due to high mountains, steep slopes, and inconvenient transportation, traditional artificial tree planting methods are labor-intensive, inefficient, and involve certain operational risks. Existing drone-based vegetation restoration methods also have limitations in this operating environment, such as low planting accuracy and low survival rate of sown vegetation seeds.

[0003] With the continuous advancement of drone technology, the multispectral or hyperspectral cameras they carry can collect positioning data and build ecological models. Using relevant software, they can monitor and understand the growth status of forest vegetation, enabling intelligent drone monitoring and rapid diagnosis of vegetation status. This provides new possibilities for addressing issues existing in traditional operation models, such as low efficiency, low accuracy, and high operating costs. The application of drone technology can not only improve the efficiency and accuracy of forestry work and reduce labor intensity, but also help enhance the level of informationization and intelligentization in forestry, providing technical support for its sustainable development. However, when drones are collecting images, random noise may interfere with the images, and details cannot be guaranteed, which in turn affects the subsequent determination of areas for garden vegetation restoration. Summary of the Invention

[0004] The purpose of the present invention is to solve at least one of the deficiencies of the prior art and to provide a method and device for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: Specifically, a precise restoration method for garden vegetation based on UAV intelligent monitoring is proposed, including the following: Step 110: Acquire a multispectral image of the target monitored garden by a drone, wherein the multispectral image includes a plurality of NIR band images and a plurality of Red band images taken continuously at the same position at very short intervals; Step 120: Process the multiple NIR band images and the multiple Red band images by image fusion to obtain an NIR fused image and a Red fused image, and then pre-process the NIR fused image and the Red fused image respectively; Step 130: constructing an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determining the vegetation restoration area of ​​the target detection garden according to the NDVI model; Step 140: Perform vegetation restoration on the area to be restored.

[0006] Furthermore, specifically, the method of restoring vegetation in the area to be restored is to spray planting capsules in the area to be restored by using a drone, wherein the planting capsule is bullet-shaped, the capsule shell is integrally formed, and the conical head cavity and the column cavity are respectively used to fill the fertilizers, seeds and medicines required for planting growth.

[0007] Furthermore, specifically, the plurality of NIR band images and the plurality of Red band images are processed by image fusion to obtain an NIR fused image and a Red fused image, including: For NIR band images or Red band images, if there are N images in total, and they are sorted in the order of acquisition to obtain the image set {TP_1, ..., TP_t, ..., TP_N}, t is a variable, t∈[1, N]; First, the N images are fused by averaging the pixel values ​​of the same pixel to obtain a classic fused image; Next, among the N images, find the detail expression area of ​​each image; Then execute the following loop program: Step 210: Initialize the variable t=1 and load the classic fusion image; Step 220: Replace the corresponding pixel position in the classic fusion image with the detail expression area in TP_t to update the classic fusion image; Step 230: Increase the value of t by 1, and then determine whether t is greater than N. If so, go to step 240; otherwise, go to step 220. Step 240: Output the classic fusion image at this time and record it as the NIR fusion image or the Red fusion image.

[0008] Furthermore, specifically, in N images, the detail expression area of ​​each image is found, including: First, for any image TP_t, let photo_t(t,mt,nt) be the pixel value of the pixel at the mth row and nth column in its image matrix photo_t(t); Next, calculate the detail rendering coefficient coefficient(t,mt,nt) corresponding to photo_t(t,mt,nt) as follows: ; in, 、 Represents the average, maximum, and minimum values ​​of the pixel values ​​of the neighborhood pixels of photo_t(t,mt,nt); Finally, the detail rendering areas of all images in N images are found based on the detail rendering coefficient, including: For TP_1, traverse its image matrix photo_1(1), find the top two pixel points with the highest detail rendering coefficient and record them as Max_point_1 and Max_point_2, connect Max_point_1 and Max_point_2 to get the straight line line_1,2, and then find the pixel point Max_point_3 with the largest detail rendering coefficient among the pixels other than the straight line line_1,2 in photo_1(1). The area formed by Max_point_1, Max_point_2 and Max_point_3 as vertices is the detail rendering area of ​​TP_1. For TP_t, calculate the average pixel value of the pixel points in the detail expression area of ​​photo_t-1(t-1) , and then traverse its image matrix photo_t(t), and find out the detail presentation coefficient greater than Let point_1 and point_2 be the random two pixels, and connect point_1 and point_2 to get line_point_1,2. Then find Max_point_4, the pixel with the largest detail presentation coefficient among the pixels other than line_point_1,2 in photo_t(t). The area formed by point_1, point_2 and Max_point_4 as vertices is the detail presentation area of ​​TP_t.

