Single-tree vegetation index matching method, device, equipment, medium and product

Abstract

The invention discloses a single-tree vegetation index matching method and device, equipment, a medium and a product, and relates to the field of remote sensing image processing, and the method comprises the steps: obtaining a multispectral remote sensing image in an orchard region of interest; performing target detection, redundancy suppression and clustering combination on the multispectral remote sensing image to obtain a single tree detection frame; calculating various vegetation indexes according to the multispectral remote sensing image; generating a global canopy mask for the multispectral remote sensing image according to the various vegetation indexes; and matching the individual tree detection frame with the global canopy mask by using a coordinate matching algorithm to obtain an individual tree instance mask. According to the invention, the accuracy of multi-spectral pixel-level crown segmentation and information matching can be improved.

Description

Technical Field

[0001] The present application relates to the field of remote sensing image processing, and in particular to a single tree vegetation index matching method, device, equipment, medium and product. Background Art

[0002] With the development of smart orchards, individual fruit tree segmentation and crown information acquisition have become key technologies for intelligent orchard management and unmanned operations. Multispectral remote sensing images capture reflectance information from ground objects in multiple specific bands, such as near-infrared (NIR), green light (Green, G), red edge (RE), red light (Red, R), and blue light (Blue, B). Although the resolution of multispectral images is often lower than that of visible light images, pixel-level crown segmentation accuracy in multispectral remote sensing images is low, and the spectral information of individual fruit tree crowns cannot be accurately matched. Therefore, a method that can address this problem is needed. Summary of the Invention

[0003] The purpose of this application is to provide a single tree vegetation index matching method, device, equipment, medium and product, which can improve the accuracy of multispectral pixel-level crown segmentation and information matching.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a single tree vegetation index matching method, comprising:

[0006] Acquire multispectral remote sensing images of the orchard's area of ​​interest;

[0007] Performing target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame;

[0008] Calculating a plurality of vegetation indices according to the multispectral remote sensing image;

[0009] Generate a global canopy mask for multispectral remote sensing images based on multiple vegetation indices;

[0010] The single tree detection frame and the global canopy mask are matched using a coordinate matching algorithm to obtain a single tree instance mask.

[0011] In one embodiment, obtaining a multispectral remote sensing image within an area of ​​interest in an orchard specifically includes:

[0012] Obtain a multi-band multispectral remote sensing grayscale image of the orchard area of ​​interest; wherein the multi-band includes a green band, a near-infrared band, a red-edge band, and a red band;

[0013] Remote sensing grayscale images of any three bands in the multi-band multispectral remote sensing grayscale image are selected and superimposed to obtain a multispectral remote sensing image of the orchard area of ​​interest; the multispectral remote sensing image of the orchard area of ​​interest is a three-channel pseudo-color remote sensing image.

[0014] In one embodiment, target detection, redundancy suppression, and cluster merging are performed on the multispectral remote sensing image to obtain a single tree detection frame, specifically including:

[0015] Crop the multispectral remote sensing image using a sliding window to obtain a plurality of sub-images;

[0016] Use a computer deep learning model to perform target detection on each sub-image and generate a local bounding box in the sub-image;

[0017] Perform coordinate transformation based on the local bounding box in the sub-image to generate the global bounding box of the remote sensing image;

[0018] removing a global bounding box located at an edge of the image from the global bounding box of the remote sensing image to obtain a removed global bounding box;

[0019] According to the removed global bounding box, clustering and merging are performed using a non-maximum suppression algorithm and a breadth-first search algorithm to obtain a single tree detection box.

[0020] In one embodiment, the plurality of vegetation indices include normalized difference vegetation index, enhanced vegetation index, normalized difference water index, normalized difference red edge vegetation index and red edge-red light difference index.

