Mangrove forest damage monitoring method and system based on mobile close-range photogrammetry

Abstract

The invention discloses a mangrove forest damage monitoring method and system based on mobile close-range photogrammetry, and the method comprises the steps: determining a target distance through testing a relation value between an image element of a field mobile phone and a distance by means of image recognition; ground control points are set through mobile phone positioning, a coordinate reference system is established based on branch wires, and a high-precision coordinate relation can be established to improve the geographic position of target recognition; a plurality of images of mangrove plants are obtained through the field mobile phone, a seamless synthetic image is output based on a preset mangrove plant image identification AI model, the problem that the view field of the mobile phone is small is solved, and the method is more suitable for large-range monitoring requirements; based on a preset mangrove plant image identification AI model, processing the seamless synthetic image to obtain mangrove plant state information, and identifying the damaged state of the mangrove plant; and obtaining mangrove plant state monitoring information in combination with the coordinate reference system and the plant state information, so that the state of the mangrove plant at a specified position can be accurately identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental protection, and particularly relates to a mangrove damage monitoring method and system based on mobile close-range photogrammetry. BACKGROUND

[0002] Human activities mostly have certain impacts on the environment, and the environment also has impacts on human activities. In recent years, water and soil loss, air pollution and water pollution caused by environmental damage have seriously affected the normal life of human beings, and therefore the importance of environmental protection has been increasingly valued.

[0003] Mangroves grow in the tidal flat zone of estuary harbors, and the soil is soft and inconvenient to walk on, so it is difficult to conduct field investigation, and the monitoring is difficult and costly. At present, large-scale port engineering construction, such as cross-sea bridge and land canal construction, may cause damage to mangrove plants around the construction site. The existence of mangrove plants can keep the soil stable and purify water and air, and therefore, as an important part of the construction site environmental protection, it is necessary to ensure that the construction does not have a bad impact on the survival of the plants, which is also a legal requirement for construction operations. Under this background, how to dynamically monitor the state of mangrove plants at low cost and high efficiency, and timely discover damaged mangrove plants in order to timely protect and repair them is a problem to be solved. SUMMARY

[0004] The present application proposes a mangrove damage monitoring method and system based on mobile close-range photogrammetry to solve the problems of insufficient funds, high cost, low precision and insufficient safety of the prior art.

[0005] To achieve the above-mentioned purposes, the technical solution adopted by the present application is:

[0006] The mangrove damage monitoring method based on mobile close-range photogrammetry comprises the following steps: testing the relationship value between the image elements and the distance of the on-site mobile phone; setting the ground control point using the positioning function of the mobile phone, and establishing a coordinate reference system based on the branch guide line; obtaining multiple images of mangrove plants through the on-site mobile phone, and outputting seamless composite images based on a pre-set mangrove plant image recognition AI model; processing the seamless composite images based on a pre-set image recognition model to obtain plant state information; and combining the coordinate reference system and the plant state information to obtain mangrove plant state monitoring information.

[0007] Further, the relationship value of the image element of the test field mobile phone and the distance comprises: a first conversion relationship of pixels and field distances under different field distance conditions by shooting and testing a target object; a second conversion relationship of image elements and distances corresponding to different geometric parameters under the same field distance condition by shooting and testing the target object; the geometric parameters include height, width, perimeter and area; and the relationship value is obtained based on the first conversion relationship and / or the second conversion relationship.

[0008] Further, the multiple images of the mangrove plants are obtained by the field mobile phone, and a seamless synthesized image is output based on a preset mangrove plant image recognition AI model, which comprises: identifying the multiple images to obtain feature homonymic points in each image, and performing multi-image synthesis based on the feature homonymic points to obtain the seamless synthesized image.

[0009] Further, the mangrove plants are mangrove forests, and the damaged forms of the mangrove forests include trunk fracture, crown withering, root system exposure and lodging; a main network of the image recognition model adopts a ResNet-34+FPN structure to balance the calculation efficiency and the feature extraction capability; the image recognition model is initialized using COCO pre-training weights for transfer learning; the size of the input image of the image recognition model is adjusted to 1024x1024 pixels, the batch size is set to 8, the learning rate is initially set to 1e-4, the learning rate is attenuated by 50% every 10 epochs, and the model accuracy is evaluated using mAP and IoU threshold.