[0009] Furthermore, specifically, the preprocessing includes: The NIR fusion image and the Red fusion image are registered, and the registration includes geometric correction and radiation correction. Then, normalization processing is performed on each of them to scale the pixel values ​​to the range of [0, 1].

[0010] Furthermore, specifically, an NDVI model is constructed based on the pre-processed NIR fusion image and the Red fusion image, and the vegetation restoration area of ​​the target detection garden is determined according to the NDVI model, including: The NDVI value is calculated using the traditional formula to construct the NDVI model, and areas with NDVI values ​​less than the preset threshold are identified as areas to be restored.

[0011] Furthermore, the method also includes constructing an NDVI model based on a pre-trained U-net network and determining an area of ​​the target garden to be detected where vegetation is to be restored.

[0012] The present invention also proposes a device for accurately restoring garden vegetation based on intelligent monitoring by drones, comprising the following: A data acquisition module is used to acquire multispectral images of the target monitored garden through a drone, wherein the multispectral images include multiple NIR band images and multiple Red band images taken continuously at the same position at very short intervals; An image fusion module is used to process multiple NIR band images and multiple Red band images by image fusion to obtain NIR fusion images and Red fusion images, and then preprocess the NIR fusion images and Red fusion images respectively; The vegetation restoration area determination module is used to construct an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determine the vegetation restoration area of ​​the target detection garden according to the NDVI model; The vegetation restoration execution module is used to restore vegetation in the area to be restored.

[0013] The beneficial effects of the present invention are: The present invention proposes a method and device for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles. On the one hand, unmanned aerial vehicle technology is used to collect multispectral images, and then image analysis is performed based on the collected images to construct an NDVI model. The vegetation restoration area of ​​the target detection garden is determined based on the NDVI model, and then the vegetation restoration area is accurately restored by the unmanned aerial vehicle, thereby overcoming the influence of environmental factors in the existing technology; on the other hand, considering that the random noise problem that may exist when the unmanned aerial vehicle collects images will interfere with the image, and the details cannot be guaranteed, which will affect the subsequent judgment of the garden vegetation restoration area, the above problems are overcome by collecting and fusing multiple images to ensure the accuracy of subsequent analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other features of the present disclosure will become more apparent through a detailed description of the embodiments shown in conjunction with the accompanying drawings. The same reference numerals in the drawings of the present disclosure represent the same or similar elements. Obviously, the drawings described below are only some embodiments of the present disclosure. It is possible for a person skilled in the art to derive other drawings based on these drawings without inventive effort. In the drawings: Figure 1 Shown is a flow chart of the garden vegetation precision restoration method based on UAV intelligent monitoring of the present invention; Figure 2The figure shows a flow chart of the construction of the garden vegetation precision restoration method based on UAV intelligent monitoring of the present invention; Figure 3 The figure shows an example of selecting a UAV route in a specific application of the present invention; Figure 4 Shown is an example diagram of the implant capsule involved in the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless there is a conflict. The same reference numerals used throughout the drawings indicate the same or similar parts.

