A wetland plant recognition method and system based on multi-dimensional feature similarity comparison

Abstract

The present application relates to the technical field of mechanical vision, and discloses a wetland plant identification method and system based on multi-dimensional feature similarity comparison, which comprises the following steps: taking color features as macroscopic information and texture features as microscopic details to construct basic feature description; counting color distribution and texture direction in image data to obtain multi-scale feature representation; matching geometric positions and descriptors in the multi-scale feature representation to establish feature manifold of the multi-scale feature representation, and guiding the multi-scale feature representation based on the feature manifold to generate fusion feature representation; selecting a primary comparison feature path according to the correlation strength in the fusion feature representation to obtain a comparison strategy; comparing the fusion feature representation with a known wetland plant feature database to obtain a similarity score; and performing consistency verification on growth environment context information and phenology context information to obtain a species identification result; the present application can improve the efficiency of wetland plant identification.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to a wetland plant identification method and system based on multidimensional feature similarity comparison. Background Technology

[0002] Existing wetland plant identification technologies have significant limitations in feature extraction. They often focus only on utilizing a single type of feature, failing to effectively combine the macroscopic global information carried by color features with the microscopic detail information corresponding to texture features. This results in a lack of comprehensiveness and completeness in the constructed basic feature description. Furthermore, the preprocessing of the original image data is inadequate, failing to adequately suppress high-frequency noise interference or reasonably expand the grayscale distribution to improve image contrast. This leads to low-quality extracted features and insufficient recognizability, creating potential problems for subsequent feature matching and recognition, and directly affecting the reliability of the recognition results.

[0003] In the process of feature fusion and recognition, existing technologies have failed to adequately address the coordination issues between multi-scale features. They lack precise alignment of the geometric positions of features at different scales and effective matching of descriptors. The construction of feature manifolds lacks scientific guidance, leading to obstructed cross-scale feature information transmission, poor fusion results, and difficulty in forming a fused feature representation that combines global relevance with local detail. Furthermore, existing comparison strategies lack specificity, failing to optimize the comparison path based on the strength of the correlation between feature dimensions and neglecting the role of contextual information such as growth environment and phenological stage in verifying the recognition results. This not only results in low comparison efficiency but also easily leads to recognition bias due to feature expression conflicts or incompatibility with environment and phenology, failing to meet the needs of efficient and accurate wetland plant identification in practical applications. Therefore, improving the efficiency of wetland plant identification has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a wetland plant identification method and system based on multidimensional feature similarity comparison to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a wetland plant identification method based on multidimensional feature similarity comparison, comprising:

[0006] S1. Using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details, construct the basic feature description of the wetland plants;

[0007] S2. Based on the basic feature description, statistically analyze the color distribution and texture direction of visually salient regions in the image data to obtain the multi-scale feature representation of the wetland plants;

[0008] S3. Match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fused feature representation of the wetland plants;

[0009] S4. Based on the strength of the correlation between different feature dimensions in the fused feature identifier, select the primary comparison feature path to obtain the comparison strategy for the wetland plants.

[0010] S5. Apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant.

[0011] S6. Based on the similarity score, perform consistency verification on the growth environment context information and phenological period context information of the wetland plants to obtain the species identification results of the wetland plants.

[0012] In a preferred embodiment, the construction of the basic feature description of the wetland plants, based on the color features of the wetland plant image data as macroscopic information and supplemented by texture features as microscopic details, includes:

[0013] High-frequency noise in wetland plant image data is suppressed to obtain denoised image data of the wetland plants;

[0014] Expanding the grayscale distribution of the denoised image data yields contrast-enhanced image data of the wetland plants;

[0015] The color space is divided from the contrast-enhanced image data, and the distribution pattern of the dominant color tone in the color space is statistically analyzed.

[0016] Based on the distribution pattern, the color features of the contrast-enhanced image data are used as the basis for the macroscopic information of the wetland plants;

[0017] The texture features of the contrast-enhanced image data are used to supplement the microscopic details of the wetland plants;

[0018] The macroscopic information base is combined with the microscopic details to construct a basic characteristic description of the wetland plants.

[0019] In a preferred embodiment, the step of statistically analyzing the color distribution and texture orientation of visually salient regions in the image data based on the fundamental feature description to obtain a multi-scale feature representation of the wetland plants includes:

[0020] Based on the aforementioned basic feature description, differential image regions in the contrast-enhanced image data are detected to obtain the visually salient regions of the wetland plants;

[0021] A concentric analysis window for the wetland plants is established with the geometric center of the visually salient region as the origin.

[0022] In the concentric analysis window, tensor synthesis is performed on the main hue and distribution ratio of the contrast-enhanced image data to obtain the color distribution vector of the wetland plants.

[0023] By tracing the grayscale gradient direction of the contrast-enhanced image data, a description of the texture direction of the wetland plants is obtained;

[0024] The texture direction description is superimposed on the color distribution vector to obtain a multi-scale feature representation of the wetland plants.

[0025] In a preferred embodiment, the step of matching the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish a feature manifold of the multi-scale feature representation, and guiding the multi-scale feature representation based on the feature manifold to generate a fused feature representation of the wetland plants, includes:

[0026] Under a unified coordinate system, the geometric positions of coarse-scale features and fine-scale features in the multi-scale feature representation are aligned to obtain the inter-scale spatial correspondence of the wetland plants.

[0027] Based on the degree of similarity between descriptors of features at different scales in the multi-scale feature representation, feature correspondence groups of the wetland plants are selected.

[0028] The topological skeleton of the multi-scale feature representation is constructed based on the spatial correspondence between the scales, and then integrated into the feature correspondence group to obtain the feature manifold of the wetland plant.

[0029] Along the topological structure of the feature manifold, the detailed information of the fine-scale features is transferred and fused to the coarse-scale features to obtain the fused feature representation of the wetland plants.

[0030] In a preferred embodiment, the process of transferring and fusing the detailed information of the fine-scale features to the coarse-scale features along the topological structure of the characteristic manifold to generate the fused feature representation of the wetland plants includes:

[0031] Based on the topological connectivity of the feature manifold, the information transmission path between the fine-scale feature and the coarse-scale feature is identified;

[0032] Along the information transmission path, the local texture structure information contained in the fine-scale features is transmitted to the region of the coarse-scale features, thereby realizing the cross-scale transfer of the detailed features;

[0033] During the cross-scale migration process, based on the geometric constraints of the characteristic manifold, the feature expression conflict between the coarse-scale features and the fine-scale features is eliminated, resulting in a coordinated feature expression of the wetland plants.

[0034] By optimizing the distribution continuity of the coordinated feature representation in the manifold space, the fusion feature representation of the wetland plants is obtained.

[0035] In a preferred embodiment, the step of selecting the primary comparison feature path based on the correlation strength between different feature dimensions in the fused feature identifier to obtain the comparison strategy for the wetland plants includes:

[0036] The connection strength between the feature dimensions is determined based on the co-occurrence relationship between the feature dimensions in the fused feature representation;

[0037] A feature topology graph of the wetland plants is constructed using the feature dimensions as nodes and the connection strength as edges.

[0038] The centrality index of the wetland plant is obtained based on the relative proportion of the number of associated edges of a node in the feature topology graph to the total number of associated edges in the feature topology graph.

[0039] The node with the highest centrality index is selected as the starting point of the path, and the feature comparison order is determined according to the edge with the strongest connection strength among the nodes, thus obtaining the comparison strategy for the wetland plants.

[0040] In a preferred embodiment, the application of the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plants includes:

[0041] Based on the feature path order determined in the comparison strategy, the wetland plant features to be compared are extracted from the fused feature representation;

[0042] The features to be compared are compared item by item with the features in the known wetland plant feature database to obtain the degree of consistency of the wetland plant features;

[0043] Based on the degree of feature consistency between features, the similarity relationship between the fused feature representation and the known wetland plant feature database is quantified to obtain the similarity score of the wetland plants.