[0021] In one embodiment, generating a global canopy mask based on a multispectral remote sensing image specifically includes:

[0022] Determining whether multiple vegetation indices of each pixel point of the multispectral remote sensing image are all greater than a set pre-threshold;

[0023] If yes, the pixel points of the multispectral remote sensing image are retained; if no, the pixel points of the multispectral remote sensing image are discarded;

[0024] Update the pixels of the multispectral remote sensing image until all pixels are judged;

[0025] A global canopy mask is generated based on the pixels of all retained multispectral remote sensing images.

[0026] In one embodiment, the single tree detection frame and the global canopy mask are matched using a coordinate matching algorithm to obtain a single tree instance mask, specifically including:

[0027] Traverse the single tree detection box and the global canopy mask;

[0028] Determining whether the pixel coordinates of the global canopy mask are within the coordinate range of the single tree detection frame;

[0029] If yes, determine the single tree instance mask based on the global canopy mask and the single tree detection frame and update the global canopy mask;

[0030] If not, update the single tree detection frame;

[0031] This continues until all single tree detection frames and the global canopy mask are matched.

[0032] In a second aspect, the present application provides a single tree vegetation index matching device, comprising:

[0033] An acquisition module is used to acquire multispectral remote sensing images of the orchard's area of ​​interest;

[0034] a single tree detection frame determination module, configured to perform target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame;

[0035] A multiple vegetation index calculation module, used to calculate multiple vegetation indices based on the multispectral remote sensing image;

[0036] A global canopy mask generation module is used to generate a global canopy mask for multispectral remote sensing images based on multiple vegetation indices;

[0037] The matching module is used to match the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask.

[0038] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the single tree vegetation index matching method.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the single tree vegetation index matching method when executed by a processor.

[0040] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the single tree vegetation index matching method when executed by a processor.

[0041] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0042] The present application provides a single tree vegetation index matching method, apparatus, equipment, medium and product, which obtain a single tree detection frame by performing target detection, redundancy suppression and cluster merging on the multispectral remote sensing image; calculate multiple vegetation indices based on the multispectral remote sensing image; generate a global canopy mask for the multispectral remote sensing image based on the multiple vegetation indices; match the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask, and improve the accuracy of multispectral pixel-level crown segmentation and information matching by generating the single tree detection frame and the global canopy mask, and then matching the single tree detection frame and the global canopy mask. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 This is the application environment map of the single tree vegetation index matching method;

[0045] Figure 2 This is a flowchart of the single tree vegetation index matching method;

[0046] Figure 3 This is a schematic diagram of the single tree vegetation index matching method;

[0047] Figure 4 This is a pseudo-color remote sensing image of an orchard;

[0048] Figure 5 Generate result map for single tree detection frame;

[0049] Figure 6 Heat maps of six vegetation indices;

[0050] Figure 7 Schematic diagram of the global canopy mask;

[0051] Figure 8 This is the result of the multispectral pixel-level single tree vegetation index matching method;

[0052] Figure 9 A schematic diagram of the functional modules of a single tree vegetation index matching device provided in another embodiment of the present application;

[0053] Figure 10 A schematic diagram of the structure of a computer device provided in one embodiment of the present application; DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0055] Multispectral bands can generate information on a variety of vegetation indices, providing greater differentiation and accuracy in subtasks such as monitoring fruit tree phenology and health, analyzing canopy nutrients like water and nitrogen, and investigating fruit tree pest and disease stress. This application achieves precise individual tree segmentation and individual tree vegetation index matching on multispectral remote sensing imagery, increasing the evaluation dimension and accuracy of fruit tree growth monitoring. It also provides data support for integrated water and fertilizer management, precise variable-rate spraying, and pest and disease control in smart orchards, providing guidance for precisely controlling inputs, reducing production costs, and achieving the green and sustainable development of smart orchards.