[0016] Example 1, reference Figure 1 as well as Figure 2 The present invention proposes a precise restoration method for garden vegetation based on UAV intelligent monitoring, which includes the following: Step 110: Acquire a multispectral image of the target monitored garden by a drone, wherein the multispectral image includes a plurality of NIR band images and a plurality of Red band images taken continuously at the same position at very short intervals; Step 120: Process the multiple NIR band images and the multiple Red band images by image fusion to obtain an NIR fused image and a Red fused image, and then pre-process the NIR fused image and the Red fused image respectively; Step 130: constructing an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determining the vegetation restoration area of ​​the target detection garden according to the NDVI model; Step 140: Perform vegetation restoration on the area to be restored.

[0017] In this embodiment 1, on the one hand, drone technology is used to collect multispectral images, and then image analysis is performed based on the collected images to construct an NDVI model. The vegetation to be restored area of ​​the target detection garden is determined based on the NDVI model, and then the vegetation to be restored area is accurately restored by the drone, thereby overcoming the influence of environmental factors in the existing technology; on the other hand, considering that the problem of random noise that may exist when the drone collects images will interfere with the picture, and the details cannot be guaranteed, which will affect the subsequent judgment of the garden vegetation to be restored area, the above problems are overcome by collecting and fusing multiple images to ensure the accuracy of subsequent analysis.

[0018] In addition, when performing specific operations, Construction Preparation 1. Before construction, it is necessary to conduct a field survey on site to understand the terrain, landform, vegetation conditions and other information of the construction area to provide a basis for subsequent drone flight path planning and data collection.

[0019] 2. Before conducting a drone aerial survey, you must first determine the coordinate system, elevation system, and control points of the area to be surveyed. Develop a flight plan based on this information. Use the DJI Map software to determine the flight frequency, define the flight range, and clarify parameters such as ground resolution, heading and lateral overlap, and route angle.

[0020] 3. Check meteorological data to ensure that the weather conditions on the day of flight meet the flight requirements, with visibility greater than 3,000 meters and wind speed less than level 5.

[0021] 4. Provide clear instructions to drone operators, clarifying the precautions and key points of operation.

[0022] UAV equipment onboard debugging Install the hyperspectral camera and other data acquisition equipment on the drone, ensuring they are stable and will not affect flight. Connect the cables of the hyperspectral camera and other equipment to the drone's interface. Turn on the drone's power and connect it to the mobile app. Use the mobile app to debug the drone's flight parameters and various parameters of the data acquisition equipment. Debug various devices to ensure normal performance during flight.

[0023] Data collection and processing 1. Upon arrival at the site, determine the drone's flight altitude, perform horizontal and compass calibration, and prepare for takeoff. Connect the drone's remote controller and control the drone's takeoff and aerial survey using the DJI Terra software. Set the flight altitude to 100 meters, the speed to 13 meters per second, the heading overlap rate to 75%, and the lateral overlap rate to 75%. The drone's flight path and the gimbal's pitch angle are automatically controlled by the software. Take aerial photos along the route. Figure 3 This is an example diagram of a specific drone route selection.

[0024] 2. The large amount of image data collected by drones can be processed and analyzed to obtain relevant information about forestry resources. Image processing techniques can be used to extract image features, such as vegetation index and vegetation coverage, to analyze forest growth and vegetation type distribution. This processing and analysis can be achieved through the establishment of a geographic information system (GIS). By combining drone data with geospatial data, the distribution and changing trends of forestry resources can be more accurately analyzed.

[0025] As a preferred embodiment of the present invention, specifically, the method of restoring vegetation in the area to be restored is to spray planting capsules in the area to be restored by using a drone, wherein the planting capsule is bullet-shaped, the capsule shell is integrally formed, and the conical head cavity and the column cavity are respectively used to fill the fertilizers, seeds and medicines required for planting growth.

[0026] Reference Figure 4 In this preferred embodiment, the termination capsule is launched by a drone, which can overcome the problem that artificial planting cannot be carried out due to the harsh environment. The specific process is as follows: 1. Use 3D modeling software to create and optimize a 3D model, then send the model to a printer for printing. The printing process typically uses layered processing and additive molding to complete 3D solid printing. After printing, the model needs to be post-processed, including removing support structures and surface treatment, to obtain the final product.