[0010] Further, the mangrove plant state monitoring information is obtained by combining the coordinate reference system and the plant state information through the GIS software set on the field mobile phone.

[0011] The mangrove forest damage monitoring system based on mobile close-range photogrammetry comprises: a first module for testing the relationship value of the image element of the field mobile phone and the distance; a second module for setting ground control points using the positioning function of the mobile phone and establishing a coordinate reference system based on a branch guide line; a third module for obtaining multiple images of mangrove plants by the field mobile phone and outputting a seamless synthesized image based on a preset mangrove plant image recognition AI model; a fourth module for processing the seamless synthesized image based on a preset image recognition model to obtain plant state information; and a fifth module for obtaining mangrove plant state monitoring information by combining the coordinate reference system and the plant state information.

[0012] Further, the relationship value of the image element and the distance of the test field mobile phone comprises: shooting and testing the first conversion relationship of the pixel and the field distance of the target object under different field distance conditions; shooting and testing the second conversion relationship of the image element and the distance corresponding to different geometric parameters under the same field distance condition; the geometric parameters comprise height, width, perimeter and area; and obtaining the relationship value based on the first conversion relationship and / or the second conversion relationship.

[0013] Further, the multiple images of the mangrove plants are acquired by the field mobile phone, and a seamless synthesis image is output based on a preset mangrove plant image recognition AI model, which comprises: identifying the multiple images to obtain feature homonymic points in each image, and performing multi-image synthesis based on the feature homonymic points to obtain the seamless synthesis image.

[0014] Further, the mangrove plants are mangrove forests, the damaged forms of the mangrove forests comprise trunk fracture, crown withering, root system exposure and lodging, the main network of the image recognition model adopts a ResNet-34+FPN structure to balance the calculation efficiency and the feature extraction capability, the image recognition model is initialized by using COCO pre-training weights for migration learning, the size of the input image of the image recognition model is adjusted to 1024*1024 pixels, the batch size is set to 8, the learning rate is initially set to 1e-4, the learning rate is attenuated by 50% every 10 epochs, and the model accuracy is evaluated by using mAP and IoU threshold.

[0015] Further, the mangrove plant state monitoring information is obtained by combining the coordinate reference system and the plant state information through the GIS software arranged on the field mobile phone.

[0016] By adopting the technical scheme, the present application has the following beneficial effects:

[0017] 1. The relationship value of the image element and the distance of the test field mobile phone can be used to determine the distance of the target by image recognition; the positioning function of the mobile phone is used to set the ground control point, and the coordinate reference system is established based on the branch line, so that the further coordinate relationship is established to improve the accuracy of target recognition; the multiple images of the mangrove plants are acquired by the field mobile phone, and a seamless synthesis image is output based on a preset mangrove plant image recognition AI model, which overcomes the problem of small field of view of the mobile phone and is more suitable for large-scale monitoring requirements; the plant state information is obtained by processing the seamless synthesis image based on the preset image recognition model, so that the state of the plant can be identified; and the mangrove plant state monitoring information can be obtained by combining the coordinate reference system and the plant state information, so that the state of the plant at the specified position can be accurately identified. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1This is a flowchart of the mangrove damage monitoring method based on mobile close-range photogrammetry proposed in this invention;

[0019] Figure 2 This is a schematic diagram illustrating the principle of geometric calibration of mobile devices proposed in this invention;

[0020] Figure 3 This is a schematic diagram of mangrove coordinate calculation proposed in this invention;

[0021] Figure 4 This is a monitoring image of mangrove plant damage status proposed in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 The method for monitoring mangrove damage based on mobile close-range photogrammetry includes: S1, testing the relationship between image elements and distance on a mobile phone at the testing site; S2, using the mobile phone's positioning function to set ground control points and establishing a coordinate reference system based on traverse lines; S3, acquiring multiple images of mangrove plants through the mobile phone at the testing site, and outputting a seamless composite image based on a preset mangrove plant image recognition AI model; S4, processing the seamless composite image based on the preset image recognition model to obtain plant status information; S5, combining the coordinate reference system and the plant status information to obtain mangrove plant status monitoring information.