[0056] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0057] The single tree vegetation index matching method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the multispectral remote sensing image to be processed to the server 104. After the server 104 receives the multispectral remote sensing image to be processed, the server 104 performs target detection, redundancy suppression and cluster merging on the multispectral remote sensing image to be processed to obtain a single tree detection frame; calculates multiple vegetation indices based on the multispectral remote sensing image; generates a global canopy mask for the multispectral remote sensing image based on the multiple vegetation indices; matches the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask. The server 104 can feed back the obtained single tree instance mask to the terminal 102. In addition, in some embodiments, the single tree vegetation index matching method can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform single tree vegetation index matching on the multispectral remote sensing image to be processed, or the server 104 can obtain the multispectral remote sensing image to be processed from the data storage system and perform single tree vegetation index matching on the multispectral remote sensing image to be processed.

[0058] Terminal 102 may include, but is not limited to, various desktop computers, laptops, smartphones, tablet computers, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers, or may be a cloud server.

[0059] In an exemplary embodiment, Figure 2 and Figure 3 As shown, a single tree vegetation index matching method is provided. The method is executed by a computer device, specifically, it can be executed by a computer device such as a terminal or a server alone, or it can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 Taking the server 104 in FIG. 1 as an example, the method includes the following steps 201 to 208. In which:

[0060] Step 201: Acquire a multispectral remote sensing image of an orchard area of ​​interest;

[0061] Step 202: performing target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame;

[0062] Step 203: Calculating multiple vegetation indices based on the multispectral remote sensing image;

[0063] Step 204: generating a global canopy mask for the multispectral remote sensing image based on multiple vegetation indices;

[0064] Step 205: Match the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask.

[0065] By performing target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image, a single tree detection frame is obtained; multiple vegetation indices are calculated based on the multispectral remote sensing image; a global canopy mask is generated for the multispectral remote sensing image based on the multiple vegetation indices; the single tree detection frame and the global canopy mask are matched using a coordinate matching algorithm to obtain a single tree instance mask. By generating the single tree detection frame and the global canopy mask, the single tree detection frame and the global canopy mask are matched, thereby improving the accuracy of multispectral pixel-level crown segmentation and information matching.

[0066] In an exemplary embodiment, a multispectral remote sensing image of an orchard region of interest is obtained, specifically comprising: obtaining a multi-band multispectral remote sensing grayscale image of the orchard region of interest; wherein the multi-band includes a green band, a near-infrared band, a red edge wave, and a red band; selecting remote sensing grayscale images of any three bands in the multi-band multispectral remote sensing grayscale image for superposition to obtain a multispectral remote sensing image of the orchard region of interest; the multispectral remote sensing image of the orchard region of interest is a three-channel pseudo-color remote sensing image. In actual applications, a drone is used to collect multispectral remote sensing grayscale images of the green (G), near-infrared (NIR), red edge (RE), red (R), and blue (B) bands of the orchard region of interest, ensuring that the grayscale images of different bands have the same ground sample distance (GSD). The grayscale images of any three bands are superimposed to form a three-channel pseudo-color remote sensing image, which is then imported into a computer system. Figure 4 Four pseudo-color remote sensing images of orchard plots are shown, among which: Figure 4 (a) is a pseudo-color remote sensing image composed of three bands: G, NIR, and RE. Figure 4 (b) is a pseudo-color remote sensing image composed of three bands: G, NIR, and R. Figure 4 (c) is a pseudo-color remote sensing image composed of three bands: G, RE, and R. Figure 4 (d) is a pseudo-color remote sensing image composed of three bands: NIR, RE, and R. Figure 4 (e) is a three-channel visible light remote sensing image composed of three bands: R, G, and B. They are referred to as G-NIR-RE, G-NIR-R, G-RE-R, NIR-RE-R, and RGB in the following text.

[0067] In an exemplary embodiment, target detection, redundancy suppression, and cluster merging are performed on the multispectral remote sensing image to obtain a single tree detection frame, specifically including: cropping the multispectral remote sensing image using a sliding window to obtain multiple sub-images; performing target detection on each sub-image using a computer deep learning model to generate a local bounding box in the sub-image; performing coordinate transformation based on the local bounding box in the sub-image to generate a global bounding box of the remote sensing image; removing the global bounding box located at the edge of the image from the global bounding box of the remote sensing image to obtain a removed global bounding box; and performing cluster merging based on the removed global bounding box using a non-maximum suppression algorithm and a breadth-first search algorithm to obtain a single tree detection frame.