[0044] In a preferred embodiment, the similarity score is calculated using the following formula:

[0045] ;

[0046] In the formula, Score the similarity. The total number of feature dimensions in the fused feature representation. The fusion feature represents the first Each feature dimension value, The first feature corresponding to the known wetland plant feature database Each feature dimension value, Preset feature dimensions The tolerance coefficient, The feature dimension in the feature topology graph and The strength of the connection between them For feature dimensions The node centrality index in the feature topology graph, The preset neighbor influence factor, The sum of the centrality indices of all feature dimension nodes in the feature topology graph is used to normalize the weights.

[0047] In a preferred embodiment, the step of performing a consistency check on the growth environment context information and phenological period context information of the wetland plants based on the similarity score to obtain the species identification result of the wetland plants includes:

[0048] Candidate wetland plant species are selected based on the ranking of candidate species in the similarity score.

[0049] The water type and soil moisture information of the area where the wetland plants are located serve as the environmental context information for the growth of the wetland plants.

[0050] The current growth stage of the wetland plants and their seasonal correspondence are used as the phenological context information of the wetland plants.

[0051] Based on the compatibility between the candidate wetland plant species and the contextual information of their growth environment, environmentally different species among the wetland plants are eliminated to obtain environmentally compatible candidate species of the wetland plants.

[0052] Verify the degree of conformity between the candidate wetland plant species and the phenological context information, and select candidate species of wetland plants that are consistent with the phenology.

[0053] The species identification results of the wetland plants are determined by combining the environmentally compatible candidate species and the phenologically consistent candidate species.

[0054] To address the aforementioned problems, the present invention also provides a wetland plant identification system based on multidimensional feature similarity comparison, the system comprising:

[0055] The basic feature construction module is used to construct a basic feature description of the wetland plants by using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details.

[0056] A multi-scale feature representation module is used to statistically analyze the color distribution and texture direction of visually salient regions in the image data based on the basic feature description, so as to obtain the multi-scale feature representation of the wetland plants.

[0057] The fusion feature generation module is used to match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fusion feature representation of the wetland plants.

[0058] The comparison strategy formulation module is used to select the primary comparison feature path based on the strength of the correlation between different feature dimensions in the fused feature identifier, and obtain the comparison strategy for the wetland plants.

[0059] The similarity comparison and scoring module is used to apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant.

[0060] The species identification result generation module is used to perform consistency verification on the growth environment context information and phenological period context information of the wetland plants based on the similarity score, and obtain the species identification result of the wetland plants.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The wetland plant identification technology of this invention achieves significant improvements in feature extraction and multi-scale feature representation. It uses the color features of wetland plant image data as the macroscopic information foundation, supplemented by texture features as microscopic details. Simultaneously, it optimizes image quality by suppressing high-frequency noise and expanding grayscale distribution, constructing a basic feature description that combines global relevance and local detail, effectively improving the completeness and recognizability of feature information. Based on this, it further generates accurate multi-scale feature representations by detecting visually salient regions, establishing concentric analysis windows to statistically analyze color distribution vectors, and tracing texture direction descriptions superimposed onto the color distribution vectors. This provides high-quality, high-recognition data support for subsequent feature fusion and similarity comparison, ensuring the effectiveness of the identification process from the source.

[0063] 2. This invention possesses significant advantages in the fusion feature generation, comparison strategy, and recognition result verification stages. By matching the geometric positions and descriptors of multi-scale features to construct a feature manifold, it guides the transfer and fusion of detailed information from fine-scale features to coarse-scale features, eliminating feature expression conflicts and optimizing distribution continuity, generating a fusion feature representation that combines global correlation and local details. Simultaneously, it selects the primary comparison path based on the strength of the correlation between feature dimensions, quantifies similarity relationships using a similarity scoring formula that includes parameters such as node centrality and connection strength, and then performs consistency verification through growth environment and phenological period context information. This not only significantly improves the targeting and efficiency of similarity comparison but also ensures the accuracy and reliability of species identification results, effectively meeting the core requirements for efficient and accurate identification in wetland plant identification scenarios. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating a wetland plant identification method based on multidimensional feature similarity comparison, provided in an embodiment of the present invention.

[0065] Figure 2 A functional module diagram of a wetland plant identification system based on multidimensional feature similarity comparison provided in an embodiment of the present invention;

[0066] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0067] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0068] This application provides a wetland plant identification method based on multidimensional feature similarity comparison. The executing entity of this wetland plant identification method based on multidimensional feature similarity comparison includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the wetland plant identification method based on multidimensional feature similarity comparison can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0069] Reference Figure 1The diagram shown is a flowchart illustrating a wetland plant identification method based on multidimensional feature similarity comparison according to an embodiment of the present invention. In this embodiment, the wetland plant identification method based on multidimensional feature similarity comparison includes:

[0070] S1. Using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details, construct the basic feature description of the wetland plants;

[0071] In this embodiment of the invention, the construction of the basic feature description of the wetland plants, based on the color features of the wetland plant image data as the macroscopic information foundation and supplemented by texture features as microscopic details, includes:

[0072] High-frequency noise in wetland plant image data is suppressed to obtain denoised image data of the wetland plants;

[0073] Expanding the grayscale distribution of the denoised image data yields contrast-enhanced image data of the wetland plants;

[0074] The color space is divided from the contrast-enhanced image data, and the distribution pattern of the dominant color tone in the color space is statistically analyzed.

[0075] Based on the distribution pattern, the color features of the contrast-enhanced image data are used as the basis for the macroscopic information of the wetland plants;

[0076] The texture features of the contrast-enhanced image data are used to supplement the microscopic details of the wetland plants;

[0077] The macroscopic information base is combined with the microscopic details to construct a basic characteristic description of the wetland plants.

[0078] The algorithm iterates through each pixel in the original wetland plant image data, selects multiple neighboring pixels around each pixel to form a local neighborhood, averages the gray values ​​of all pixels in this local neighborhood, and replaces the original pixel's gray value with the average value. In this way, the high-frequency noise part of the image with abrupt changes in gray value is reduced, thus obtaining denoised image data.

[0079] By examining the grayscale values ​​of all pixels in the denoised image data, the minimum and maximum grayscale values ​​are determined. The grayscale values ​​of all pixels that were originally concentrated in a smaller grayscale range are linearly mapped according to the interval from the minimum to the maximum grayscale value. This expands the originally narrow grayscale distribution to the entire available grayscale range, making the bright parts of the image brighter and the dark parts darker, enhancing the grayscale differences between different areas, and thus obtaining contrast-enhanced image data.

[0080] The contrast-enhanced image data is split into three independent single-channel image data according to the red, green and blue color channels. Each channel corresponds to the distribution of a basic color. Then, the number of pixels corresponding to different gray values ​​in each channel is counted. The color corresponding to the gray value with the most pixels in each channel is identified as the dominant color of that channel. By combining the dominant colors of the three channels and their respective proportions in the image, the distribution patterns such as the frequency and distribution range of the dominant color in the entire contrast-enhanced image data are summarized.

[0081] Based on the distribution patterns of the dominant colors obtained from the previous statistics, we identified the dominant colors with the highest proportion in the contrast-enhanced image data, clarified the distribution area and coverage of these dominant colors in the image, and the transition relationship between different dominant colors. This set of information that can reflect the overall color appearance of wetland plants is the macroscopic information basis of wetland plants, which can present the color characteristics of wetland plants as a whole.

[0082] By observing the grayscale value changes between adjacent pixels in the contrast-enhanced image data, and recording the grayscale value combinations of adjacent pixels one by one along four different directions (horizontal, vertical, 45 degrees, and 135 degrees), the arrangement patterns and repetition rules of grayscale values ​​in the image are analyzed by counting the number of occurrences of these different grayscale value combinations. This information, which can reflect the fine structure of the surface of wetland plants, is the extracted texture feature, which serves as a supplement to the microscopic details of wetland plants.

[0083] The macroscopic information base reflecting the overall color characteristics of wetland plants obtained above is integrated with the microscopic details reflecting the fine structure of the surface of wetland plants. This ensures that the main color distribution and range in the macroscopic information correspond to and are fully presented with the texture arrangement and patterns in the microscopic information, forming a comprehensive information set that includes both global color information and local detailed features. This set is the basic characteristic description of wetland plants.