[0024] A mobile phone's camera function, based on its own performance and corresponding image processing software, will generally have relatively stable image accuracy if there is no performance degradation or software malfunction. Therefore, by actually testing the relationship between image elements in the image and their actual distances, the correlation value (or relationship value) of the corresponding relative positions (distances) can be determined. This relationship value allows for distance measurement through calculation rather than actual measurement, thus reducing the cost of on-site distance measurement.

[0025] Now the smart phone, many with positioning function, can realize the global positioning based on self-positioning. The branch guide line is from the known point, and then the angle and distance are measured at each to-be-determined point, and a straight line is used to connect each to-be-determined point in turn, forming a free stretching polyline. This guide line form is called branch guide line. Through global positioning and branch guide line, a coordinate system centered on the mobile phone can be set up, which can further improve the accuracy of identifying target points rather than being limited to distance. Especially in the construction site, the construction site is strict in size accuracy, not only various reference materials are set in the field, but also many measuring tools, which can assist in the setting of the branch guide line.

[0026] The mobile phone can of course shoot a large angle image, but this will cause the image accuracy to decrease, causing subsequent identification to be abnormal, therefore, a small range is shot multiple times, and then seamlessly synthesized, which can fully utilize the function of the mobile phone without causing the shooting range to be too small.

[0027] Through the trained image recognition model, the corresponding result, i.e., plant state information, can be output.

[0028] By combining the specific plant state information and the corresponding coordinate parameters (distance parameters), the mangrove plant state monitoring information in the research range can be obtained.

[0029] The relationship value between the image elements of the test site mobile phone and the distance includes: shooting and testing the first conversion relationship between the pixels and the field distance of the target object under different field distance conditions; shooting and testing the second conversion relationship between the image elements and the distance corresponding to different geometric parameters under the same field distance condition; the geometric parameters include height, width, perimeter and area; based on the first conversion relationship and / or the second conversion relationship, the relationship value is obtained.

[0030] Through actual experiments, the first conversion relationship between pixels and field distance of the target object under different field distance conditions can be shot and tested, i.e., the resolution / pixel accuracy is different under different distances, basically, the closer the distance, the higher the resolution / pixel accuracy, and the corresponding matching relationship is the first conversion relationship.

[0031] Similarly, the shape of different targets will also change when the distance changes, and the corresponding change belongs to geometric change, and the geometric parameters include height, width, perimeter and area, etc., and the corresponding matching relationship is the second conversion relationship.

[0032] The first conversion relationship and / or the second conversion relationship can be selected for combination and processing to obtain the relationship value, and the specific selection can be obtained according to actual experiments.

[0033] The method comprises the following steps: acquiring multiple images of the mangrove plant by the field mobile phone, and outputting seamless synthesized images based on a preset mangrove plant image recognition AI model, which comprises the following steps: identifying the multiple images to obtain feature homonymic points in each image, and performing multi-image synthesis based on the feature homonymic points to obtain the seamless synthesized images.

[0034] Due to the short shooting time and small shooting position change, the appearance of the same target in the images will not change substantially, and thus the features of the appearance, such as the number of branches, the height of branches, the color of leaves, and the shape of branches, are preserved. By naming the feature points, the same feature in different images can be matched, and the same points in the images can be merged to obtain the seamless synthesized images.

[0035] The mangrove plant is a mangrove forest, and the damaged forms of the mangrove forest include broken trunks, withered canopies, exposed root systems, and lodging. The main network of the image recognition model adopts a ResNet-34+FPN structure to balance the calculation efficiency and feature extraction capability. The image recognition model is initialized using COCO pre-training weights for transfer learning. The size of the input image of the image recognition model is adjusted to 1024x1024 pixels, the batch size is set to 8, the initial learning rate is set to 1e-4, the learning rate is decayed by 50% every 10 epochs, and the model accuracy is evaluated using mAP and IoU thresholds.

[0036] The field GIS (geographic information system) software is used to obtain mangrove plant state monitoring information in combination with the coordinate reference system and the plant state information.

[0037] The GIS software can realize image processing and information output on a mobile phone, a tablet computer, or a notebook computer, without the need for special equipment, thereby reducing the actual measurement cost.