[0068] In practical applications, a computer system calls a single tree detection frame determination module and crops subimages from a multispectral remote sensing image using a fixed-size sliding window. A computer vision deep learning model trained on an orchard dataset performs target detection prediction on each subimage, generating local bounding boxes within the subimage. A global bounding box within the remote sensing image is then generated based on coordinate transformation. The computer vision deep learning model used in this application utilizes an existing model capable of performing instance segmentation tasks. Through two-stage instance refinement, the global bounding boxes are suppressed for redundancy and clustered and merged: the first step is to remove the global bounding boxes at the edge of the multispectral remote sensing image because they are targets of no interest; the second step is to use the non-maximum suppression algorithm to remove redundant and low-confidence global bounding boxes; the third step is to merge those global bounding boxes whose overlap rate exceeds the threshold based on the intersection-union ratio of the global bounding boxes; the fourth step is to use the breadth-first search algorithm to traverse all the global bounding boxes in the image, and based on the distance between the center points of the two global bounding boxes, put the global bounding boxes with a distance less than the threshold into a set to achieve cluster merging of global bounding boxes; the fifth step is to use the union-find data structure to store and classify the global bounding boxes on both sides of the slice line, and based on the degree of overlap and proximity of the edges of the global bounding boxes, put the global bounding boxes on both sides of the slice line but belonging to the same fruit tree into the same set to achieve cluster merging; the sixth step is to merge those pairs with sufficiently high overlap based on the intersection-union ratio of the global bounding boxes again. At this point, the algorithm generates a unique single tree detection frame with an independent ID for each fruit tree in the image, realizing single tree segmentation at the bounding box level in multispectral remote sensing images. Figure 5 The results of generating single tree detection frames for an orchard with high crown closure rate are shown. The first row in the figure shows the actual annotations, and the second row shows the predicted single tree detection frames. Figure 5 (a) is a G-NIR-RE pseudo-color remote sensing image. Figure 5 (b) is a G-NIR-R pseudo-color remote sensing image. Figure 5 (c) is a G-RE-R pseudo-color remote sensing image Figure 5 (d) is a NIR-RE-R pseudo-color remote sensing image Figure 5 (e) in the figure is an RGB three-channel remote sensing image. In an exemplary embodiment, the plurality of vegetation indices include normalized difference vegetation index, enhanced vegetation index, normalized difference water index, normalized difference red edge vegetation index and red edge-red light difference index.

[0069] The computer system calls a variety of vegetation index calculation modules to calculate multiple vegetation indices based on multispectral remote sensing images of green (G), near infrared (NIR), red edge (RE), and red (R) bands. Figure 6 Shown are heat maps of six common vegetation indices on an orchard plot. Figure 6(a) in the figure is the normalized difference vegetation index (NDVI) heat map. Figure 6 (b) is the Enhanced Vegetation Index (EVI) heat map. Figure 6 (c) is the red edge-red light difference index (DRE) heat map. Figure 6 (d) in the figure is the normalized difference water index (NDWI) heat map. Figure 6 (e) is the near-infrared-red light difference index (DVI) heat map. Figure 6 (f) in the figure is the normalized difference red edge index (NDRE) heat map.

[0070] In an exemplary embodiment, generating a global canopy mask based on a multispectral remote sensing image specifically includes: determining whether multiple vegetation indices of each pixel point of the multispectral remote sensing image are greater than a set pre-threshold; if so, retaining the pixel point of the multispectral remote sensing image; if not, discarding the pixel point of the multispectral remote sensing image; updating the pixel points of the multispectral remote sensing image until the determination of all pixel points is completed; and generating a global canopy mask based on all retained pixel points of the multispectral remote sensing image.

