Lake invasion aquatic plant real-time identification and positioning method based on edge calculation
By using edge computing and image processing technologies, the problem of identifying and locating invasive aquatic plants in lake environments has been solved, enabling real-time and accurate plant monitoring and location, and improving identification efficiency and control accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-03-13
Smart Images
Figure CN121661411A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of species invasion identification technology, and in particular relates to a method for real-time identification and localization of invasive aquatic plants in lakes based on edge computing. Background Technology
[0002] Lake ecosystems are a crucial component in maintaining regional ecological balance, and their health directly impacts water quality, biodiversity, and surrounding human activities. However, the spread of invasive aquatic plants poses a serious threat to lake ecosystems, such as disrupting native species structures, blocking waterways, and affecting water circulation. These plants, due to their rapid reproduction and strong adaptability, often spread on a large scale within a short period, posing a significant challenge to lake management. Therefore, developing real-time, precise identification and location technologies for invasive aquatic plants has become an urgent need to ensure the ecological security of lakes.
[0003] Current methods for detecting invasive aquatic plants mainly rely on manual patrols or drone aerial photography combined with post-processing image analysis, but these methods have significant drawbacks. Manual patrols are inefficient, unable to cover large areas of lakes, and are prone to missed or incorrect detections due to limitations imposed by the observer's experience. While drone aerial photography can acquire large-area images, it is limited by data processing latency and interference from complex environments, making real-time analysis difficult. Furthermore, in complex lake environments, existing technologies often suffer from insufficient identification accuracy due to variations in water surface lighting or the diversity of plant morphology, especially at different distances and angles, making it difficult to accurately determine the distribution range and coverage area of plants.
[0004] The core technical challenge in identifying and locating invasive aquatic plants in lake environments lies in achieving accurate image analysis and spatial localization under dynamic water conditions. First, surface undulations and light refraction in lakes can distort image features; for example, the shape and color of plant leaves may differ significantly under different lighting conditions, affecting the stability of feature extraction. Second, this instability in feature extraction further exacerbates the difficulty of spatial localization, as pixel coordinates in images are difficult to accurately convert to actual geographic coordinates at different shooting distances and angles. For instance, close-up shots may magnify the coverage area of plant communities, while long-distance shots may lose details due to insufficient resolution, making it impossible for management departments to accurately determine the true distribution range of invasive plants. Summary of the Invention
[0005] This invention proposes a real-time identification and localization method for invasive aquatic plants in lakes based on edge computing, in order to solve the problems existing in the prior art.
[0006] To achieve the above objectives, this invention provides a method for real-time identification and localization of invasive aquatic plants in lakes based on edge computing, comprising the following steps: The morphological features, growth density, and distribution range parameters of suspected invasive aquatic plants in the images collected by the monitoring equipment are extracted, and the data are matched with historical invasion data through a convolutional neural network algorithm to obtain a preliminary invasion confidence score. Based on the preliminary invasion confidence score, the scores are compared and verified with the corresponding plants in the local species database to determine the comprehensive confidence score. If the overall confidence score exceeds the preset threshold, the proportional calibration algorithm of the edge computing node is activated to establish a spatial coordinate system and obtain the transformation matrix from pixel coordinates to geographic coordinates. Based on the transformation matrix, the accuracy of the corrected scale is determined by incorporating water surface ripple and light refraction compensation factors into the dynamic correction model, thereby obtaining an accurate value of the plant coverage area. Based on the plant cover area values, a threat level distribution map was determined; By using the threat level distribution map, a subset of growth density data for high-threat areas is obtained, and a precise location report of invasive aquatic plants is generated. Based on the precise location report, relevant alarm signals are extracted from the early warning mechanism, and a decision tree algorithm is used to assess the overall lake intrusion risk level to obtain the final threat level determination result.
[0007] Optionally, obtaining the preliminary intrusion confidence score includes: Image data is collected from monitoring equipment, and morphological features, growth density, and distribution range parameters are extracted to obtain the first parameter set; A convolutional neural network algorithm is used to process the first parameter set and historical data, analyze the degree of matching, and obtain a matching score; By comparing the matching score with a preset threshold, if the matching score is higher than the preset threshold, it is judged as a suspected intrusion and an intrusion identifier is obtained. Based on the combination of invasion markers and growth density, the coverage ratio of the distribution range is calculated, and the coverage index is determined. The correlation between coverage indicators and morphological features is obtained, and the classification result is obtained by using the support vector machine algorithm. Based on the correspondence between the classification results and historical data, the confidence score is determined to obtain the initial intrusion confidence score.
[0008] Optionally, determining the comprehensive confidence score includes: By querying the local species database using the initial invasion confidence score, standard data on the corresponding plant leaf shape are extracted to obtain the shape matching index; Based on the shape matching index and the color feature standard data, a support vector machine model is used for classification to determine the color consistency score; If the color consistency score is higher than the preset threshold, the texture similarity level is determined by comparing it with the standard texture pattern data. The texture similarity level and preliminary intrusion confidence score were obtained, and the K-means clustering algorithm was used to group the data and obtain the cluster center values. Based on the integrated verification results of cluster center values, a weighted average is calculated to determine the intermediate comprehensive index; Key components are extracted from the intermediate comprehensive indicators, and a linear regression model is used for prediction to obtain the final comprehensive confidence score.
[0009] Optionally, obtaining the transformation matrix from pixel coordinates to geographic coordinates includes: If the confidence score exceeds the preset threshold, an early warning mechanism is triggered, and an activation signal is received. Based on the activation signal, a calibration algorithm is initiated at the edge node to determine the position of the reference object; By using the positions of reference points, a spatial coordinate system is constructed, and a coordinate frame is obtained; A coordinate framework is used to map pixel coordinates, and the mapping result is judged. If the mapping results are consistent, obtain the transformation matrix, determine the matrix parameters, and then determine the transformation matrix based on the matrix parameters.