[0027] 2. The planting capsule is bullet-shaped, with an integral shell. The conical head cavity and column cavity are used to fill the fertilizer, seeds, and medication required for planting and growth, respectively. Fertilizer, seeds, rodent repellent and insect repellent, and water-retaining agent are filled into the capsule in that order, and the capsule ends are sealed using heat pressing.

[0028] 3. The planting capsule is made of a biodegradable material (polylactic acid, a substance extracted from corn starch). It is 2mm thick, 5-7cm long, and 2-3cm in diameter. It has a certain thickness and hardness to withstand impact with the ground, maintaining its integrity and protecting the seeds from birds, rodents, and other animals. The ratio of the capsule head cavity length to the seed cavity length is 1:2. This optimal ratio helps the seed capsule maintain a vertical drop after being ejected.

[0029] Assembly and debugging of high-pressure jet planting device 1. When planting capsules, a high-pressure jet device is used to launch the seeding. The high-pressure jet device has a circular capsule storage compartment on the top, which can hold 50 capsules at a time. During construction, the capsules are loaded into the compartment in sequence with the conical heads facing downward. Then, the sensor line of the high-pressure jet device is passed through the hose at the bottom of the drone and connected to the "BS" interface inside the aircraft. After the sensor line is connected, the seeding box is installed on the bottom of the drone.

[0030] 2. Calibrate the drone's positioning system before flight. Leveraging 5G technology, create a new "Flow Calibration" module and debug the high-pressure jet device to ensure it can operate from the drone. Set the jet pressure (0.1-0.3MPa), flight speed (10-13m / s), and select flow calibration to ensure timely feedback on the app during the jetting process.

[0031] Installation of intelligent pouring device Using the model obtained from drone monitoring, the coordinates to be restored are extracted through professional software and input into the drone controller. After completing the system calibration, the unmanned aerial vehicle is turned on and flown to the sky above the work area. After descending to the specified height, the high-field spraying function is started, so that the capsule is launched at an extremely high speed to penetrate the weeds on the ground and allow the planting capsule to be buried 3-5cm into the ground.

[0032] Drone monitoring and maintenance 1. After the restoration work is completed, the construction area will be monitored regularly. If necessary, drones equipped with small water tanks will be used to water the planted areas to promote germination and growth. Water enters through the small holes in the side walls of the seed cavity. The water-retaining agent in the cavity absorbs water and expands, providing moisture to the seed and exerting pressure on the capsule shell. The capsule shell is made of biodegradable biomass material, which degrades rapidly when exposed to water. During the degradation process, the biomass capsule also provides some nutrients for the seed's growth, which helps the seed germinate and break out of the shell when it encounters water again.

[0033] 2. Use drones to monitor plants in the area, including vegetation growth conditions and pests and diseases, to guide maintenance workers in their work.

[0034] As a preferred embodiment of the present invention, specifically, a plurality of NIR band images and a plurality of Red band images are processed by image fusion to obtain an NIR fused image and a Red fused image, including: For NIR band images or Red band images, if there are N images in total, and they are sorted in the order of acquisition to obtain the image set {TP_1, ..., TP_t, ..., TP_N}, t is a variable, t∈[1, N]; First, the N images are fused by averaging the pixel values ​​of the same pixel to obtain a classic fused image; Next, among the N images, find the detail expression area of ​​each image; Then execute the following loop program: Step 210: Initialize the variable t=1 and load the classic fusion image; Step 220: Replace the corresponding pixel position in the classic fusion image with the detail expression area in TP_t to update the classic fusion image; Step 230: Increase the value of t by 1, and then determine whether t is greater than N. If so, go to step 240; otherwise, go to step 220. Step 240: Output the classic fusion image at this time and record it as the NIR fusion image or the Red fusion image.