[0084] The beneficial effects are as follows: By suppressing high-frequency noise in wetland plant image data to obtain denoised image data, the interference of noise on subsequent image feature extraction can be effectively reduced, ensuring the purity of the image's basic information and providing a clear and reliable image data source for subsequent processing. Expanding the gray-level distribution of the denoised image data to obtain contrast-enhanced image data can significantly improve the gray-level differences in different regions of the image, making it easier to identify and extract details such as the outlines and textures of wetland plants, thus enhancing the recognizability of image details. Dividing the color space from the contrast-enhanced image data and statistically analyzing the distribution patterns of the dominant color tone can systematically and accurately capture the global distribution of color features in the image, avoiding the omission of key color information and providing comprehensive color data support for subsequently determining the macroscopic information basis. Using the color features of the contrast-enhanced image data as the macroscopic information basis for wetland plants based on the above distribution patterns can make the color features... By fully leveraging the macroscopic and global descriptive role, the overall color appearance of wetland plants is clearly presented, providing a global reference for feature description. Extracting texture features from contrast-enhanced image data and using them as supplementary microscopic details of wetland plants fills the gap in the description of local fine structures by macroscopic color features, enriching feature dimensions and making feature descriptions more closely aligned with the actual morphological characteristics of wetland plants. Finally, combining the macroscopic information foundation with the microscopic detail supplementation to construct the basic feature description of wetland plants allows the basic features to possess both global color correlation and local detail specificity, avoiding the limitations of single-type feature descriptions and significantly improving the completeness and recognizability of the basic features. This provides a high-quality and highly usable feature foundation for subsequent image recognition processes based on these basic features, such as multi-scale feature representation, fusion feature generation, and similarity comparison of wetland plants, effectively ensuring the accuracy and effectiveness of subsequent recognition processes.

[0085] S2. Based on the basic feature description, statistically analyze the color distribution and texture direction of visually salient regions in the image data to obtain the multi-scale feature representation of the wetland plants;

[0086] In this embodiment of the invention, the step of statistically analyzing the color distribution and texture direction of visually salient regions in the image data based on the basic feature description to obtain a multi-scale feature representation of the wetland plants includes:

[0087] Based on the aforementioned basic feature description, differential image regions in the contrast-enhanced image data are detected to obtain the visually salient regions of the wetland plants;

[0088] A concentric analysis window for the wetland plants is established with the geometric center of the visually salient region as the origin.

[0089] In the concentric analysis window, tensor synthesis is performed on the main hue and distribution ratio of the contrast-enhanced image data to obtain the color distribution vector of the wetland plants.

[0090] By tracing the grayscale gradient direction of the contrast-enhanced image data, a description of the texture direction of the wetland plants is obtained;

[0091] The texture direction description is superimposed on the color distribution vector to obtain a multi-scale feature representation of the wetland plants.

[0092] Based on the macroscopic color features and microscopic texture features of wetland plants contained in the basic feature description, the contrast-enhanced image data is divided into regions. The image is divided into multiple local regions of equal size. Then, the color features and texture features of each local region are compared with the surrounding local regions one by one. If the main color of a local region is significantly different from the main color of the surrounding regions, and the texture arrangement pattern of the local region is also different from the texture arrangement pattern of the surrounding regions, then the local region is the difference image region. All difference image regions that meet the above conditions are summarized, and the resulting set is the visual salience region of wetland plants.

[0093] First, the geometric center of the region is determined by calculating the average of the x-coordinates and y-coordinates of all pixels within the visually salient region. The average x-coordinates and y-coordinates together constitute the coordinates of the geometric center. Then, using this geometric center as a fixed center, multiple circular windows are drawn with their radii gradually increasing at equal intervals. The centers of these circular windows completely overlap, and their radii increase sequentially. Adjacent circular windows form annular regions of equal width. The analysis structure composed of these concentric circular windows is the concentric analysis window for wetland plants.

[0094] Within each circular window of the concentric analysis window, the dominant hue of the contrast-enhanced image data is statistically analyzed. This involves analyzing the colors of all pixels within each circular window and identifying the most frequently occurring color as the dominant hue for that window. Simultaneously, the proportion of this dominant hue among all pixels within the circular window is calculated, representing the distribution ratio. Then, the dominant hue of each circular window is converted into a specific numerical value according to a preset color-to-value correspondence rule. This value is then associated and combined with its corresponding distribution ratio value. Following the order of the circular windows from the inside out, the associated combination results of all circular windows are arranged sequentially, forming an ordered numerical sequence that constitutes the color distribution vector of wetland plants.

[0095] In contrast-enhanced image data, the grayscale values ​​of adjacent pixels are compared one by one along four fixed directions: horizontal, vertical, 45-degree tilt, and 135-degree tilt. The difference between the grayscale values ​​of two adjacent pixels is calculated, and the direction of grayscale change is determined based on the difference. If the grayscale value of the later pixel is higher than that of the earlier pixel, it is the direction of grayscale increase; if it is lower than that of the earlier pixel, it is the direction of grayscale decrease; if they are equal, it is the direction of grayscale no change. The number of adjacent pixel pairs with grayscale increase, decrease, and no change in each direction is counted. The direction of grayscale change that occurs most frequently is determined as the texture direction of that region. The texture direction of all regions is recorded in text form, and the resulting record is a description of the texture direction of wetland plants.

[0096] First, each texture direction in the texture direction description is converted into a corresponding numerical identifier according to preset rules. For example, a horizontal direction is assigned a specific fixed value, and a vertical direction is assigned another specific fixed value. Then, following the order of the circular windows in the concentric analysis window from the inside out, the texture direction numerical identifier corresponding to each circular window is added to the end of the corresponding value in the color distribution vector. This associates each value in the color distribution vector with its corresponding texture direction numerical identifier, forming an ordered dataset that simultaneously contains color distribution information and texture direction information. This ordered dataset is the multi-scale feature representation of wetland plants.

[0097] The beneficial effects are as follows: Based on the description of fundamental features, differential image regions are detected in contrast-enhanced image data to obtain the visually salient regions of wetland plants. This relies on the comprehensive information of macroscopic color and microscopic texture in the fundamental features to accurately locate the core regions in the image that best represent the characteristics of wetland plants, effectively eliminating interference from background or irrelevant regions. This provides a precise range for subsequent feature analysis of key plant parts, avoiding the influence of invalid information on feature extraction. Establishing a concentric analysis window with the geometric center of the visually salient region as the origin allows for the construction of an analysis framework with clear spatial hierarchy. This enables subsequent statistical analysis of color and texture to proceed along an ordered spatial dimension from the center to the periphery, ensuring that feature information from different spatial locations is captured by the system, providing a structured spatial carrier for the formation of multi-scale features. Tensor synthesis is performed on the dominant hue and distribution ratio of the contrast-enhanced image data within the concentric analysis window to obtain a color distribution vector. This deeply integrates the category information of the dominant hue with the quantitative information of the distribution ratio, rather than presenting the two in isolation. The data allows the color distribution vector to reflect both the core color of each region and the proportion of that color within the region, accurately preserving the spatial distribution correlation of color features. Tracking the grayscale gradient direction of contrast-enhanced image data to obtain a texture direction description can accurately capture the directional patterns of the fine surface structures of wetland plants, filling the gap in microscopic morphological information that cannot be represented by color features alone, making the feature description more consistent with the actual morphological characteristics of plants. Superimposing the texture direction description onto the color distribution vector to obtain a multi-scale feature representation of wetland plants can achieve the organic integration of color spatial distribution information and texture direction information. The final multi-scale feature includes color differences at different spatial levels and covers texture details at each level, forming a comprehensive feature expression with both spatial and morphological dimensions. This provides a high-quality, high-discrimination feature foundation for subsequent image recognition processes such as wetland plant feature matching and species identification based on this feature, significantly improving the accuracy and reliability of subsequent identification processes in distinguishing wetland plant species.