[0010] Optionally, obtaining the accurate plant cover area value includes: By applying a dynamic correction model to image data under shooting distance and angle using a transformation matrix, and incorporating a water surface ripple compensation factor, a preliminary corrected image is determined. Based on the initial corrected image incorporating a light refraction compensation factor, a convolutional neural network is used to process the refraction deviation, resulting in a refraction-compensated image; If the scale deviation in the refraction-compensated image exceeds a preset threshold, the angle parameter is adjusted by the transformation matrix, and the accuracy of the adjusted scale is judged. The plant region pixels are obtained based on the adjusted scale accuracy, and the coverage boundary is segmented using the k-means clustering algorithm to determine the plant coverage pixel set. The total coverage area is calculated by combining the set of plant cover pixels with the corrected scale, and the plant cover area value is obtained. If the deviation between the plant coverage area value and historical data is greater than a preset threshold, the support vector machine algorithm is used to verify the impact of the fluctuation and obtain the final verified area.
[0011] Optionally, determining the threat level distribution map includes: Plant cover area was extracted using spectral analysis methods to generate a cover area dataset. By using a coordinate system mapping method, the coverage area dataset is extended to the entire lake area to generate a full coverage distribution map; The K-means clustering algorithm was used to group the community location coordinates in the full-coverage distribution map to obtain a set of location points of suspected invasive communities. If the density of the location point set is greater than a preset threshold, it is judged as a high threat level; If the density is less than or equal to the preset threshold, it is determined to be a low threat level, and a threat level assessment result is generated. Based on the threat level assessment results and combined with spatial distribution characteristics, a threat level distribution map of the lake area is generated; By overlay analysis, the threat level distribution map is matched with the lake area to determine the boundary of the high-threat area; Spatial interpolation is used to smooth the data within the boundary range to obtain the final threat level distribution map.
[0012] Optionally, the generation of a precise location report for invasive aquatic plants includes: Growth density subsets are extracted from the high-threat area delineation results, and cluster analysis is used to group the subsets to obtain density distribution characteristics; The density distribution characteristics are obtained and compared with the pre-stored morphological feature data. The Pearson correlation coefficient method is used to obtain the comparison consistency value. If the consistency value is higher than the preset threshold, the high-threat area is segmented by a convolutional neural network to obtain the preliminary location of the invasive aquatic plants. By comparing the initial location points with the morphological feature data in a second step, and using the template matching method to calculate the matching degree, accurate location information is obtained. Spatial coordinate data are extracted from precise location information, and coordinate mapping is performed using a geographic information system to obtain the geographical distribution of invasive aquatic plants. Location information records are generated based on geographical distribution, and these records are stored using database storage technology to obtain the final location dataset.
[0013] Optionally, obtaining the final threat level determination result includes: Alarm signals are obtained from location reports, and similar signals are grouped using clustering algorithms to identify signal clusters; Intrusion risk indicators are extracted from the signal cluster. If the intrusion risk indicators exceed a preset threshold, the high-risk cluster is marked, and the marked cluster is obtained. For labeled clusters, a risk model is constructed using a decision tree algorithm to determine the risk level; By combining the risk level with the alarm signal strength, a comprehensive risk score is obtained; If the overall risk score is higher than the threshold, the threat level is determined to be high, and a threat level is obtained. Based on the threat level, the system correlates lake intrusion data to generate a judgment result.
[0014] Compared with the prior art, the present invention has the following advantages and technical effects: This invention discloses an intelligent monitoring and precise location method for invasive aquatic plants in lakes. It extracts morphological features, growth density, and distribution range of suspected invasive plants from image data collected by monitoring equipment. A convolutional neural network algorithm is used to analyze the matching degree with historical data, generating a preliminary invasion confidence score. The method then compares leaf shape, color, and texture features with a local species database to determine a comprehensive confidence score. If the score exceeds a threshold, the invention triggers an early warning mechanism. A proportional calibration algorithm is activated at the edge computing node, establishing a spatial coordinate system using known-sized reference objects, generating a pixel-to-geographic coordinate transformation matrix, and dynamically correcting the image data by incorporating water surface fluctuation and light refraction compensation factors to calculate the precise plant coverage area. Based on the coverage area and distribution range, the invention generates a threat level distribution map using a clustering algorithm, performs a secondary comparison of the growth density and morphological features of high-threat areas to generate a precise location report, and finally assesses the overall invasion risk of the lake using a decision tree algorithm, outputting the threat level determination result. This invention achieves automated monitoring from image analysis to precise location, significantly improving the identification efficiency and control accuracy of invasive aquatic plants, and providing technical support for lake ecological protection. Attached Figure Description
[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation
[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0018] Example 1 like Figure 1 As shown, this embodiment provides a method for real-time identification and localization of invasive aquatic plants in lakes based on edge computing, including the following steps: The morphological features, growth density, and distribution range parameters of suspected invasive aquatic plants in the images collected by the monitoring equipment are extracted, and the data are matched with historical invasion data through a convolutional neural network algorithm to obtain a preliminary invasion confidence score. Based on the preliminary invasion confidence score, the corresponding plants in the local species database are compared and verified to determine the comprehensive confidence score; If the overall confidence score exceeds the preset threshold, the proportional calibration algorithm of the edge computing node is activated to establish a spatial coordinate system and obtain the transformation matrix from pixel coordinates to geographic coordinates. Based on the transformation matrix, the accuracy of the corrected scale is determined by incorporating water surface ripple and light refraction compensation factors into the dynamic correction model, and the accurate value of the plant coverage area is obtained. Based on the vegetation cover area value, a threat level distribution map was determined; By using the threat level distribution map, we can obtain a subset of growth density data for high-threat areas and generate a precise location report for invasive aquatic plants. Based on the precise location report, relevant alarm signals are extracted from the early warning mechanism, and the decision tree algorithm is used to assess the overall lake intrusion risk level to obtain the final threat level determination result.