[0038] The mangrove forest damage monitoring system based on mobile close-range photogrammetry comprises the following modules: a first module for testing the relationship between image elements and distance of a field mobile phone; a second module for setting ground control points using the positioning function of the mobile phone and establishing a coordinate reference system based on a branch guide line; a third module for acquiring multiple images of a mangrove plant by the field mobile phone and outputting seamless synthesized images based on a preset mangrove plant image recognition AI model; a fourth module for processing the seamless synthesized images based on a preset image recognition model to obtain plant state information; and a fifth module for obtaining mangrove plant state monitoring information in combination with the coordinate reference system and the plant state information.

[0039] Testing the relationship value of the image element of the field mobile phone and the distance, including: shooting and testing the first conversion relationship of the pixel and the field distance of the target object under different field distance conditions; shooting and testing the second conversion relationship of the image element and the distance corresponding to different geometric parameters under the same field distance condition; the geometric parameters include height, width, perimeter and area; obtaining the relationship value based on the first conversion relationship and / or the second conversion relationship.

[0040] The multiple images of the mangrove plants are acquired by the field mobile phone, and a seamless synthesized image is output based on a preset mangrove plant image recognition AI model, including: identifying the multiple images to obtain feature homonymic points in each image, and performing multi-image synthesis based on the feature homonymic points to obtain the seamless synthesized image.

[0041] The mangrove plants are mangrove forests, and the damaged forms of the mangrove forests include trunk fracture, canopy withering, root system exposure and lodging; the main network of the image recognition model selects a ResNet-34+FPN structure to balance the calculation efficiency and the feature extraction capability; the image recognition model is initialized using COCO pre-training weights for transfer learning; the size of the input image of the image recognition model is adjusted to 1024x1024 pixels, the batch size is set to 8, the learning rate is initially set to 1e-4, the learning rate is attenuated by 50% every 10 epochs, and the model accuracy is evaluated using mAP and IoU threshold.

[0042] The mangrove plant state monitoring information is obtained by combining the coordinate reference system and the plant state information through the GIS software set on the field mobile phone.

[0043] The scheme takes mobile close-range photogrammetry as the core, acquires high-definition images of mangrove forests through standardized shooting, combines ground control points (GCPs) and mobile sensor parameter calibration, realizes image geometric correction and geographic positioning, synthesizes millimeter-level resolution side-view images of mangrove forests based on feature point automatic identification and image inlay technology, solves the problem of extracting structural features of mangrove forests under complex terrain, constructs a special annotation dataset for the morphological features (such as trunk fracture, canopy withering, root system exposure and lodging) of damaged mangrove forests, trains a Faster R-CNN model through transfer learning and data enhancement (lighting changes and texture changes), automatically extracts the geometric size, quantity and spatial distribution of damaged mangroves by integrating GIS spatial analysis functions, generates a heat map and a repair priority zoning, and supports multi-period image comparison to evaluate the damage dynamics.

[0044] 1. Close-range photogrammetry based on geometric calibration of mobile and portable intelligent devices

[0045] By performing geometric calibration in the laboratory, the pixel-to-real-world distance conversion relationship (scale parameter) at different shooting distances can be accurately obtained, enabling high-precision extraction of geometric parameters such as the height, width, perimeter, and area of ​​the target object, providing a reliable geometric data foundation for damage assessment.

[0046] 2. Spatialization of mobile data

[0047] Ground control points are used to correct photos taken with mobile phones; and the mobile phone is used as a survey station to transmit coordinates to the mangrove forest through traverse lines, thus solving the problem of missing spatial reference in mobile phone photos.

[0048] 3. Seamless synthesis of mangrove side view images

[0049] By using mobile portable devices to capture side-view images of mangroves, and then using feature homologous point matching, multiple images are synthesized to obtain a seamless composite image.

[0050] 4. Deep Learning-Based Target Recognition in Damaged Mangrove Forests

[0051] Using samples of damaged mangroves, deep learning labels were created to train a damaged mangrove recognition model, which was then used to identify the damaged mangroves and obtain their location and distribution range.