[0098] S3. Match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fused feature representation of the wetland plants;

[0099] In this embodiment of the invention, the step of matching the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish a feature manifold of the multi-scale feature representation, and guiding the multi-scale feature representation based on the feature manifold to generate a fused feature representation of the wetland plants, includes:

[0100] Under a unified coordinate system, the geometric positions of coarse-scale features and fine-scale features in the multi-scale feature representation are aligned to obtain the inter-scale spatial correspondence of the wetland plants.

[0101] Based on the degree of similarity between descriptors of features at different scales in the multi-scale feature representation, feature correspondence groups of the wetland plants are selected.

[0102] The topological skeleton of the multi-scale feature representation is constructed based on the spatial correspondence between the scales, and then integrated into the feature correspondence group to obtain the feature manifold of the wetland plant.

[0103] Along the topological structure of the feature manifold, the detailed information of the fine-scale features is transferred and fused to the coarse-scale features to obtain the fused feature representation of the wetland plants.

[0104] The topological structure along the characteristic manifold transfers and fuses the detailed information of the fine-scale features to the coarse-scale features to generate the fused feature representation of the wetland plants, including:

[0105] Based on the topological connectivity of the feature manifold, the information transmission path between the fine-scale feature and the coarse-scale feature is identified;

[0106] Along the information transmission path, the local texture structure information contained in the fine-scale features is transmitted to the region of the coarse-scale features, thereby realizing the cross-scale transfer of the detailed features;

[0107] During the cross-scale migration process, based on the geometric constraints of the characteristic manifold, the feature expression conflict between the coarse-scale features and the fine-scale features is eliminated, resulting in a coordinated feature expression of the wetland plants.

[0108] By optimizing the distribution continuity of the coordinated feature representation in the manifold space, the fusion feature representation of the wetland plants is obtained.

[0109] First, a unified coordinate system is established, with the geometric center of the visually salient region of wetland plants as the origin. The horizontal direction to the right is set as the positive x-axis, and the vertical direction upward is set as the positive y-axis. The coordinate calculation rules for each pixel in this coordinate system are determined. Next, coarse-scale features and fine-scale features are extracted from the multi-scale feature representation. Coarse-scale features correspond to the feature information of larger areas in the image, while fine-scale features correspond to the feature information of smaller areas. By measuring and recording the coordinates of each vertex in the region covered by the coarse-scale features and the coordinates of each vertex in the region covered by the fine-scale features in the same coordinate system, the coordinate values ​​of the fine-scale feature vertices are adjusted so that the vertex coordinates of the spatially overlapping areas of the fine-scale and coarse-scale feature regions are completely consistent. This achieves the alignment of the geometric positions of the coarse-scale and fine-scale features, ultimately obtaining the inter-scale spatial correspondence of wetland plants.

[0110] First, the specific content of the descriptors for features at different scales in the multi-scale feature representation is clarified. The descriptor for a coarse-scale feature includes the dominant hue, color distribution ratio, and overall texture trend of the feature region at that scale. The descriptor for a fine-scale feature includes the local dominant hue, local texture direction, and texture density of the feature region at that scale. Then, the descriptors of the coarse-scale features are compared item by item with the descriptors of each fine-scale feature. If the dominant hue of the coarse-scale feature is the same as the local dominant hue of a certain fine-scale feature, and the overall texture trend of the coarse-scale feature is consistent with the local texture direction of the fine-scale feature, and the color distribution ratio of the coarse-scale feature can include the color distribution ratio of the fine-scale feature, then the descriptors of the two are determined to be similar. The coarse-scale feature is paired with the corresponding fine-scale feature. After summarizing all feature pairs that meet the above similarity conditions, the feature correspondence group of wetland plants is obtained.

[0111] Based on the established spatial correspondences between scales, the geometric positions of coarse-scale features in a unified coordinate system are used as core nodes of the topological skeleton, with each core node representing a coarse-scale feature region. The geometric positions of fine-scale features that have spatial correspondences with each core node are then used as branch nodes of the topological skeleton, with each branch node representing a fine-scale feature region. Straight line segments connect the core nodes to their corresponding branch nodes, forming a preliminary topological skeleton structure that reflects the spatial relationships between features of different scales. Subsequently, selected feature correspondence groups are incorporated, and the connecting line segments between coarse-scale and fine-scale features in each feature correspondence group are marked to distinguish between ordinary spatial correspondences and correspondences similar to feature descriptors. Through this construction and fusion, the feature manifold of wetland plants is finally obtained.

[0112] Following the topology of the feature manifold, the information transmission path from branch nodes to core nodes is first determined, i.e., information is transmitted through the connecting lines between the two nodes in the feature manifold. Detailed information of fine-scale features is extracted from each branch node, including texture changes, local color transition details, and minute texture structure features within the fine-scale feature region. This detailed information is then transmitted to the corresponding core node according to the transmission path. During the transmission process, the detailed information of the fine-scale features is integrated with the existing information of the coarse-scale features. For example, the texture changes of the fine-scale features are added to the overall texture trend of the coarse-scale features, and the local color transition details of the fine-scale features are added to the color distribution information of the coarse-scale features. This ensures that the detailed information of the fine-scale features can be accurately integrated into the global information of the coarse-scale features, forming a feature information set that combines global features and local details. This set is the fused feature representation of wetland plants.

[0113] Observe the topological connectivity of the feature manifold. This connectivity is specifically represented by the connecting lines between the core nodes representing coarse-scale features and the branch nodes representing fine-scale features, as well as the feature-corresponding association information marked on these lines. For each branch node corresponding to a fine-scale feature, trace its connection path to the coarse-scale feature core node along the existing connecting lines in the feature manifold. Confirm that the branch node is directly associated with the specific core node through only one connecting line. This connecting line from the fine-scale feature branch node to the coarse-scale feature core node, and the node associations it passes through, constitute the information transmission path between the fine-scale and coarse-scale features that needs to be identified.

[0114] Local texture structure information is extracted from fine-scale features. This information includes the shape of texture units within the fine-scale feature region, the spacing between texture units, the density of local textures, and subtle variations in texture direction within a small area. Then, following the identified information transmission path, this local texture structure information is progressively transferred from the branch node region corresponding to the fine-scale feature to the core node region of the coarse-scale feature pointed to by the path. This ensures the integrity of the local texture structure information during the transmission process; that is, every local texture detail of the fine-scale feature is accurately transmitted to the coarse-scale feature region. This process achieves cross-scale transfer of detailed features.

[0115] The geometric constraints of the feature manifold are specifically defined as the spatial boundary range of coarse-scale and fine-scale feature regions in a unified coordinate system, the positional overlap ratio between regions, and the shape matching requirements of feature regions. During the cross-scale migration of detailed features, if the local texture direction of the fine-scale feature is inconsistent with the overall texture trend of the coarse-scale feature, or if the texture density of the fine-scale feature conflicts with the texture density distribution pattern within the coarse-scale feature region, the spatial boundary range and positional overlap ratio in the geometric constraints are used to determine whether the fine-scale feature region is completely within the coarse-scale feature region. If it is, the overall texture trend and density distribution pattern of the coarse-scale feature are used as a benchmark to adjust the conflicting texture information in the fine-scale feature to match the feature expression of the coarse-scale feature, eliminating the feature expression conflict between the two and obtaining a coordinated feature expression of wetland plants.

[0116] This analysis examines the distribution of coordinated feature representations within the manifold space, which is composed of the topological structure of the feature manifold and the coordinates of each feature region. Distribution continuity refers to the naturalness of the transition between coordinated feature information between adjacent feature nodes, without significant abrupt changes. If a sudden change in texture direction or color distribution is found in the coordinated features of two adjacent feature nodes, transitional feature information between these two features is supplemented based on their connection relationship and spatial distance within the manifold space. For example, between nodes with abrupt changes in texture direction, texture direction information that gradually transitions from one direction to another is added, ensuring a smooth connection between the coordinated features of adjacent nodes. This optimizes the distribution continuity of coordinated feature representations within the manifold space, resulting in a fused feature representation of wetland plants.