[0019] Specifically, the following steps are included: Step S101: Extract the morphological characteristics, growth density, and distribution range parameters of suspected invasive aquatic plants from the image data collected by the monitoring equipment, and use a convolutional neural network algorithm to analyze the degree of matching between these parameters and historical invasion data to obtain a preliminary invasion confidence score.
[0020] Specifically, image data is collected from monitoring equipment, and morphological features, growth density, and distribution range parameters are extracted to obtain a first parameter set. A convolutional neural network algorithm is used to process the first parameter set and historical data, analyzing the matching degree to obtain a matching score. The matching score is compared with a preset threshold; if the matching score is higher than the preset threshold, it is judged as a suspected intrusion, and an intrusion identifier is obtained. Based on the intrusion identifier and growth density, the coverage ratio of the distribution range is calculated to determine the coverage index. The correlation between the coverage index and morphological features is obtained, and a support vector machine algorithm is used for classification to obtain the classification result. Based on the correspondence between the classification result and historical data, a confidence score is determined to obtain a preliminary intrusion confidence score.
[0021] In this embodiment, acquiring image data from monitoring equipment is the starting point for the intrusion detection system. For example, in the field of ecological monitoring, drones are used to capture images of lake areas to identify invasive alien plants. This acquisition ensures real-time data collection, helps to detect potential threats early, and thus improves the efficiency of ecological protection.
[0022] Specifically, morphological features such as leaf shape and color, growth density such as the number of plants per unit area, and distribution range such as the coverage area parameter are extracted to form the first parameter set.
[0023] For example, morphological characteristics can be quantified as a leaf length-to-width ratio of 2.5, a density of 15 plants per square meter, and a range of 500 square meters. These parameters are obtained through image processing algorithms, which support the accuracy of subsequent analysis.
[0024] In this embodiment, a convolutional neural network algorithm is used to process the first parameter set and historical data, and the matching degree is analyzed to obtain a matching score.
[0025] For example, CNNs extract deep features from images and compare the current leaf shape with historical invasive plant data. If the similarity reaches 0.85, the score is 85. This method utilizes the pattern recognition capabilities of neural networks to improve matching accuracy, avoid human error, and facilitate rapid screening of suspected invasive events.
[0026] Specifically, by comparing the matching score with a preset threshold, if the score is higher than 80, it is judged as a suspected intrusion and an intrusion identifier is obtained.
[0027] For example, in lake monitoring, if the current data score is 90 points, which exceeds the threshold, it is marked as "suspected intrusion". This helps to prioritize responses to high-risk areas and optimize resource allocation.
[0028] In this embodiment, the coverage index is determined by calculating the coverage ratio of the distribution range based on the combination of invasion markers and growth density.
[0029] For example, a density of 20 plants per square meter, a distribution area covering 30% of the monitoring area, a coverage ratio of 0.6, and an index value of 60% reflect the severity of the invasion. This calculation integrates multi-dimensional parameters to form a comprehensive assessment, which is conducive to quantifying the impact of invasion and promoting targeted interventions.
[0030] Specifically, the correlation between coverage indicators and morphological features is obtained, and the classification result is obtained by using the support vector machine algorithm.
[0031] For example, SVM classifies intrusions as "moderate" based on morphology such as leaf texture and a coverage index of 0.7. This classification utilizes machine learning's boundary hyperplane to improve classification robustness and reduce misclassification.
[0032] In this embodiment, based on the correspondence between the classification results and historical data, a preliminary intrusion confidence score is obtained by determining the confidence score.
[0033] For example, the current moderate classification matches historical similar cases with a score of 0.9 and a confidence score of 90%. Statistical correspondence analysis ensures the reliability of the results, which is beneficial for decision support and avoids overreaction.
[0034] Specifically, this confidence assessment integrates historical experience to enhance the reliability of the system. For example, in the same lake scenario, the confidence level reaches 95% after multiple verifications, supporting the formulation of ecological management strategies. The entire process forms a closed loop from data collection to confidence judgment, enhancing the efficiency and accuracy of intrusion detection.
[0035] Step S102: Based on the obtained preliminary invasion confidence score, standard data of leaf shape, color characteristics and texture pattern of the corresponding plant are obtained from the local species database, compared and verified, and the comprehensive confidence score is determined.
[0036] Specifically, the initial invasion confidence score is used to query the local species database to extract standard data on the corresponding plant leaf shapes, resulting in a shape matching index. Based on the shape matching index, standard color feature data is fused, and a support vector machine model is used for classification to determine a color consistency score. If the color consistency score is higher than a preset threshold, it is compared with standard texture pattern data to determine the texture similarity level. The texture similarity level and the initial invasion confidence score are obtained, and K-means clustering is used to group the data, obtaining cluster center values. The cluster center values are integrated and validated, and a weighted average is calculated to determine an intermediate comprehensive index. Key components are extracted from the intermediate comprehensive index, and a linear regression model is used for prediction to obtain the final comprehensive confidence score.
[0037] For example, in the field of aquatic plant invasion detection, by querying a local species database using a preliminary invasion confidence score, standard data on the corresponding plant leaf shapes can be extracted, thus obtaining a shape matching index. This method first uses a confidence score generated from historical invasion data as the query key. For example, when the score is 0.75, the database returns the geometric parameters of standard leaf shapes such as ellipses or needles. By comparing the leaf contours in the current image, a matching index of 0.85 is calculated. This helps improve the accuracy of detection because it associates the preliminary judgment with standard data, reduces false alarms, and enhances the reliability of invasion identification.
[0038] Specifically, by fusing color feature standard data with shape matching index and using support vector machine model for classification, a color consistency score can be determined.