[0071] In practical applications, the computer system calls the global canopy mask generation module to generate a pixel-level canopy segmentation mask (canopymask). Background pixels are filtered using the mean of multiple vegetation indices, retaining foreground pixels belonging to the canopy. This involves using an AND gate to retain pixels where each vegetation index is greater than a threshold, thereby generating a global canopy mask. Each masked pixel stores multiple vegetation index values. The mean() function is defined to represent the mean of each vegetation index. Taking the six common vegetation indices mentioned above as an example, the conditions for filtering multiple vegetation indices using an AND gate can be written as: canopy mask = (NDVI > mean(NDVI)) & (EVI > mean(NDVI)) & (NDWI > mean(NDWI)) & (DVI > mean(DVI)) & (NDRE > mean(NDRE)) & (DRE > mean(DRE)). Figure 7 A global canopy mask is shown for an orchard plot with high canopy closure.

[0072] In an exemplary embodiment, the single tree detection frame and the global canopy mask are matched using a coordinate matching algorithm to obtain a single tree instance mask, specifically including: traversing the single tree detection frame and the global canopy mask; judging whether the pixel coordinates of the global canopy mask are within the coordinate range of the single tree detection frame; if so, determining the single tree instance mask based on the global canopy mask and the single tree detection frame and updating the global canopy mask; if not, updating the single tree detection frame; until all single tree detection frames and the global canopy mask are matched.

[0073] In practice, the computer system calls the matching module to coordinate-match the global canopy mask and the individual tree detection frame, generating a unique instance mask for each individual tree detection frame, also known as the individual tree instance mask. The instance mask is a two-dimensional array of the same size as the original image. For each individual tree detection frame and the corresponding instance mask, each pixel in the instance mask is assigned a unique instance ID (i.e., a unique tree ID) and stores multiple vegetation index values. This allows the acquisition of multiple vegetation indices for the canopy of an individual fruit tree, achieving vegetation index matching.

[0074] The logic of the coordinate matching algorithm is: assuming that the coordinates of the instance mask pixel are (x, y), and the single tree detection frame is represented as (x1, y1, x2, y2), traverse each single tree detection frame and the pixel points in the canopy mask. Assuming that the current traversal reaches the single tree detection frame Box i and pixel i , respectively expressed as (x i1 ,y i1 ,x i2 ,y i2 ) and (x i ,y i ). Then: if x i ∈[x i1 ,x i2 ]&y i ∈[y i1 ,y i2 ], then we can introduce pixel i ∈Box i , that is, the pixel pixel i Belong to Box i . Figure 8 A heat map was used to show the matching results of individual tree vegetation indices on orchard plots with high crown closure rates. Figure 8 (a) in the figure is the instance mask obtained by single tree segmentation. Figure 8 (b) in the figure is the matching result of NDVI vegetation index. Figure 8 (c) in the figure is the EVI vegetation index matching result. Figure 8 (d) in the figure is the matching result of NDWI vegetation index. Figure 8(e) in the figure is the matching result of NDRE vegetation index. Figure 8 (f) in the figure is the DVI vegetation index matching result. Figure 8 (g) in the figure is the DRE vegetation index matching result.

[0075] The present application relates to the field of agricultural remote sensing image processing technology, specifically a method for matching the vegetation index of individual fruit trees based on sliding window-assisted deep learning target detection and multiple vegetation index filtering. In response to the problems of low pixel-level crown segmentation accuracy and inaccurate matching of individual fruit tree crown spectral information in multispectral remote sensing images, the present application proposes a multispectral pixel-level individual tree vegetation index matching method, which realizes multispectral pixel-level crown segmentation and accurate information matching through algorithm modular design. The multispectral pixel-level individual tree vegetation index matching method consists of the following three core steps in sequence: (1) Generate a single tree detection frame, which is a pre-selected frame of a series of individual fruit tree crowns on a multispectral orthophoto map. (2) Generate a pixel-level canopy segmentation mask, calculate multiple vegetation indices based on multispectral remote sensing images in the green (G), near-infrared (NIR), red edge (RE), and red (R) bands, and use the mean of multiple vegetation indices to filter background pixels to generate a global canopy mask, where each mask pixel stores multiple vegetation index values. (3) Coordinate matching is performed between the global canopy mask and the single tree detection frame, and a unique instance mask, i.e., the single tree instance mask, is generated for each single tree detection frame. Multiple vegetation index values ​​are stored, i.e., multiple vegetation indices of the crown of a single fruit tree can be obtained, thus achieving vegetation index matching.