[0117] The beneficial effects are as follows: Aligning the geometric positions of coarse-scale and fine-scale features in multi-scale feature representation under a unified coordinate system to obtain the inter-scale spatial correspondence of wetland plants can eliminate the spatial deviation of features at different scales, enabling coarse and fine-scale features to establish a clear association in the same spatial dimension, laying a spatial consistency foundation for subsequent feature matching and information fusion; Filtering feature correspondence groups of wetland plants based on the similarity of descriptors of features at different scales can accurately select coarse and fine-scale feature pairs that are correlated in attributes, eliminating interference from unrelated features, improving the accuracy of feature correspondence, and ensuring the effectiveness of subsequent information fusion; Constructing a topological skeleton based on the inter-scale spatial correspondence and integrating it with feature correspondence groups to obtain the feature manifold of wetland plants. This feature manifold not only contains the spatial distribution structure of features at different scales but also integrates the attribute associations between features, providing a clear and orderly structural framework for the transmission of detailed information from fine-scale features to coarse-scale features, avoiding information transmission chaos; Identifying the information transmission path between fine-scale and coarse-scale features based on the topological connectivity of the feature manifold can clarify the transmission direction and path of detailed information, ensuring the accuracy of the local texture structure of fine-scale features. Information can be precisely and directionally transmitted to the corresponding coarse-scale feature regions, avoiding mistransmission or omission. Local texture structure information of fine-scale features is transferred to coarse-scale feature regions along the information transmission path to achieve cross-scale migration of detailed features. This compensates for the shortcomings of coarse-scale features in describing local fine structures, allowing coarse-scale features to possess both global relevance and local detail, enriching the dimension of feature expression. During cross-scale migration, the geometric constraints of the feature manifold eliminate feature expression conflicts between coarse and fine-scale features to obtain coordinated feature expressions of wetland plants. This resolves potential contradictions in texture direction and density of features at different scales, ensuring that the fused feature information remains consistent in logic and attributes, thus improving feature reliability. Optimizing the continuity of the distribution of coordinated feature expressions in the manifold space to obtain fused feature representations of wetland plants allows for a smooth transition in spatial distribution of fused features, avoiding information gaps caused by abrupt feature mutations. Ultimately, this results in high-quality fused features that combine global structure, local detail, and spatial coherence, providing accurate and complete feature support for subsequent image recognition processes such as wetland plant feature comparison and species identification, effectively improving the accuracy and reliability of subsequent recognition processes.

[0118] S4. Based on the strength of the correlation between different feature dimensions in the fused feature identifier, select the primary comparison feature path to obtain the comparison strategy for the wetland plants.

[0119] In this embodiment of the invention, the step of selecting the primary comparison feature path based on the correlation strength between different feature dimensions in the fused feature identifier to obtain the comparison strategy for the wetland plants includes:

[0120] The connection strength between the feature dimensions is determined based on the co-occurrence relationship between the feature dimensions in the fused feature representation;

[0121] A feature topology graph of the wetland plants is constructed using the feature dimensions as nodes and the connection strength as edges.

[0122] The centrality index of the wetland plant is obtained based on the relative proportion of the number of associated edges of a node in the feature topology graph to the total number of associated edges in the feature topology graph.

[0123] The node with the highest centrality index is selected as the starting point of the path, and the feature comparison order is determined according to the edge with the strongest connection strength among the nodes, thus obtaining the comparison strategy for the wetland plants.

[0124] First, the feature dimensions included in the fusion feature representation are defined. These dimensions specifically cover dimensions directly related to plant characteristics, such as the distribution of the dominant hue, local texture direction, color transition patterns, and texture density variations of wetland plants. Then, fusion feature data from multiple wetland plant samples are collected, and the co-occurrence of any two feature dimensions in these samples is statistically analyzed. For example, the number of times "a certain dominant hue" and "a specific texture direction" co-occur in the samples is counted. The more times two feature dimensions co-occur, the stronger their co-occurrence relationship, and the higher the corresponding connection strength. Conversely, the fewer times they co-occur, the lower the connection strength. The connection strength between all feature dimensions is determined in this way.

[0125] Each feature dimension in the fused feature representation is treated as an independent node. A fixed position is assigned to each node on the plane, and the corresponding feature dimension name is labeled. Next, the connection strength between the feature dimensions determined in the previous step is examined. For two feature dimension nodes with a connection strength, a line segment is used to connect them; this line segment becomes an edge in the feature topology graph. The connection strength of the edge is labeled on it. Through this node drawing and edge connection operation, a feature topology graph of wetland plants is constructed.

[0126] First, traverse the entire feature topology graph and count the total number of edges. This count represents the total number of associated edges in the feature topology graph. Then, for each node in the feature topology graph, count the number of edges directly connected to that node. This count represents the number of associated edges for that node. Next, divide the number of associated edges for each node by the total number of associated edges to obtain a ratio. This ratio represents the centrality index of that node in the feature topology graph. For example, if a node has a certain number of associated edges and the total number of associated edges is a certain number, the value obtained by dividing the two is the centrality index of that node. Calculate the centrality index of all nodes in the same way to obtain the centrality index of wetland plants.

[0127] First, compare the centrality indices of all nodes and find the node with the highest value. This node is then designated as the primary starting point for the feature alignment path. Next, observe all edges connected to this starting node, checking the connection strength of each edge. Select the edge with the strongest connection strength; the node connected to this edge will be the next feature dimension to be compared. Then, using this newly determined node as the current node, examine all its connected edges again, selecting the edge with the strongest connection strength to determine the next feature dimension to be compared. Continue this process until all feature dimensions are included in the alignment sequence. The resulting ordered feature alignment sequence is the alignment strategy for wetland plants.

[0128] The beneficial effects are as follows: Determining the connection strength between feature dimensions based on the co-occurrence relationship in the fused feature representation allows for the objective quantification of the correlation between features based on their co-occurrence in actual wetland plant samples. This avoids biases caused by subjective judgment, providing accurate and reliable correlation basis for subsequent feature topology construction and ensuring that the connections between features reflect the true plant characteristic attribute relationships. Constructing a feature topology graph of wetland plants using feature dimensions as nodes and connection strength as edges transforms abstract feature dimension relationships into an intuitive graphical structure, clearly presenting the connection status and strength differences between each feature dimension. This facilitates rapid location of core features and key correlations, avoiding inefficient subsequent analysis due to chaotic feature correlations. Furthermore, obtaining the wetland plant centrality index based on the relative proportion of the number of associated edges to the total number of associated edges in the feature topology graph accurately reflects the correlation between features. Identifying the core feature dimensions with the widest association range and the greatest impact on other features in the feature network, these core features often play a more decisive role in distinguishing wetland plant species, providing an objective quantitative standard for determining the key features to be prioritized for comparison. Selecting the node with the highest centrality index as the starting point of the path and determining the feature comparison order according to the edge with the strongest connection strength in the node yields a wetland plant comparison strategy. This allows the comparison process to prioritize core features and the most strongly associated features, reducing invalid comparisons of secondary features and significantly improving feature comparison efficiency. Simultaneously, it ensures that the comparison focuses on key feature dimensions that are more discriminative for wetland plant identification, providing a precise and efficient comparison direction for subsequent similarity comparisons with known wetland plant feature databases. This effectively improves the effectiveness and reliability of the feature comparison stage in the wetland plant identification process, laying an efficient comparison foundation for the final accurate identification of wetland plant species.

[0129] S5. Apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant.

[0130] In this embodiment of the invention, the application of the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain the similarity score of the wetland plants includes:

[0131] Based on the feature path order determined in the comparison strategy, the wetland plant features to be compared are extracted from the fused feature representation;

[0132] The features to be compared are compared item by item with the features in the known wetland plant feature database to obtain the degree of consistency of the wetland plant features;

[0133] Based on the degree of feature consistency between features, the similarity relationship between the fused feature representation and the known wetland plant feature database is quantified to obtain the similarity score of the wetland plants.

[0134] The formula for calculating the similarity score is as follows:

[0135] ;

[0136] In the formula, Score the similarity. The total number of feature dimensions in the fused feature representation. The fusion feature represents the first Each feature dimension value, The first feature corresponding to the known wetland plant feature database Each feature dimension value, Preset feature dimensions The tolerance coefficient, The feature dimension in the feature topology graph and The strength of the connection between them For feature dimensions The node centrality index in the feature topology graph, The preset neighbor influence factor, The sum of the centrality indices of all feature dimension nodes in the feature topology graph is used to normalize the weights.