[0035] In this preferred embodiment, considering that the existing method often performs pixel averaging when performing image fusion, this method cannot effectively suppress random noise and may lose key details. Therefore, we propose the above-mentioned improved image fusion method, which gradually replaces the corresponding content in the classic fusion image obtained in the existing method with the detail expression area calculated in each image, so that the classic fusion image is always updated in the direction of more details (here, when looking for the detail expression area of ​​all images, it can be seen that each time the area searched is an estimated area higher than the mean of the previous image), thereby making the final fused image richer in details and suppressing random noise to a certain extent.

[0036] As a preferred embodiment of the present invention, specifically, finding the detail expression area of ​​each image in N images includes: First, for any image TP_t, let photo_t(t,mt,nt) be the pixel value of the pixel at the mth row and nth column in its image matrix photo_t(t); Next, calculate the detail rendering coefficient coefficient(t,mt,nt) corresponding to photo_t(t,mt,nt) as follows: ; in, 、 Represents the average, maximum, and minimum values ​​of the pixel values ​​of the neighborhood pixels of photo_t(t,mt,nt); Finally, the detail rendering areas of all images in N images are found based on the detail rendering coefficient, including: For TP_1, traverse its image matrix photo_1(1), find the top two pixel points with the highest detail rendering coefficient and record them as Max_point_1 and Max_point_2, connect Max_point_1 and Max_point_2 to get the straight line line_1,2, and then find the pixel point Max_point_3 with the largest detail rendering coefficient among the pixels other than the straight line line_1,2 in photo_1(1). The area formed by Max_point_1, Max_point_2 and Max_point_3 as vertices is the detail rendering area of ​​TP_1. For TP_t, calculate the average pixel value of the pixel points in the detail expression area of ​​photo_t-1(t-1) , and then traverse its image matrix photo_t(t), and find out the detail presentation coefficient greater than Let point_1 and point_2 be the random two pixels, and connect point_1 and point_2 to get line_point_1,2. Then find Max_point_4, the pixel with the largest detail presentation coefficient among the pixels other than line_point_1,2 in photo_t(t). The area formed by point_1, point_2 and Max_point_4 as vertices is the detail presentation area of ​​TP_t.

[0037] As a preferred embodiment of the present invention, specifically, the pretreatment includes: The NIR fusion image and the Red fusion image are registered, and the registration includes geometric correction and radiation correction. Then, normalization processing is performed on each of them to scale the pixel values ​​to the range of [0, 1].

[0038] As a preferred embodiment of the present invention, specifically, an NDVI model is constructed based on the pre-processed NIR fusion image and the Red fusion image, and the vegetation restoration area of ​​the target detection garden is determined according to the NDVI model, including: The NDVI value is calculated using the traditional formula to construct the NDVI model, and areas with NDVI values ​​less than the preset threshold are identified as areas to be restored.

[0039] As a preferred embodiment of the present invention, the method further includes constructing an NDVI model based on a pre-trained U-net network and determining the vegetation restoration area of ​​the target detection garden.

[0040] In this preferred embodiment, considering that the U-net network is suitable for image-to-image translation and can generate a higher quality NDVI model, the target detection and determination of the vegetation restoration area of ​​the garden are performed based on the pre-trained U-net network. The specific implementation process is as follows: The model design of the U-net network is as follows: Input: NIR image (single channel) and Red image (single channel), merged into 2-channel input.

[0041] Output: NDVI image (single channel, range [-1,1] or [0,1]).

[0042] Network structure: Encoder-Decoder Architecture (like U-Net): Encoder: Extracts multi-scale features (pre-trained models such as ResNet can be reused).

[0043] Decoder: restores image resolution and fuses high- and low-level features.

[0044] Skip connection: preserves detail information and improves edge and texture restoration capabilities.

[0045] Loss function: Mean squared error (MSE) or perceptual loss (PerceptualLoss), matching the true NDVI distribution.