[0076] The method provided in this application has the following advantages:

[0077] (1) This application achieves accurate pixel-level segmentation of individual trees on multispectral remote sensing images from drones. Since the resolution of multispectral images is often lower than that of visible light images, there is a lack of research on directly using multispectral remote sensing images for pixel-level segmentation of individual fruit trees. Existing research focuses on using multispectral data to complete tree species identification and orchard area identification, or using vegetation indices calculated from multispectral bands to complete image background removal or image filtering, and then further using other image features to complete fruit tree identification; multispectral images are usually single-channel grayscale images, and their image features do not match the current computer vision SOTA large model architecture. These large models usually receive three-channel images as input, and fruit tree identification based on multispectral images mostly uses traditional processing algorithms, or uses machine learning models to classify fruit tree groups in multispectral images into tree species. There is a lack of technical means to use deep learning models for high-precision individual tree crown segmentation and crown width parameter extraction. This application receives three-channel multispectral pseudo-color images and inputs them into a deep learning model for training and inference. Satisfactory single-tree segmentation results are achieved on multispectral pseudo-color images with four band combinations. The target detection mAP50 on the sub-image reaches 0.99; the accuracy of fruit tree recognition and counting reaches 100%.

[0078] (2) This application achieves accurate spectral information matching on multispectral remote sensing images from drones. Existing research lacks technical means to achieve vegetation index registration for individual fruit trees using multispectral images. This application generates a single tree detection frame to achieve frame-level single tree segmentation; further generates a global canopy mask; and finally generates a single tree segmentation mask to match multiple vegetation indices to the crown of a single fruit tree. The mAP50 of the single tree detection frame reaches 1; the best mAP50 of the single tree segmentation mask reaches 0.8.

[0079] In another exemplary embodiment, a deep learning model can be used to directly complete instance segmentation level tasks on remote sensing images, but the resolution of multispectral grayscale images or pseudo-color images is low, the crown boundaries are blurred, and standardization and training are difficult. The cost and accuracy of achieving pixel-level single tree segmentation are not as good as the solution proposed in this application.

[0080] Based on the same inventive concept, embodiments of the present application also provide a single tree vegetation index matching device for implementing the single tree vegetation index matching method described above. The solution provided by this device is similar to the solution described in the method described above. Therefore, the specific limitations of one or more of the following single tree vegetation index matching device embodiments can be found in the limitations of the single tree vegetation index matching method described above and will not be repeated here.

[0081] In an exemplary embodiment, Figure 9 As shown, a single tree vegetation index matching device is provided, comprising:

[0082] Acquisition module 901, for acquiring multispectral remote sensing images of an orchard area of ​​interest;

[0083] A single tree detection frame determination module 902 is configured to perform target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame;

[0084] A multiple vegetation index calculation module 903 is used to calculate multiple vegetation indices based on the multispectral remote sensing image;

[0085] A global canopy mask generation module 904 is configured to generate a global canopy mask for a multispectral remote sensing image based on a plurality of vegetation indices;

[0086] The matching module 905 is configured to match the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask.

[0087] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store single tree vegetation index matching data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a single tree vegetation index matching method is implemented.

[0088] Those skilled in the art will understand that Figure 10 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the above-mentioned method embodiments when executing the computer program.

[0089] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the above-mentioned method embodiments when executed by a processor.

[0090] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the above method embodiments are implemented.