[0137] First, the sequence of feature paths determined in the comparison strategy is clarified. This sequence is a sequence of feature dimensions arranged from high to low correlation strength around the core feature dimensions. For example, the sequence would be: distribution of the main hue of wetland plants, local texture direction, color transition patterns, and texture density changes. Then, based on this feature path sequence, specific information for each corresponding feature dimension is extracted from the fused feature representation. For example, when extracting the main hue distribution feature, the types of main hues and the proportion of each main hue in the plant area recorded in the fused feature representation are obtained; when extracting the local texture direction feature, the texture direction and distribution range of different areas recorded in the fused feature representation are obtained. This set of feature information extracted in the path sequence is used as the comparison feature for wetland plants.

[0138] First, a database of known wetland plant characteristics is retrieved. This database stores standard characteristics of various known wetland plants, and each plant's standard characteristics include a corresponding feature dimension for the feature to be compared. Next, following the extraction order of the features to be compared, the first feature dimension of the feature to be compared is compared with the standard information of the corresponding feature dimension for each known plant in the database. This determines whether the dominant hue type of the feature to be compared is consistent with the standard dominant hue type and whether the proportion of the dominant hue of the feature to be compared is consistent with the standard proportion range. Then, in the same way, subsequent feature dimensions of the feature to be compared are compared item by item with the standard characteristics of the corresponding plants in the database. The consistency of each feature dimension is recorded, and the consistency of all feature dimensions is summarized to form the feature consistency degree of wetland plants.

[0139] First, quantification rules are established for the degree of feature consistency. If a feature dimension is completely consistent, a high similarity contribution value is assigned to that feature dimension; if a feature dimension is partially consistent, a medium similarity contribution value is assigned based on the consistency ratio; if a feature dimension is completely inconsistent, a low similarity contribution value or zero similarity contribution value is assigned to that feature dimension. Then, according to these quantification rules, the sum of similarity contribution values ​​for all feature dimensions in the features to be compared is calculated. This sum is then combined with the total number of feature dimensions of the standard features of the plant in the known wetland plant feature database to convert the sum of similarity contribution values ​​into a numerical value that can intuitively reflect the degree of similarity. For example, the higher the sum of similarity contribution values, the larger the corresponding numerical value. This numerical value is the similarity score of the wetland plant, which directly quantifies the similarity relationship between the fused feature representation and the corresponding plant features in the known wetland plant feature database.

[0140] In the similarity score calculation formula, The parameter representing the total number of feature dimensions in the fused feature representation is derived from the fused feature representation itself. This fused feature representation is generated by fusing detailed information from fine-scale features to coarse-scale features along the topological structure of the feature manifold. The specific value of this parameter is the number of all independent feature dimensions contained in the fused feature representation.

[0141] The first representing the fusion feature representation The parameters of each feature dimension are derived from the fused feature representation. In the fused feature representation, each feature dimension has a corresponding specific numerical description. For example, the dominant hue distribution dimension has a dominant hue percentage value, and the local texture direction dimension has a texture direction angle value, etc. The parameter of the first feature dimension is selected. The specific numerical value corresponding to each feature dimension is the value of that parameter.

[0142] The first corresponding feature in the known wetland plant feature database The parameter for the first feature dimension value is derived from a database of known wetland plant features. This database stores standard features of various known wetland plants, and each standard feature contains a feature dimension corresponding to the fused feature representation. A standard feature of a known plant is selected from the database and compared with the first feature dimension of the fused feature representation. The standard value corresponding to each feature dimension is the value of that parameter.

[0143] Represents the preset feature dimensions The tolerance coefficient parameter is derived from the actual variation of wetland plant characteristics. After extensive statistical analysis of the natural differences in similar characteristic dimensions of wetland plants, a fixed value is determined based on this range of differences to allow the fusion of feature representations with standard features in the database. There are reasonable differences in each dimension, and this fixed value is the specific value of the tolerance coefficient.

[0144] Feature dimension in the feature topology graph and The parameter for the connection strength between features originates from the feature topology graph, which is constructed with feature dimensions as nodes and connection strengths as edges. The connection strength is determined based on the co-occurrence relationships between feature dimensions in the fused feature representation. In the feature topology graph, the feature dimensions... and The value of the connection strength between them is the value of this parameter.

[0145] Representative feature dimension The parameter of the node centrality index in the feature topology graph comes from the centrality index calculation result. This centrality index is obtained based on the relative proportion of the number of associated edges of a node in the feature topology graph to the total number of associated edges. (Feature dimension) The value of this parameter is the ratio obtained by dividing the number of associated edges of the corresponding node in the feature topology graph by the total number of associated edges.

[0146] The parameter representing the preset neighbor influence factor is based on the actual influence of neighbor associations between feature dimensions on similarity. After analyzing the influence of neighbor feature dimension similarity on the overall similarity judgment in a large number of wetland plant feature comparison cases, a fixed value is determined to control the influence intensity of the neighbor feature dimension on the similarity contribution of the current feature dimension. This fixed value is the specific value of the neighbor influence factor.

[0147] The parameter represents the sum of the centrality indices of all feature dimension nodes in the feature topology graph. It is derived from the cumulative calculation of the centrality indices of all nodes. The sum of the centrality indices of each feature dimension node in the feature topology graph is the value of this parameter.

[0148] The significance of the similarity score calculation formula lies in obtaining a similarity score that can intuitively reflect the degree of similarity between the quantified fusion feature representation and the features in the known wetland plant feature database.

[0149] In the formula, the proportion of the centrality index of each feature dimension to the sum of the centrality indices of all feature dimension nodes is used as the weight because feature dimensions with high centrality indices have a wider range of associations in the feature topology graph and play a stronger decisive role in wetland plant species identification. By using this weight, the key feature dimensions can have a more significant impact on the similarity score, avoiding interference from secondary feature dimensions in the scoring results.

[0150] In the formula, the first exponential part is used to calculate the similarity contribution of a single feature dimension itself, by comparing the fused feature representations. The smaller the difference between a dimension value and the corresponding dimension value in the database, the larger the value of that index part and the higher the similarity contribution. The preset tolerance coefficient controls the degree to which that dimension accepts differences. The larger the tolerance coefficient, the smaller the impact of small differences on the value of that index part, ensuring that reasonable differences do not excessively reduce the similarity contribution.

[0151] In the formula, the second exponent is used to calculate the first exponent. The similarity contribution of neighboring feature dimensions in each feature dimension, in the feature topology graph. and The connection strength reflects the degree of correlation between the two. The higher the connection strength, the greater the impact of the similar contribution of the neighbor dimension on the current dimension. The preset neighbor influence factor controls the overall influence of the neighbor dimension on the similar contribution of the current dimension. The more similar the neighbor dimension is to the database features, the larger the value of this index part, which supplements the lack of similar contribution of the current dimension itself and more comprehensively reflects the feature similarity.

[0152] The formula finally multiplies the (self-similar contribution + neighbor similarity contribution) of each feature dimension by the corresponding weight and adds them all together. The sum is the similarity score. The higher the score, the higher the similarity between the fused feature representation and the corresponding plant features in the known wetland plant feature database. Based on this score, the matching degree between the wetland plant to be identified and the known plants in the database can be determined more accurately, providing a quantitative basis for subsequent species identification.

[0153] The beneficial effects are as follows: Extracting the wetland plant features to be compared from the fused feature representation according to the feature path sequence determined in the comparison strategy allows for precise focus on core feature dimensions with high discriminative power for wetland plant identification. This avoids comparison redundancy caused by extracting irrelevant or secondary features, ensuring a high degree of fit between the features to be compared and subsequent similarity comparison requirements, thus laying the foundation for efficient comparison. Furthermore, comparing the features to be compared with features in a known wetland plant feature database item by item yields the degree of feature consistency, allowing for detailed verification of the matching of each key feature dimension, without overlooking subtle differences between features, and accurately reflecting the consistency between the features to be compared and the database standard features. The alignment of features across various dimensions provides a real and comprehensive basis for subsequent quantification of similarity relationships. By quantifying the similarity relationship between the fused feature representations and the known wetland plant feature database based on the degree of feature consistency between features, a similarity score for wetland plants is obtained. This transforms the abstract feature matching situation into an intuitive numerical expression, clearly presenting the degree of similarity between the wetland plant to be identified and the known plants in the database, avoiding subjective judgment bias. It provides an accurate and quantifiable reference standard for subsequent species identification by combining growth environment and phenological information, effectively improving the efficiency and reliability of the wetland plant similarity comparison process and ensuring the accuracy of the final identification results.