[0046] Data processing, Data alignment: Ensure strict registration of NIR and Red images (geometric correction + radiometric correction).

[0047] Normalization: Scale pixel values ​​to the range [0,1].

[0048] Data enhancement: random cropping, rotation, and flipping to increase the diversity of training samples.

[0049] Training strategy Label generation: Calculate NDVI as a supervised label using the traditional formula: , Optimizer: Adam (learning rate = 1e-4), batch size = 8.

[0050] Evaluation metrics: Peak signal-to-noise ratio (PSNR), structural similarity (SSIM).

[0051] The vegetation restoration area of ​​the target detection garden is determined according to the NDVI model as follows: Threshold segmentation: Classification based on NDVI values ​​(example thresholds require domain knowledge adjustment): Healthy vegetation: NDVI ≥ 0.3, Mild degradation: 0.1≤NDVI<0.3, Severe degradation: NDVI < 0.1 (areas to be restored), Output a binary mask or classification map (e.g., using colors to mark different levels).

[0052] Morphological post-processing: The dilation / erosion operation eliminates isolated noise points and merges connected areas.

[0053] Dynamic threshold optimization (as the preferred solution): Adaptive thresholds based on historical data or statistical distribution (such as Otsu algorithm).

[0054] The present invention also proposes a device for accurately restoring garden vegetation based on intelligent monitoring by drones, comprising the following: A data acquisition module is used to acquire multispectral images of the target monitored garden through a drone, wherein the multispectral images include multiple NIR band images and multiple Red band images taken continuously at the same position at very short intervals; An image fusion module is used to process multiple NIR band images and multiple Red band images by image fusion to obtain NIR fusion images and Red fusion images, and then preprocess the NIR fusion images and Red fusion images respectively; The vegetation restoration area determination module is used to construct an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determine the vegetation restoration area of ​​the target detection garden according to the NDVI model; The vegetation restoration execution module is used to restore vegetation in the area to be restored.

[0055] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.

[0056] If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or system that can carry the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0057] Although the present invention has been described in considerable detail and with particularity with respect to several described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, but rather should be construed as providing a broad possible interpretation of these claims in view of the prior art by reference to the appended claims, thereby effectively encompassing the intended scope of the invention. In addition, the invention has been described above in terms of embodiments foreseen by the inventors for the purpose of providing a useful description, and those insubstantial modifications of the invention that are not currently foreseen may still represent equivalent modifications of the invention.

[0058] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. As long as the technical effects of the present invention are achieved by the same means, they shall fall within the scope of protection of the present invention. Within the scope of protection of the present invention, various modifications and variations of the technical solutions and / or implementation methods may be made.

Claims

1. A precise restoration method for garden vegetation based on UAV intelligent monitoring, characterized by: These include: Step 110: Acquire a multispectral image of the target monitored garden by a drone, wherein the multispectral image includes a plurality of NIR band images and a plurality of Red band images taken continuously at the same position at very short intervals; Step 120: Process the multiple NIR band images and the multiple Red band images by image fusion to obtain an NIR fused image and a Red fused image, and then pre-process the NIR fused image and the Red fused image respectively; Step 130: constructing an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determining the vegetation restoration area of ​​the target detection garden according to the NDVI model; Step 140: Perform vegetation restoration on the area to be restored.

2. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 1 is characterized in that: Specifically, the method of restoring vegetation in the area to be restored is to use a drone to spray planting capsules in the area to be restored. The planting capsules are bullet-shaped, the capsule shell is integrally formed, and the conical head cavity and the column cavity are respectively used to fill the fertilizers, seeds and medicines required for planting and growth.

3. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 1 is characterized in that: Specifically, multiple NIR band images and multiple Red band images are processed by image fusion to obtain NIR fused images and Red fused images, including: For NIR band images or Red band images, if there are N images in total, and they are sorted in the order of acquisition to obtain the image set {TP_1, ..., TP_t, ..., TP_N}, t is a variable, t∈[1, N]; First, the N images are fused by averaging the pixel values ​​of the same pixel to obtain a classic fused image; Next, find the detail-representing area of ​​each image in N images; Then execute the following loop program: Step 210: Initialize the variable t=1 and load the classic fusion image; Step 220: Replace the corresponding pixel position in the classic fusion image with the detail expression area in TP_t to update the classic fusion image; Step 230: Increase the value of t by 1, and then determine whether t is greater than N. If so, go to step 240; if not, go to step 220; Step 240: Output the classic fusion image at this time and record it as the NIR fusion image or the Red fusion image.

4. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 3 is characterized in that: Specifically, among N images, find the detail expression area of ​​each image, including: First, for any image TP_t, let photo_t(t,mt,nt) be the pixel value of the pixel at the mth row and nth column in its image matrix photo_t(t); Next, calculate the detail rendering coefficient coefficient(t,mt,nt) corresponding to photo_t(t,mt,nt) as follows: ； in, 、 Represents the average, maximum, and minimum values ​​of the pixel values ​​of the neighborhood pixels of photo_t(t,mt,nt); Finally, the detail rendering areas of all images in N images are found based on the detail rendering coefficient, including: For TP_1, traverse its image matrix photo_1(1), find the top two pixel points with the highest detail rendering coefficient and record them as Max_point_1 and Max_point_2, connect Max_point_1 and Max_point_2 to get the straight line line_1,2, and then find the pixel point Max_point_3 with the largest detail rendering coefficient among the pixels other than the straight line line_1,2 in photo_1(1). The area formed by Max_point_1, Max_point_2 and Max_point_3 as vertices is the detail rendering area of ​​TP_1. For TP_t, calculate the average pixel value of the pixel points in the detail expression area of ​​photo_t-1(t-1) , and then traverse its image matrix photo_t(t), and find out the detail presentation coefficient greater than Let point_1 and point_2 be the random two pixels, and connect point_1 and point_2 to get line_point_1,2. Then find Max_point_4, the pixel with the largest detail presentation coefficient among the pixels other than line_point_1,2 in photo_t(t). The area formed by point_1, point_2 and Max_point_4 as vertices is the detail presentation area of ​​TP_t.

5. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 1 is characterized in that: Specifically, the preprocessing includes: The NIR fusion image and the Red fusion image are registered, and the registration includes geometric correction and radiation correction. Then, normalization processing is performed on each of them to scale the pixel values ​​to the range of [0, 1].

6. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 1 is characterized in that: Specifically, an NDVI model is constructed based on the preprocessed NIR fusion image and the Red fusion image, and the vegetation restoration area of ​​the target detection garden is determined according to the NDVI model, including: The NDVI value is calculated using the traditional formula to construct the NDVI model, and areas with NDVI values ​​less than the preset threshold are identified as areas to be restored.

7. The method for accurately restoring garden vegetation based on intelligent monitoring by unmanned aerial vehicles according to claim 6 is characterized in that: The method further includes constructing an NDVI model based on a pre-trained U-net network and determining a vegetation restoration area of ​​the target detection garden.

8. A device for accurately restoring garden vegetation based on intelligent monitoring by drones, characterized in that: These include: A data acquisition module is used to acquire multispectral images of the target monitored garden through a drone. The multispectral images include multiple NIR band images and multiple Red band images taken continuously at the same location at very short intervals; An image fusion module is used to process multiple NIR band images and multiple Red band images by image fusion to obtain NIR fusion images and Red fusion images, and then preprocess the NIR fusion images and Red fusion images respectively; The vegetation restoration area determination module is used to construct an NDVI model based on the pre-processed NIR fusion image and the Red fusion image, and determine the vegetation restoration area of ​​the target detection garden according to the NDVI model; The vegetation restoration execution module is used to restore vegetation in the area to be restored.

Images