[0039] In this embodiment, if the shape matching index is 0.85, and standard color data such as the RGB value range of green is fused, the support vector machine can perform hyperplane separation classification and output a score such as 0.90. This supports the robustness of invasion judgment from multiple aspects. For example, a small color deviation indicates that the plant may be an invasive species, thereby improving the accuracy of the overall assessment and facilitating timely intervention in aquatic ecosystems.
[0040] For example, if the color consistency score is higher than a preset threshold such as 0.80, the texture similarity level is determined by comparing it with standard texture pattern data.
[0041] Specifically, texture templates such as mesh or stripe patterns from the database are used to compare with the current image to obtain a similarity level such as 0.88. This step examines the possibility of intrusion from a texture perspective, with support from multiple aspects such as roughness and smoothness, ensuring that the judgment is unbiased, effectively avoiding errors caused by a single feature, and improving the comprehensiveness of the detection.
[0042] Specifically, texture similarity level and initial intrusion confidence score are obtained, and K-means clustering algorithm is used to group the data to obtain cluster center values.
[0043] In this embodiment, a similarity of 0.88 and a confidence of 0.75 are used as inputs. The K-means is used to form a center value of 0.82 through iterative grouping. This integrates data from a clustering perspective, supports the logical chain of multidimensional analysis, is beneficial for discovering hidden patterns, and improves the robustness of intrusion prediction.
[0044] For example, based on the integrated verification results of cluster center values, a weighted average is calculated to determine the intermediate comprehensive index.
[0045] Specifically, the central value of 0.82 is assigned a weight of 0.6 and averaged with other validation parameters such as density parameter with a weight of 0.4, resulting in an index of 0.79. This step uses a weighted mechanism to progressively scale from the core to the extended schemes, ensuring that the index reflects comprehensive validation, effectively supports the final decision, and optimizes resource allocation to prevent the spread of intrusion.
[0046] Specifically, key components are extracted from intermediate comprehensive indicators, and a linear regression model is used for prediction to obtain the final comprehensive confidence score.
[0047] In this embodiment, the shape and color components of index 0.79 are extracted, and the predicted score is 0.92 through regression fitting. This provides support from multiple aspects, such as historical matching, forming a consistent argument, improving the credibility of the confidence score, and helping to generate reliable alerts and maintain the balance of aquatic ecosystems.
[0048] Step S103: If the overall confidence score exceeds the preset threshold, an early warning mechanism is triggered. At the same time, the proportional calibration algorithm is activated at the edge computing node to establish a spatial coordinate system using known size reference objects in the image and obtain the transformation matrix from pixel coordinates to geographic coordinates.
[0049] Specifically, if the confidence score exceeds a preset threshold, an early warning mechanism is triggered, generating an activation signal. Based on the activation signal, a calibration algorithm is initiated at the edge nodes to determine the reference object's position. A spatial coordinate system is constructed using the reference object's position, obtaining a coordinate frame. The pixel coordinates are mapped using this coordinate frame, and the mapping result is evaluated. If the mapping results are consistent, the transformation matrix is obtained, and the matrix parameters are determined.
[0050] For example, in a plant invasion detection system, when the overall confidence score exceeds a preset threshold, such as 0.85, the system immediately triggers an early warning mechanism and generates an activation signal. This mechanism ensures that potential invasive species are identified in a timely manner by monitoring score changes in real time.
[0051] In this embodiment, assuming that the leaf characteristic score of a suspected alien plant reaches 0.9, the early warning mechanism will send a signal to the relevant nodes after activation, prompting further verification, thereby avoiding ecological risks caused by misjudgment.
[0052] For example, in river monitoring scenarios, if the score exceeds the threshold, an activation signal can be triggered to initiate a scan by drone equipment, enabling early detection of signs of intrusion and facilitating rapid response to ecological protection needs.
[0053] In this embodiment, a calibration algorithm is initiated at the edge node based on the activation signal to determine the position of the reference object. This involves using image processing techniques to locate and calibrate on-site reference objects such as fixed marker points.
[0054] For example, after receiving the signal, the edge node runs an algorithm to analyze the image captured by the camera and calculates the coordinates of the reference object in the frame, such as (150, 200), thereby calibrating the overall field of view. This method helps maintain detection accuracy in dynamic environments and avoids deviations caused by device shake.
[0055] For example, in greenhouse plant monitoring, calibration algorithms can lock soil markers as references. Once the location is determined, it supports subsequent coordinate construction, which helps improve the system's accurate tracking of leaf positions.
[0056] For example, a spatial coordinate system is constructed using the position of a reference point to obtain a coordinate frame. This step extends the reference coordinates into a three-dimensional frame, such as establishing an x, y, z axis system centered on the reference point.
[0057] In this embodiment, if the reference point is located at (0,0,0), the system can generate a frame that covers the entire monitoring area, which facilitates the unification of plant image data.
[0058] For example, in field intrusion detection, constructing a coordinate frame can standardize leaf images taken from different angles, which is beneficial for subsequent mapping operations and ensures the consistency of feature comparison.
[0059] In this embodiment, a coordinate frame is used to map pixel coordinates, and the mapping result is judged. This includes mapping the original pixel points to the new frame and checking whether it meets the expected consistency.
[0060] For example, assuming the original pixel coordinates are (100, 150), after mapping they become (120, 170), the system determines that the deviation from the standard frame is less than 5 pixels, and therefore considers them consistent. This mapping helps correct image distortion and improves the reliability of invasive species identification.
[0061] For example, in leaf texture analysis, if the mapping results are consistent, it confirms the accurate extraction of color features, which is beneficial to the accuracy of the overall confidence assessment.
[0062] For example, if the mapping results are consistent, the transformation matrix is obtained and the matrix parameters are determined. This involves calculating transformation parameters from the original to the frame, such as a rotation angle of 15 degrees and a scaling factor of 1.2.
[0063] In this embodiment, the system generates a matrix based on the consistent results for subsequent image transformation, ensuring that all data is processed under unified coordinates.