[0154] S6. Based on the similarity score, perform consistency verification on the growth environment context information and phenological period context information of the wetland plants to obtain the species identification results of the wetland plants.

[0155] In this embodiment of the invention, the step of performing consistency verification on the growth environment context information and phenological period context information of the wetland plants based on the similarity score to obtain the species identification result of the wetland plants includes:

[0156] Candidate wetland plant species are selected based on the ranking of candidate species in the similarity score.

[0157] The water type and soil moisture information of the area where the wetland plants are located serve as the environmental context information for the growth of the wetland plants.

[0158] The current growth stage of the wetland plants and their seasonal correspondence are used as the phenological context information of the wetland plants.

[0159] Based on the compatibility between the candidate wetland plant species and the contextual information of their growth environment, environmentally different species among the wetland plants are eliminated to obtain environmentally compatible candidate species of the wetland plants.

[0160] Verify the degree of conformity between the candidate wetland plant species and the phenological context information, and select candidate species of wetland plants that are consistent with the phenology.

[0161] The species identification results of the wetland plants are determined by combining the environmentally compatible candidate species and the phenologically consistent candidate species.

[0162] First, obtain the existing wetland plant similarity scores. These scores quantify the similarity between the fused feature representation and the features of various plants in the known wetland plant feature database. Arrange the corresponding plant species in the known wetland plant feature database according to the similarity scores from high to low, forming a candidate species ranking. Then, review this ranking, retaining plant species whose similarity scores demonstrate significant similarity to the wetland plant to be identified. For example, exclude species with excessively low scores or those that differ significantly from the features of the plant to be identified. Summarize these retained plant species to obtain the candidate wetland plant species.

[0163] Travel to the wetland area where the wetland plants to be identified are located, observe the water conditions in the area, and determine the type of water body. Specifically, this can be done by observing the water source, water flow speed, and water morphology to determine whether it is a freshwater lake, marsh, river tributary, or mudflat, etc. Simultaneously, use a soil moisture monitoring tool, insert it into the soil around the roots of the wetland plants in the area, leave it for a period of time, and read the soil moisture content data displayed by the monitoring tool to determine the soil moisture information. Combine the determined water body type with the detected soil moisture information to provide contextual information about the growth environment of the wetland plants.

[0164] The growth status of the wetland plants to be identified is determined through field observation. This involves examining the development of plant organs, such as the presence of new buds, flowering, fruiting, and leaf growth, to ascertain the plant's current growth stage, such as germination, seedling, mature plant, flowering, fruiting, or withering. The season of observation is also recorded, such as spring, summer, autumn, or winter. A correspondence is established between the current growth stage and the recorded season. For example, if a wetland plant is in its flowering stage and the observation is conducted in summer, this "flowering stage - summer" correspondence provides the phenological context information for the wetland plant.

[0165] The suitable growth environment information for each candidate wetland plant species is queried from the known wetland plant characteristic database. This information includes the range of water types and soil moisture ranges in which each candidate plant can grow normally. The suitable water types for each candidate wetland plant species are compared with the actual water types in the growth environment context information. At the same time, the suitable soil moisture range of the candidate plant is compared with the actual soil moisture information. If the suitable water type of a candidate species does not match the actual water type at all, or if its suitable soil moisture range does not include the actual soil moisture, then the candidate species is determined to be incompatible with the growth environment context information and is removed from the candidate wetland plant species. The remaining candidate species are the environmentally compatible candidate species for wetland plants.

[0166] The phenological patterns of each candidate wetland plant species are retrieved from botanical literature or databases of known wetland plant characteristics. This information clarifies the typical growth stage of each candidate plant in different seasons; for example, a candidate species is typically in the germination stage in spring and the mature stage in summer. The phenological pattern information of each candidate wetland plant species is compared with the phenological context information of the wetland plants. If the phenological pattern of a candidate species completely matches the current growth stage and season—for example, if the candidate species' pattern is "summer-mature stage," and the actual phenological context information is also "mature stage-summer"—then the candidate species is determined to be consistent with the phenological context information. All matching candidate species are then selected to obtain the wetland plant phenologically consistent candidate species.

[0167] The obtained environmentally compatible candidate species are compared with the phenologically consistent candidate species to identify plant species that exist in both sets, forming an intersection candidate species. If the intersection candidate species contains only one plant species, that plant species is directly identified as the wetland plant species. If the intersection candidate species contains multiple plant species, the previous similarity scores are retrieved again to check the similarity scores of these intersection species in the candidate species ranking. The plant species with the highest similarity score is selected as the wetland plant species identification result.

[0168] The beneficial effects are as follows: By ranking candidate wetland plant species according to similarity scores, priority can be given to species with higher similarity to the plant to be identified, while excluding obviously mismatched species, significantly narrowing the scope of subsequent verification and avoiding efficiency losses caused by invalid verification, thus laying the foundation for accurate identification. Using the water type and soil moisture information of the wetland plant's location as contextual information of the growth environment is crucial, as water type and soil moisture are core environmental conditions for wetland plant survival, directly determining whether plants can grow normally. This information accurately reflects the actual survival adaptation needs of the plant to be identified, providing an objective and critical basis for subsequent compatibility judgment. Using the correspondence between the current growth stage of wetland plants and the season as contextual information of phenology is also beneficial, as phenology is an inherent growth pattern of wetland plants. Different plant species exhibit fixed growth stages in specific seasons, and this correspondence accurately reflects the growth characteristics of the plant to be identified, avoiding misjudgments caused by relying solely on feature similarity while ignoring differences in growth patterns. Based on the compatibility of candidate wetland plant species with their growth environment context information, species with environmental differences are eliminated to obtain environmentally compatible candidate species. This process excludes species with similar characteristics to the plant to be identified but unable to survive in the current wetland environment, thus selecting candidate species that are more in line with the actual situation from the perspective of environmental adaptability and reducing erroneous candidates due to environmental mismatch. Verifying the degree of conformity between candidate wetland plant species and phenological period context information to select phenologically consistent candidate species can further eliminate candidate species whose growth stage does not match the current season, verifying the rationality of candidate species from the perspective of plant growth patterns, narrowing the candidate range and improving the reliability of candidate species. Combining environmentally compatible candidate species and phenologically consistent candidate species to determine the wetland plant species identification results allows for dual verification of candidate species selected from feature similarity from both environmental adaptability and growth pattern perspectives. This compensates for the possible biases that may exist in relying solely on feature similarity comparison, ultimately greatly improving the accuracy and reliability of wetland plant species identification results and avoiding identification errors caused by single-dimensional judgment.

[0169] like Figure 2 The diagram shown is a functional block diagram of a wetland plant identification system based on multidimensional feature similarity comparison provided in an embodiment of the present invention.

[0170] The wetland plant identification system 100 based on multidimensional feature similarity comparison described in this invention can be installed in an electronic device. Depending on the functions implemented, the wetland plant identification system 100 may include a basic feature construction module 101, a multi-scale feature representation module 102, a fusion feature generation module 103, a comparison strategy formulation module 104, a similarity comparison scoring module 105, and a species identification result generation module 106. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0171] In this embodiment, the functions of each module / unit are as follows:

[0172] The basic feature construction module 101 is used to construct a basic feature description of the wetland plants by using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details.

[0173] The multi-scale feature representation module 102 is used to statistically analyze the color distribution and texture direction of visually salient regions in the image data based on the basic feature description, so as to obtain the multi-scale feature representation of the wetland plants.

[0174] The fusion feature generation module 103 is used to match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fusion feature representation of the wetland plants.