[0052] Close-range photogrammetry, in particular, involves capturing images of a target object within a 300-meter radius and determining its 3D coordinates based on a 2D image. By integrating close-range photogrammetry onto mobile devices, it can perform information acquisition and image processing tasks in diverse scenarios, enabling information recognition and possessing broad application value. Image recognition of images acquired by mobile devices primarily involves image processing, feature extraction, pattern recognition, and deep learning. Image recognition technology is a crucial branch of artificial intelligence, aiming to automatically identify and analyze objects, features, and scenes in images. With the continuous advancement of computer vision technology, the technological path of image recognition is constantly evolving, gradually transitioning from relying on traditional algorithms to a new stage driven by deep learning. Compared to traditional computer image recognition technology, AI-driven image recognition solutions offer significant advantages. Firstly, this technology combines high precision with high efficiency. Through data-driven model training mechanisms combined with multi-dimensional feature verification, it can significantly improve recognition efficiency and accuracy. Secondly, when handling high-load tasks, edge computing architecture enables the rational allocation of system performance resources, effectively reducing data processing latency.

[0053] like Figure 4 The image shows a monitoring diagram of the damage status of mangrove plants (exposed roots, lodging, and withering).

[0054] In response to the limitations of traditional monitoring technologies, such as the tidal flat environment, bulky equipment, and low operational efficiency, this project proposes to deeply integrate mobile close-range photogrammetry with artificial intelligence to build a lightweight and intelligent mangrove damage detection system. This system will enable efficient, accurate, and automated detection operations, and promote the transformation of ecological protection work from reliance on manual inspections to intelligent perception and decision support.

[0055] The beneficial effects of this plan include:

[0056] 1. A closed-loop process of "calibration-modeling-recognition": an integrated process from equipment geometric calibration to intelligent recognition, breaking through the problems of traditional methods and data fragmentation;

[0057] 2. Multi-source data fusion driven: Integrating GNSS positioning, image geometric correction and deep learning to achieve centimeter-level to sub-meter-level accuracy, enabling accurate and efficient analysis of mangrove side view structures;

[0058] 3. Lightweight engineering implementation: Using consumer-grade mobile phones (around 2,000 yuan) to replace professional photogrammetry equipment (around 50,000 yuan) reduces the cost of single-point data collection by 96%; based on mainstream deep learning algorithms, it significantly reduces the technical threshold and implementation cost.

[0059] like Figure 2 As shown:

[0060] 1) Geometric calibration of mobile devices

[0061] 1.1 Data Acquisition: In the laboratory calibration field, different object distances di (e.g., 1m, 2m, ..., 10m) are set along the optical axis. The mobile phone is kept vertically placed using a mobile phone stabilization platform, and a precision standard calibration plate of known size (checkerboard L = 1mm) is photographed.

[0062] 1.2 Pixel Measurement: Extracting the pixel span N of feature points on the calibration board i (For example, the number of pixels corresponding to the spacing between the corner points of a chessboard grid).

[0063] 1.3 Parameter Fitting: The parameters of the nonlinear model are solved using the least squares method. Due to lens distortion and perspective projection, the scale parameter λ(d) has a nonlinear relationship with the object distance d. A polynomial model is commonly used for fitting.

[0064] λ(d)=k1*d 2 +k2*d+k3 Formula 1

[0065] Where K1 and K2 are undetermined coefficients.

[0066] 1.4 Practical Application: For any shooting distance d, the actual size Lreal of the target object can be calculated using the following formula:

[0067] Lreal=N*λ(d) Formula 2

[0068] Where: N is the target pixel span in the image, d is the current shooting distance (obtained by laser ranging).

[0069] (2) Spatialization of mobile phone photography images

[0070] 2.1 Accurately measure the location of the mobile phone point of position using GNSS RTK (Global Navigation Satellite System Real-Time Kinematic).

[0071] 2.2 Use the mobile phone as a survey station and transfer the coordinates to the mangrove through a branch traverse.

[0072] 2.3 Use the mangrove location as a control point to correct the position of the photo.

[0073] As shown in Figure 3 , according to the branch traverse measurement mode and polar coordinate calculation method, the coordinate calculation formula of the mangrove can be obtained:

[0074] X p = X o + d cos(a) Y p = Y o + d sin(a)

[0075] Where p is the unknown point to be determined, a is the azimuth angle, and d is the horizontal distance from the mobile phone to the mangrove (obtained by laser ranging).