[0091] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0092] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0093] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0094] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0095] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A single tree vegetation index matching method, characterized in that: The single tree vegetation index matching method includes: Acquire multispectral remote sensing images of the orchard's area of ​​interest; Performing target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame; Calculating a plurality of vegetation indices according to the multispectral remote sensing image; Generate a global canopy mask for multispectral remote sensing images based on multiple vegetation indices; The single tree detection frame and the global canopy mask are matched using a coordinate matching algorithm to obtain a single tree instance mask.

2. The single tree vegetation index matching method according to claim 1, characterized in that: Acquire multispectral remote sensing images of the orchard area of ​​interest, including: Obtain a multi-band multispectral remote sensing grayscale image of the orchard area of ​​interest; wherein the multi-band includes a green band, a near-infrared band, a red-edge band, and a red band; Remote sensing grayscale images of any three bands in the multi-band multispectral remote sensing grayscale image are selected and superimposed to obtain a multispectral remote sensing image of the orchard area of ​​interest; the multispectral remote sensing image of the orchard area of ​​interest is a three-channel pseudo-color remote sensing image.

3. The single tree vegetation index matching method according to claim 1, characterized in that: Target detection, redundancy suppression, and cluster merging are performed on the multispectral remote sensing image to obtain a single tree detection frame, specifically including: Crop the multispectral remote sensing image using a sliding window to obtain a plurality of sub-images; Use a computer deep learning model to perform target detection on each sub-image and generate a local bounding box in the sub-image; Perform coordinate transformation based on the local bounding box in the sub-image to generate the global bounding box of the remote sensing image; removing a global bounding box located at an edge of the image from the global bounding box of the remote sensing image to obtain a removed global bounding box; According to the removed global bounding box, clustering and merging are performed using a non-maximum suppression algorithm and a breadth-first search algorithm to obtain a single tree detection box.

4. The single tree vegetation index matching method according to claim 1, characterized in that: The various vegetation indices include the Normalized Difference Vegetation Index, the Enhanced Vegetation Index, the Normalized Difference Water Index, the Normalized Difference Red Edge Vegetation Index, and the Red Edge-Red Light Difference Index.

5. The single tree vegetation index matching method according to claim 1, characterized in that: Generate a global canopy mask based on multispectral remote sensing images, including: Determining whether multiple vegetation indices of each pixel point of the multispectral remote sensing image are all greater than a set pre-threshold; If yes, the pixel points of the multispectral remote sensing image are retained; if no, the pixel points of the multispectral remote sensing image are discarded; Update the pixels of the multispectral remote sensing image until all pixels are judged; A global canopy mask is generated based on the pixels of all retained multispectral remote sensing images.

6. The single tree vegetation index matching method according to claim 1, characterized in that: Matching the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask, specifically including: Traverse the single tree detection box and the global canopy mask; Determining whether the pixel coordinates of the global canopy mask are within the coordinate range of the single tree detection frame; If yes, determine the single tree instance mask based on the global canopy mask and the single tree detection frame and update the global canopy mask; If not, update the single tree detection frame; This continues until all single tree detection frames and the global canopy mask are matched.

7. A single tree vegetation index matching device, characterized in that: The single tree vegetation index matching device includes: An acquisition module is used to acquire multispectral remote sensing images of the orchard's area of ​​interest; a single tree detection frame determination module, configured to perform target detection, redundancy suppression, and cluster merging on the multispectral remote sensing image to obtain a single tree detection frame; A multiple vegetation index calculation module, used to calculate multiple vegetation indices based on the multispectral remote sensing image; A global canopy mask generation module is used to generate a global canopy mask for multispectral remote sensing images based on multiple vegetation indices; The matching module is used to match the single tree detection frame and the global canopy mask using a coordinate matching algorithm to obtain a single tree instance mask.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the single tree vegetation index matching method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the single tree vegetation index matching method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the single tree vegetation index matching method according to any one of claims 1 to 6 is implemented.

Images