[0175] The comparison strategy formulation module 104 is used to select the primary comparison feature path based on the correlation strength between different feature dimensions in the fused feature identifier, and obtain the comparison strategy for the wetland plants.

[0176] The similarity comparison and scoring module 105 is used to apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant.

[0177] The species identification result generation module 106 is used to perform consistency verification on the growth environment context information and phenological period context information of the wetland plants based on the similarity score, so as to obtain the species identification result of the wetland plants.

[0178] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0179] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0181] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0182] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A wetland plant identification method based on multidimensional feature similarity comparison, characterized in that, The method includes: S1. Using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details, construct the basic feature description of the wetland plants; S2. Based on the basic feature description, statistically analyze the color distribution and texture direction of visually salient regions in the image data to obtain the multi-scale feature representation of the wetland plants; S3. Match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fused feature representation of the wetland plants; S4. Based on the strength of the correlation between different feature dimensions in the fused feature representation, select the primary comparison feature path to obtain the comparison strategy for the wetland plants. S5. Apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant. S6. Based on the similarity score, perform consistency verification on the growth environment context information and phenological period context information of the wetland plants to obtain the species identification results of the wetland plants.

2. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, characterized in that, The basic feature description of the wetland plants is constructed by using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details, including: High-frequency noise in wetland plant image data is suppressed to obtain denoised image data of the wetland plants; Expanding the grayscale distribution of the denoised image data yields contrast-enhanced image data of the wetland plants; The color space is divided from the contrast-enhanced image data, and the distribution pattern of the dominant color tone in the color space is statistically analyzed. Based on the distribution pattern, the color features of the contrast-enhanced image data are used as the basis for the macroscopic information of the wetland plants; The texture features of the contrast-enhanced image data are used to supplement the microscopic details of the wetland plants; The macroscopic information base is combined with the microscopic details to construct a basic characteristic description of the wetland plants.

3. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 2, characterized in that, Based on the aforementioned basic feature description, the color distribution and texture orientation of visually salient regions in the image data are statistically analyzed to obtain a multi-scale feature representation of the wetland plants, including: Based on the aforementioned basic feature description, differential image regions in the contrast-enhanced image data are detected to obtain the visually salient regions of the wetland plants; A concentric analysis window for the wetland plants is established with the geometric center of the visually salient region as the origin. In the concentric analysis window, tensor synthesis is performed on the main hue and distribution ratio of the contrast-enhanced image data to obtain the color distribution vector of the wetland plants. By tracing the grayscale gradient direction of the contrast-enhanced image data, a description of the texture direction of the wetland plants is obtained; The texture direction description is superimposed on the color distribution vector to obtain a multi-scale feature representation of the wetland plants.

4. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, characterized in that, The process of matching the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guiding the multi-scale feature representation based on the feature manifold to generate the fused feature representation of the wetland plants, includes: Under a unified coordinate system, the geometric positions of coarse-scale features and fine-scale features in the multi-scale feature representation are aligned to obtain the inter-scale spatial correspondence of the wetland plants. Based on the degree of similarity between descriptors of features at different scales in the multi-scale feature representation, feature correspondence groups of the wetland plants are selected. The topological skeleton of the multi-scale feature representation is constructed based on the spatial correspondence between the scales, and then integrated into the feature correspondence group to obtain the feature manifold of the wetland plant. Along the topological structure of the feature manifold, the detailed information of the fine-scale features is transferred and fused to the coarse-scale features to obtain the fused feature representation of the wetland plants.

5. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 4, characterized in that, The topological structure along the characteristic manifold transfers and fuses the detailed information of the fine-scale features to the coarse-scale features to generate the fused feature representation of the wetland plants, including: Based on the topological connectivity of the feature manifold, the information transmission path between the fine-scale feature and the coarse-scale feature is identified; Along the information transmission path, the local texture structure information contained in the fine-scale features is transmitted to the region of the coarse-scale features, thereby realizing the cross-scale transfer of the fine-scale features; During the cross-scale migration process, based on the geometric constraints of the characteristic manifold, the feature expression conflict between the coarse-scale features and the fine-scale features is eliminated, resulting in a coordinated feature expression of the wetland plants. By optimizing the distribution continuity of the coordinated feature representation in the manifold space, the fusion feature representation of the wetland plants is obtained.

6. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, characterized in that, The step of selecting the primary comparison feature path based on the strength of the correlation between different feature dimensions in the fused feature representation to obtain the comparison strategy for the wetland plants includes: The connection strength between the feature dimensions is determined based on the co-occurrence relationship between the feature dimensions in the fused feature representation; A feature topology graph of the wetland plants is constructed using the feature dimensions as nodes and the connection strength as edges. The centrality index of the wetland plants is obtained based on the relative proportion of the number of associated edges of nodes in the feature topology graph to the total number of associated edges in the feature topology graph. The node with the highest centrality index is selected as the starting point of the path, and the feature comparison order is determined according to the edge with the strongest connection strength among the nodes, thus obtaining the comparison strategy for the wetland plants.

7. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, characterized in that, The comparison strategy described above is applied to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plants, including: Based on the feature path order determined in the comparison strategy, the wetland plant features to be compared are extracted from the fused feature representation; The features to be compared are compared item by item with the features in the known wetland plant feature database to obtain the degree of consistency of the wetland plant features; Based on the degree of feature consistency between features, the similarity relationship between the fused feature representation and the known wetland plant feature database is quantified to obtain the similarity score of the wetland plants.

8. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 6, characterized in that, The formula for calculating the similarity score is as follows: ； In the formula, Score the similarity. The total number of feature dimensions in the fused feature representation. The fusion feature represents the first Each feature dimension value, The first feature corresponding to the known wetland plant feature database Each feature dimension value, Preset feature dimensions The tolerance coefficient, The feature dimension in the feature topology graph and The strength of the connection between them For feature dimensions The node centrality index in the feature topology graph, The preset neighbor influence factor, The sum of the centrality indices of all feature dimension nodes in the feature topology graph is used to normalize the weights.

9. The wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, characterized in that, The consistency verification of the growth environment context information and phenological period context information of the wetland plants based on the similarity score is performed to obtain the species identification results of the wetland plants, including: Candidate wetland plant species are selected based on the ranking of candidate species in the similarity score. The water type and soil moisture information of the area where the wetland plants are located serve as the environmental context information for the growth of the wetland plants. The current growth stage of the wetland plants and their seasonal correspondence are used as the phenological context information of the wetland plants. Based on the compatibility between the candidate wetland plant species and the contextual information of their growth environment, environmentally different species among the wetland plants are eliminated to obtain environmentally compatible candidate species of the wetland plants. Verify the degree of conformity between the candidate wetland plant species and the phenological context information, and select candidate species of wetland plants that are consistent with the phenology. The species identification results of the wetland plants are determined by combining the environmentally compatible candidate species and the phenologically consistent candidate species.

10. A wetland plant identification system based on multidimensional feature similarity comparison, used to implement the wetland plant identification method based on multidimensional feature similarity comparison as described in claim 1, the system comprising: The basic feature construction module is used to construct a basic feature description of the wetland plants by using the color features of the wetland plant image data as the macroscopic information basis and the texture features as the microscopic details. A multi-scale feature representation module is used to statistically analyze the color distribution and texture direction of visually salient regions in the image data based on the basic feature description, so as to obtain the multi-scale feature representation of the wetland plants. The fusion feature generation module is used to match the geometric positions and descriptors of features at different scales in the multi-scale feature representation to establish the feature manifold of the multi-scale feature representation, and guide the multi-scale feature representation based on the feature manifold to generate the fusion feature representation of the wetland plants. The comparison strategy formulation module is used to select the primary comparison feature path based on the strength of the correlation between different feature dimensions in the fused feature representation, and obtain the comparison strategy for the wetland plants. The similarity comparison and scoring module is used to apply the comparison strategy to compare the fused feature representation with a known wetland plant feature database to obtain a similarity score for the wetland plant. The species identification result generation module is used to perform consistency verification on the growth environment context information and phenological period context information of the wetland plants based on the similarity score, and obtain the species identification result of the wetland plants.

Images