[0076] (3) Seamless synthesis of mangrove side view images

[0077] Use a mobile device to take side view images of the mangrove, match the same name features and points, and synthesize multiple images to obtain a seamless synthesis of the mangrove image.

[0078] (4) Target recognition of damaged mangroves based on deep learning

[0079] Use damaged mangrove samples to create deep learning labels and train a damaged mangrove recognition model to identify damaged mangroves in the study area and obtain their location and distribution range.

[0080] (5) Model training and verification

[0081] The target recognition model is trained using Faster R-CNN, and the backbone network uses the ResNet-34+FPN structure to balance the calculation efficiency and feature extraction capability. The model is initialized using COCO pre-trained weights for transfer learning. The input image size is adjusted to 1024x1024 pixels, and the batch size is set to 8.

[0082] The learning rate is initially set to 1e-4 and is attenuated by 50% every 10 epochs. The model accuracy is evaluated using mAP (mean average precision) and IoU threshold.

[0083] (6) Dynamic monitoring of damaged mangrove forests

[0084] Mobile phones are used for close-range photogrammetry of target areas, deep learning models are used to identify the geometric parameters of damaged mangroves, and GIS software is used to obtain the location and range of damaged mangrove forests.

[0085] The technical route of this project can be summarized as follows: The technical route of this project takes "equipment calibration - spatial correction - image synthesis - intelligent identification - dynamic monitoring" as the main line. A pixel-actual distance nonlinear model is constructed through laboratory multi-distance calibration, and GNSS RTK and branch coordinate transfer are combined to realize the geospatialization of mobile phone images. Feature matching algorithms are used to complete the seamless synthesis of side-view images of mangrove forests, and Faster R-CNN deep learning models are used to accurately identify damaged mangrove forest targets. Through the GIS platform, geometric parameters and spatiotemporal data are integrated to form a "data collection - processing - analysis - decision-making" full-chain closed loop, and ultimately realize low-cost, automated dynamic monitoring of mangrove forests with centimeter to submeter accuracy.

[0086] Application cases

[0087] The Ping-Lu Canal under construction in Guangxi Zhuang Autonomous Region is the first canal project in China since its founding that connects rivers to the sea, and is also the backbone project of the new land-sea channel in the west. The canal is an effective means of transportation, but the reconstruction of the canal will involve large-scale land reconstruction, which will have an adverse impact on the vegetation nearby. Mangroves are a type of plant that is easily damaged, difficult to restore, and has high ecological value. The Ping-Lu Canal is being built according to the standards of an inland I-class waterway, and in the future, the passage of 5000-ton ships will cause significant ship waves. Superimposed with the accelerated flow caused by large-scale straightening and dredging and widening of the river channel, the erosion of the riverbank will be further intensified, and the mangrove ecosystem will face greater scouring pressure, and its survival foundation will be continuously weakened. In this context, this technology is used to carry out high-frequency dynamic monitoring of the damage status of mangrove forests, identify damaged mangrove forests, and repair them in a timely manner.

[0088] The above description is a detailed description of the preferred embodiments of the present application, but the embodiments are not intended to limit the scope of the patent application of the present application. Any equivalent changes or modifications made under the technical spirit of the present application should be included in the scope of the patent.

Claims

1. A method for monitoring mangrove damage based on mobile close-range photogrammetry, characterized in that, include: The relationship between image elements of the mobile phone and distance at the test site; Use the mobile phone's location function to set ground control points and establish a coordinate reference system based on the traverse line; Multiple images of mangrove plants are acquired using the mobile phone at the site, and a seamless composite image is output based on a preset mangrove plant image recognition AI model. Based on a preset mangrove plant image recognition model, the seamlessly synthesized image is processed to obtain plant state information; By combining the coordinate reference system and the plant status information, mangrove plant status monitoring information is obtained.