[0064] For example, in plant database comparisons, determining the matrix parameters allows for optimization of shape matching index fusion, which improves the calculation efficiency and accuracy of color consistency scores. Through these steps, the system forms a robust logical chain, progressively strengthening the integrity of intrusion detection from early warning triggering to matrix determination.
[0065] Step S104: Using the obtained transformation matrix, apply a dynamic correction model to the image data under shooting distance and angle, incorporate water surface ripple and light refraction compensation factors, determine the accuracy of the corrected scale, and obtain an accurate value of plant coverage area.
[0066] Specifically, a dynamic correction model is applied to image data at different shooting distances and angles using a transformation matrix, incorporating a water surface ripple compensation factor to determine the initial corrected image. Based on this initial corrected image, a light refraction compensation factor is incorporated, and a convolutional neural network is used to process refraction deviations, resulting in a refraction-compensated image. If the scale deviation in the refraction-compensated image exceeds a preset threshold, the angle parameters are adjusted using the transformation matrix to determine the accuracy of the adjusted scale. Based on the adjusted scale accuracy, the pixels of the plant region are obtained, and a k-means clustering algorithm is used to segment the coverage boundary, determining the plant coverage pixel set. The total coverage area is calculated by combining the plant coverage pixel set with the corrected scale, yielding the plant coverage area value. If the plant coverage area value deviates from historical data by a greater than a preset threshold, a support vector machine algorithm is used to verify the impact of fluctuations, obtaining the final verified area.
[0067] In this embodiment, a dynamic correction model is applied to the image data under shooting distance and angle by a transformation matrix. First, the influence of water surface fluctuations is considered. For example, when monitoring aquatic plants in a lake, fluctuations may cause image distortion. After incorporating a compensation factor, these deformations can be effectively corrected, thereby obtaining a more accurate preliminary corrected image. This helps to improve the accuracy of subsequent analysis and reduce error accumulation.
[0068] Specifically, assuming a shooting distance of 5 meters and an angle of 30 degrees, the model will adjust the pixel position according to fluctuations such as 0.5 meters in height and width to ensure that the image reflects the real geographical layout. This will bring higher spatial consistency and improve the reliability of environmental monitoring.
[0069] For example, when processing the initial corrected image, a light refraction compensation factor is incorporated, and a convolutional neural network is used to handle refraction deviation. The principle is that the network extracts features through multiple convolutions to compensate for the visual distortion caused by the bending of underwater light, thus obtaining a refraction-compensated image. This is particularly useful in the assessment of aquatic plant cover, as it can avoid misjudging plant boundaries due to refraction, thereby improving the clarity and accuracy of the image.
[0070] In this embodiment, if the image shows blurred plant edges, the network will perform deviation correction for refractive index with a refractive angle such as 1.33. The analysis process includes feature extraction and reconstruction, and the final image deviation can be reduced to 20% of the original. This not only supports accurate measurement, but also enhances the robustness of the system.
[0071] Specifically, if the scale deviation in the refraction compensation image exceeds a preset threshold such as 5%, the angle parameter is adjusted by the transformation matrix, for example, from the initial 30 degrees to 28 degrees. The accuracy of the adjusted scale is then judged. This ensures that the mapping from pixels to actual distances is more accurate, avoids area calculation deviations in plant monitoring scenarios, thereby maintaining data consistency and supporting long-term trend analysis.
[0072] For example, the pixels of the plant area are obtained according to the adjusted scale accuracy, and the k-means clustering algorithm is used to segment the coverage boundary. The principle is that the algorithm clusters according to pixel color and texture. For example, green plant pixels are grouped into one class to determine the set of plant coverage pixels. This supports the accuracy of the boundary from multiple perspectives in the monitoring of a single lake. For example, setting the cluster center to 3 can handle light and shadow variations, bring more reliable segmentation results, and is beneficial to quantifying coverage changes.
[0073] In this embodiment, the total coverage area is calculated by combining the plant cover pixel set with the corrected scale. For example, if the pixel set is 10,000 and the scale is 1 pixel equals 0.01 square meters, then the area value is 100 square meters. This step emphasizes the conversion logic from pixels to actual area, ensuring the accuracy of the calculation and providing support for ecological assessment.
[0074] Specifically, if the deviation between the plant cover area value and historical data is greater than a preset threshold, such as 10%, a support vector machine algorithm is used to verify the impact of fluctuations and obtain the final verification area. The algorithm separates normal fluctuations from abnormal changes through classification boundaries. For example, if the historical area is 90 square meters and the current area is 105 square meters, it is adjusted to 98 square meters after verification. This explains in principle how to reduce environmental noise interference. Multiple aspects, such as fluctuation compensation and historical comparison, support each other to form a consistent verification mechanism, which is beneficial for accurately tracking plant dynamics and optimizing resource management.
[0075] Step S105: Based on the obtained plant cover area value, combined with the distribution range parameter extended to the coordinate system mapping of the entire lake area, a clustering algorithm is used to group the location points of suspected invasive communities to determine the threat level distribution map.
[0076] Specifically, spectral analysis was used to extract plant cover area, generating a cover area dataset. This dataset was then extended to the entire lake area using coordinate mapping, generating a global cover distribution map. K-means clustering was employed to group the community location coordinates in the global cover distribution map, obtaining a set of locations of suspected invasive communities. If the density of the location set exceeded a preset threshold, it was classified as a high-threat level; if the density was less than or equal to the preset threshold, it was classified as a low-threat level, generating a threat level assessment result. Based on the threat level assessment result and spatial distribution characteristics, a threat level distribution map of the lake area was generated. Overlay analysis was used to match the threat level distribution map with the lake area, determining the boundary of the high-threat area. Spatial interpolation was used to smooth the data within the boundary range, yielding the final threat level distribution map.