2. The mangrove damage monitoring method based on mobile close-range photogrammetry according to claim 1, characterized in that, The relationship between image elements and distance on the mobile phone at the test site includes: Capture and test the first conversion relationship between pixels and real-world distance for a target object under different real-world distance conditions; Capture and test the second transformation relationship between image elements and distance for different geometric parameters of a target object under the same real-world distance conditions; The geometric parameters include height, width, perimeter, and area; The relationship value is obtained based on the first transformation relationship and / or the second transformation relationship.

3. The mangrove damage monitoring method based on mobile close-range photogrammetry according to claim 2, characterized in that, The process involves acquiring multiple images of mangrove plants via a mobile phone at the site, and outputting a seamlessly synthesized image based on a pre-set mangrove plant image recognition AI model, including: The multiple images are identified, and the corresponding feature points in each image are obtained. Based on the corresponding feature points, the multiple images are synthesized to obtain the seamless synthesized image.

4. The mangrove damage monitoring method based on mobile close-range photogrammetry according to claim 3, characterized in that, The mangrove plant is a mangrove forest, and the damaged forms of the mangrove forest include trunk breakage, canopy withering, exposed root system and lodging; The backbone network of the image recognition model adopts the ResNet-34+FPN structure to balance computational efficiency and feature extraction capability. The image recognition model is initialized using COCO pre-trained weights, and transfer learning is then performed. The input image size of the image recognition model is adjusted to 1024×1024 pixels, the batch size is set to 8, the initial learning rate is set to 1e-4, and the learning rate decreases by 50% every 10 epochs. The model accuracy is evaluated using mAP and IoU thresholds.

5. The mangrove damage monitoring method based on mobile close-range photogrammetry according to claim 4, characterized in that, By using the GIS software set on the mobile phone at the site, combined with the coordinate reference system and the plant status information, mangrove plant status monitoring information is obtained.

6. A mangrove damage monitoring system based on mobile close-range photogrammetry, characterized in that, include: The first module is used to test the relationship between image elements of the mobile phone and distance in the field. The second module is used to set ground control points using the mobile phone's positioning function and establish a coordinate reference system based on the traverse line; The third module is used to acquire multiple images of mangrove plants through the mobile phone on site, and output a seamless synthesized image based on a preset mangrove plant image recognition AI model; The fourth module is used to process the seamlessly synthesized image based on a preset image recognition model to obtain plant state information; The fifth module is used to combine the coordinate reference system and the plant status information to obtain mangrove plant status monitoring information.

7. The mangrove damage monitoring system based on mobile close-range photogrammetry according to claim 6, characterized in that, The relationship between image elements and distance on the mobile phone at the test site includes: Capture and test the first conversion relationship between pixels and real-world distance for a target object under different real-world distance conditions; Capture and test the second transformation relationship between image elements and distance for different geometric parameters of a target object under the same real-world distance conditions; The geometric parameters include height, width, perimeter, and area; The relationship value is obtained based on the first transformation relationship and / or the second transformation relationship.

8. The mangrove damage monitoring system based on mobile close-range photogrammetry according to claim 7, characterized in that, The process involves acquiring multiple images of mangrove plants via a mobile phone at the site, and outputting a seamlessly synthesized image based on a pre-set mangrove plant image recognition AI model, including: The multiple images are identified, and the corresponding feature points in each image are obtained. Based on the corresponding feature points, the multiple images are synthesized to obtain the seamless synthesized image.

9. The mangrove damage monitoring system based on mobile close-range photogrammetry according to claim 8, characterized in that, The mangrove plant is a mangrove forest, and the damaged forms of the mangrove forest include trunk breakage, canopy withering, exposed root system and lodging; The backbone network of the image recognition model adopts the ResNet-34+FPN structure to balance computational efficiency and feature extraction capability. The image recognition model is initialized using COCO pre-trained weights, and transfer learning is then performed. The input image size of the image recognition model is adjusted to 1024×1024 pixels, the batch size is set to 8, the initial learning rate is set to 1e-4, and the learning rate decreases by 50% every 10 epochs. The model accuracy is evaluated using mAP and IoU thresholds.

10. The mangrove damage monitoring system based on mobile close-range photogrammetry according to claim 9, characterized in that, By using the GIS software set on the mobile phone at the site, combined with the coordinate reference system and the plant status information, mangrove plant status monitoring information is obtained.

Images