[0077] In this embodiment, when extracting plant cover area using spectral analysis, the reflectance spectral characteristics of the lake water are first captured through multispectral images. For example, specific bands such as near-infrared and visible light are distinguished. The principle lies in the unique absorption and reflection of the spectrum by plant chlorophyll, thereby identifying the cover area. This method can effectively avoid water surface interference, improve extraction accuracy, and benefit subsequent ecological monitoring.
[0078] Specifically, when processing historical image data, if the lake area is 500 hectares, plant pixels with a coverage of 30% can be extracted from the sample area to form a dataset, which helps to quantify the spread trend of invasive plants.
[0079] For example, when extending the coverage dataset to the entire lake area using coordinate system mapping methods, geographic coordinate transformation is used to project local data onto the global grid, such as mapping from the UTM coordinate system to the lake boundary, generating a coverage distribution map. This extension can integrate data after water surface fluctuation compensation, ensuring global consistency and facilitating the discovery of hidden intrusion hotspots.
[0080] In this embodiment, if the dataset covers an initial 10-hectare area, it is extended to the entire lake through interpolation mapping. The distribution map shows that the coverage density is higher in the north than in the south, which corresponds to the historical light refraction correction and supports a comprehensive evaluation.
[0081] Specifically, when using the K-means clustering algorithm to group the community location coordinates in the global coverage distribution map, the algorithm clusters the coordinate points into several groups based on Euclidean distance. For example, by setting K to a value of 5, a set of points suspected of being invasive communities is obtained. This grouping can reveal community patterns and is beneficial for early intervention with invasive plants.
[0082] For example, when coordinate points are densely clustered in the eastern part of the lake, three main sets are formed after clustering, which is consistent with the historical scale accuracy judgment and provides location support.
[0083] In this embodiment, when determining the density of location point sets, if the preset threshold is 10 points per square kilometer, a density exceeding this threshold is considered a high threat level; otherwise, it is considered low, and an assessment result is generated. This threshold-based judgment can quantify risk, which is beneficial for resource allocation.
[0084] Specifically, in a set, if the number of points is 15, the area is 1 square kilometer, and the density is 15, exceeding the threshold, it is judged as a high threat. This complements the historical fluctuation compensation verification, forming a consistent risk view.
[0085] For example, when generating a threat level distribution map of a lake area based on threat level assessment results and spatial distribution characteristics, high-threat points are overlaid with features of adjacent water areas, such as incorporating depth and flow velocity data, to create a color gradient map. This generation method visualizes the distribution and is beneficial for management decision-making.
[0086] In this embodiment, the high-threat area is shown in red, covering 20% of the central part of the lake, which expands the utility of historical coverage area calculation.
[0087] Specifically, when overlaying and analyzing the threat level distribution map with the lake area, GIS tools are used to overlay boundary layers to determine the boundaries of high-threat areas, for example, with meter-level accuracy, delineating irregular polygons. This matching can exclude irrelevant areas, which is beneficial for targeted governance.
[0088] For example, if the total area of a lake is 1000 hectares, the matched high-threat boundary surrounds 50 hectares, which is supported by the historical k-means dividing boundary.
[0089] In this embodiment, when smoothing boundary range data using spatial interpolation, Kriging interpolation is used to fill in blank points, for example, by performing a weighted average of density data to obtain a smoothed distribution map. This smoothing reduces noise and is beneficial for accurately predicting intrusion spread.
[0090] Specifically, after interpolation at the boundary, the threat level gradually becomes uniform from the edge, which enriches the diversity of historical support vector machine verification and forms a complete ecological solution.
[0091] Step S106: Obtain the growth density subset data of high-threat areas through the determined threat level distribution map, and perform a secondary comparison with morphological features. If the secondary comparison results show that the consistency is higher than the threshold, a precise location report of invasive aquatic plants is generated.
[0092] A subset of growth density is extracted from the high-threat area segmentation results. Cluster analysis is used to group this subset, yielding density distribution characteristics. These density distribution characteristics are then compared with pre-stored morphological feature data using the Pearson correlation coefficient method to obtain a consistency value. If the consistency value exceeds a preset threshold, a convolutional neural network is used to segment the high-threat area, obtaining preliminary location points for the invasive aquatic plants. A secondary comparison is performed between these preliminary location points and the morphological feature data, using template matching to calculate the matching degree and obtain precise location information. Spatial coordinate data is extracted from the precise location information and mapped using a geographic information system (GIS) to obtain the geographical distribution of the invasive aquatic plants. Location information records are generated based on the geographical distribution and stored using database technology to obtain the final location dataset.
[0093] For example, when extracting a subset of growth density from the results of high-threat area delineation in lakes, one can first filter pixels with density values higher than the average level from the generated threat level distribution map to form a subset of data. This extraction helps to focus on potential intrusion hotspots, avoids wasting resources on full-area computation, and thus improves monitoring efficiency.
[0094] Specifically, assuming the total coverage area of the lake region is 500 square kilometers, with 20% being high-threat areas, the extracted subset may contain point sets with density values above 0.6, which can provide an accurate basis for subsequent analysis.
[0095] In this embodiment, cluster analysis is used to group subsets. For example, the K-means algorithm is used to divide density points into three clusters, each cluster representing a different growth density, such as cluster 1 with a density of 0.7-0.9 and cluster 2 with a density of 0.5-0.7. This helps to reveal the distribution patterns of invasive plants and improve the accuracy of threat assessment.
[0096] For example, after obtaining density distribution characteristics, when comparing them with pre-stored morphological characteristic data, the Pearson correlation coefficient method is used to calculate consistency. For example, the density characteristics show a leaf density of 0.8, while the pre-stored data is 0.75, and the coefficient value is 0.9. This indicates a high correlation, which can identify the spread risk of invasive species such as water hyacinth at an early stage.
[0097] Specifically, if the consistency value is higher than the preset threshold of 0.85, the high-threat area is segmented by a convolutional neural network. For example, the network processes satellite images and outputs preliminary positioning points such as coordinates (120.5, 30.2). This can automatically separate plant outlines, improve positioning accuracy and reduce manual intervention.
[0098] In this embodiment, a second comparison is performed between the initial positioning point and morphological feature data, and the matching degree is calculated using a template matching method. For example, if the matching degree reaches 95%, accurate positioning information is obtained, which enhances the reliability of identification and avoids misjudging native plants.
[0099] For example, spatial coordinate data can be extracted from precise location information, and coordinate mapping can be performed using a geographic information system, such as converting pixel coordinates into latitude and longitude (120.6, 30.3), to obtain the geographical distribution of invasive aquatic plants. This facilitates the generation of visual maps and supports ecological management decisions.
[0100] Specifically, location information records are generated based on geographical distribution, and database storage technology is used to save the records. For example, SQL database is used to store point set data to obtain the final location dataset. This ensures data persistence and queryability, which is conducive to long-term monitoring of intrusion dynamics and the formulation of prevention and control strategies.
[0101] Step S107: Based on the generated precise location report, relevant alarm signals are extracted from the early warning mechanism, and the overall lake intrusion risk level is assessed using a decision tree algorithm to obtain the final threat level determination result.
[0102] Specifically, alarm signals are obtained from location reports, and cleaned signals are obtained through data cleaning. Based on the cleaned signals, clustering algorithms are used to group similar signals and determine signal clusters. Intrusion risk indicators are extracted from the signal clusters. If the intrusion risk indicators exceed a preset threshold, high-risk clusters are marked, resulting in marked clusters. For the marked clusters, a decision tree algorithm is used to construct a risk model to determine the risk level. Based on the risk level, the alarm signal strength is fused to obtain a comprehensive risk score. If the comprehensive risk score is higher than a threshold, the threat level is determined to be high, resulting in a threat level. Based on the threat level, lake intrusion data is correlated to generate a determination result.
[0103] In this embodiment, the process of obtaining alarm signals from the location report involves reading a previously generated dataset of precisely located invasive aquatic plants. These signals typically include indicative data of abnormal density distribution or morphological deviations. A cleansing signal is then obtained through data cleaning.
[0104] For example, median filtering can be used to remove noise interference and ensure signal accuracy. This cleaning method can effectively eliminate false positives caused by environmental factors such as water flow fluctuations, thereby improving the reliability of subsequent analysis and facilitating the early identification of ecological threats to lakes.
[0105] In this embodiment, similar signals are grouped using a clustering algorithm based on the purification signals.
[0106] For example, K-means clustering can be used to classify signals by density value and location coordinates to identify signal clusters. Assuming that the density value in the cleansing signal ranges from 0.5 to 2.0 square meters per plant, the algorithm can cluster similar density signals into three clusters. This helps to reveal the aggregation pattern of invasive plants, avoid misjudgments caused by scattered signals, and improve the accuracy of monitoring.
[0107] In this embodiment, intrusion risk indicators are extracted from the signal cluster. If the indicator exceeds a preset threshold, such as 1.5, the cluster is marked as high-risk.
[0108] For example, risk indicators are calculated based on the plant growth rate within a cluster; if the rate increases by 20% per month, the cluster is flagged. This method uses threshold screening to highlight potentially hazardous areas, which is beneficial for optimizing resource allocation and preventing the invasion from spreading throughout the entire lake.
[0109] In this embodiment, a decision tree algorithm is used to construct a risk model for the labeled cluster to determine the risk level.
[0110] For example, decision trees use density, morphological consistency, and location as nodes to classify risks from low to high. This can simulate the intrusion evolution path, provide predictive insights, and help in developing prevention strategies.
[0111] In this embodiment, a comprehensive risk score is obtained by fusing the risk level with the alarm signal strength.
[0112] For example, a weighted average of the decision tree output level and the signal strength at an 80% confidence level is calculated; a score of 0.75 indicates significant risk. This fusion enhances the comprehensiveness of the assessment and helps to integrate multi-source data to reduce bias.
[0113] In this embodiment, if the overall risk score is higher than a threshold such as 0.8, the threat level is determined to be high.
[0114] For example, in lake monitoring, high scores correspond to densely intruded areas, which triggers an alarm mechanism and facilitates timely intervention to maintain ecological balance.
[0115] In this embodiment, a judgment result is generated based on the threat level associated with lake intrusion data.
[0116] For example, matching high-threat levels with historical intrusion records outputs a report including location coordinates and recommended removal options. This correlation forms a closed-loop feedback loop, which is beneficial for long-term tracking of intrusion dynamics and improves overall management efficiency. Through these steps, from signal acquisition to final judgment, the logical chain is rigorous, ensuring the systematic nature and effectiveness of lake aquatic plant invasion monitoring. The implementation of each technical topic supports the accuracy and practicality of risk assessment.
[0117] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for real-time identification and localization of invasive aquatic plants in lakes based on edge computing, characterized in that, Includes the following steps: The morphological features, growth density, and distribution range parameters of suspected invasive aquatic plants in the images collected by the monitoring equipment are extracted, and the data are matched with historical invasion data through a convolutional neural network algorithm to obtain a preliminary invasion confidence score. Based on the preliminary invasion confidence score, the scores are compared and verified with the corresponding plants in the local species database to determine the comprehensive confidence score. If the overall confidence score exceeds the preset threshold, the proportional calibration algorithm of the edge computing node is activated to establish a spatial coordinate system and obtain the transformation matrix from pixel coordinates to geographic coordinates. Based on the transformation matrix, the accuracy of the corrected scale is determined by incorporating water surface ripple and light refraction compensation factors into the dynamic correction model, thereby obtaining an accurate value of the plant coverage area. Based on the plant cover area values, a threat level distribution map was determined; By using the threat level distribution map, a subset of growth density data for high-threat areas is obtained, and a precise location report of invasive aquatic plants is generated. Based on the precise location report, relevant alarm signals are extracted from the early warning mechanism, and a decision tree algorithm is used to assess the overall lake intrusion risk level to obtain the final threat level determination result.
2. The method according to claim 1, characterized in that, The preliminary intrusion confidence score obtained includes: Image data is collected from monitoring equipment, and morphological features, growth density, and distribution range parameters are extracted to obtain the first parameter set; A convolutional neural network algorithm is used to process the first parameter set and historical data, analyze the degree of matching, and obtain a matching score; By comparing the matching score with a preset threshold, if the matching score is higher than the preset threshold, it is judged as a suspected intrusion and an intrusion identifier is obtained. Based on the combination of invasion markers and growth density, the coverage ratio of the distribution range is calculated, and the coverage index is determined. The correlation between coverage indicators and morphological features is obtained, and the classification result is obtained by using the support vector machine algorithm. Based on the correspondence between the classification results and historical data, the confidence score is determined to obtain the initial intrusion confidence score.
3. The method according to claim 1, characterized in that, The determination of the overall confidence score includes: By querying the local species database using the initial invasion confidence score, standard data on the corresponding plant leaf shape are extracted to obtain the shape matching index; Based on the shape matching index and the color feature standard data, a support vector machine model is used for classification to determine the color consistency score; If the color consistency score is higher than the preset threshold, the texture similarity level is determined by comparing it with the standard texture pattern data. The texture similarity level and preliminary intrusion confidence score were obtained, and the K-means clustering algorithm was used to group the data and obtain the cluster center values. Based on the integrated verification results of cluster center values, a weighted average is calculated to determine the intermediate comprehensive index; Key components are extracted from the intermediate comprehensive indicators, and a linear regression model is used for prediction to obtain the final comprehensive confidence score.
4. The method according to claim 1, characterized in that, The process of obtaining the transformation matrix from pixel coordinates to geographic coordinates includes: If the confidence score exceeds the preset threshold, an early warning mechanism is triggered, and an activation signal is received. Based on the activation signal, a calibration algorithm is initiated at the edge node to determine the position of the reference object; By using the positions of reference points, a spatial coordinate system is constructed, and a coordinate frame is obtained; A coordinate framework is used to map pixel coordinates, and the mapping result is judged. If the mapping results are consistent, obtain the transformation matrix, determine the matrix parameters, and then determine the transformation matrix based on the matrix parameters.
5. The method according to claim 1, characterized in that, The accurate plant cover area value obtained includes: By applying a dynamic correction model to image data under shooting distance and angle using a transformation matrix, and incorporating a water surface ripple compensation factor, a preliminary corrected image is determined. Based on the initial corrected image incorporating a light refraction compensation factor, a convolutional neural network is used to process the refraction deviation, resulting in a refraction-compensated image; If the scale deviation in the refraction-compensated image exceeds a preset threshold, the angle parameter is adjusted by the transformation matrix, and the accuracy of the adjusted scale is judged. The plant region pixels are obtained based on the adjusted scale accuracy, and the coverage boundary is segmented using the k-means clustering algorithm to determine the plant coverage pixel set. The total coverage area is calculated by combining the set of plant cover pixels with the corrected scale, and the plant cover area value is obtained. If the deviation between the plant coverage area value and historical data is greater than a preset threshold, the support vector machine algorithm is used to verify the impact of the fluctuation and obtain the final verified area.
6. The method according to claim 1, characterized in that, The determination of the threat level distribution map includes: Plant cover area was extracted using spectral analysis methods to generate a cover area dataset. By using a coordinate system mapping method, the coverage area dataset is extended to the entire lake area to generate a full coverage distribution map; The K-means clustering algorithm was used to group the community location coordinates in the full-coverage distribution map to obtain a set of location points of suspected invasive communities. If the density of the location point set is greater than a preset threshold, it is judged as a high threat level; If the density is less than or equal to the preset threshold, it is determined to be a low threat level, and a threat level assessment result is generated. Based on the threat level assessment results and combined with spatial distribution characteristics, a threat level distribution map of the lake area is generated; By overlay analysis, the threat level distribution map is matched with the lake area to determine the boundary of the high-threat area; Spatial interpolation is used to smooth the data within the boundary range to obtain the final threat level distribution map.
7. The method according to claim 1, characterized in that, The generated precise location report of the invasive aquatic plants includes: Growth density subsets are extracted from the high-threat area delineation results, and cluster analysis is used to group the subsets to obtain density distribution characteristics; The density distribution characteristics are obtained and compared with the pre-stored morphological feature data. The Pearson correlation coefficient method is used to obtain the comparison consistency value. If the consistency value is higher than the preset threshold, the high-threat area is segmented by a convolutional neural network to obtain the preliminary location of the invasive aquatic plants. By comparing the initial location points with the morphological feature data in a second step, and using the template matching method to calculate the matching degree, accurate location information is obtained. Spatial coordinate data are extracted from precise location information, and coordinate mapping is performed using a geographic information system to obtain the geographical distribution of invasive aquatic plants. Location information records are generated based on geographical distribution, and these records are stored using database storage technology to obtain the final location dataset.
8. The method according to claim 1, characterized in that, The final threat level determination result includes: Alarm signals are obtained from location reports, and similar signals are grouped using clustering algorithms to identify signal clusters; Intrusion risk indicators are extracted from the signal cluster. If the intrusion risk indicators exceed a preset threshold, the high-risk cluster is marked, and the marked cluster is obtained. For labeled clusters, a risk model is constructed using a decision tree algorithm to determine the risk level; By combining the risk level with the alarm signal strength, a comprehensive risk score is obtained; If the overall risk score is higher than the threshold, the threat level is determined to be high, and a threat level is obtained. Based on the threat level, the system correlates lake intrusion data to generate a judgment result.