A forest and grass resource precision identification method based on image analysis

Abstract

The present application relates to the technical field of forest and grass resources identification, and particularly relates to a forest and grass resource accurate identification method based on image analysis, comprising the following steps: step S1: collecting multi-view sequence images of target forest and grass sample plots, synchronously acquiring real-time positioning data, shooting posture parameters and environment site basic data corresponding to the sample plots; step S2: uniformly preprocessing the collected sequence images; step S3: matching feature points, positioning and posture data based on the sequence images; step S4: multi-scale segmentation of the standardized images, extracting the contours and feature parameters of different types of vegetation targets such as trees, shrubs and herbaceous plants; step S5: completing forest and grass species classification, resource parameter inversion and health grade determination based on the extracted feature parameters, and outputting the forest and grass resource investigation results matched with geographic coordinates. The present application is free from the limitation of physical scales, is suitable for multi-plant structure detection, and realizes full-plant parameter calculation without the constraint of close-range targets.

Description

Technical Field

[0001] This invention relates to the field of forest and grassland resource identification technology, and in particular to a method for accurate identification of forest and grassland resources based on image analysis. Background Technology

[0002] Forest and grassland resources are the core carriers for maintaining the stability of terrestrial ecosystems and are the core survey objects for ecological protection and resource supervision. Manual field surveys suffer from low efficiency, limited coverage, and poor data timeliness. Therefore, image analysis technology is needed to conduct surveys to achieve large-scale and efficient acquisition of forest and grassland resource data.

[0003] Existing technologies, such as the one disclosed in CN115131667A, include a plant trait detection method and system based on image analysis. The method includes: acquiring an image with a ruler and a part of the plant to be tested placed side-by-side; performing image enhancement processing to identify the scale markings on the ruler in the image; and performing geometric analysis on the plant part in the image based on the scale markings to obtain phenotypic traits related to the geometric size and shape of the plant. This improves the accuracy of plant trait detection and can simultaneously identify multiple phenotypic traits related to the geometric size and shape of the plant, thus improving the efficiency of plant trait detection.

[0004] However, this technical solution has the following problems in practical use: Image-based plant phenotypic trait detection technology acquires images of a ruler placed side-by-side with the plant part to be tested in a laboratory environment, and then uses image enhancement and ruler recognition calibration to detect plant phenotypic traits. However, because it relies on close-range calibration with a physical ruler, it is only suitable for detecting single plant organs and cannot cover large-scale survey scenarios of forest and grassland communities in the wild. This limits the scope of application of the technology and restricts the improvement of efficiency in field surveys. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that the existing technology relies on physical rulers for close-range calibration, which can only be adapted to the detection of single plant organs and cannot cover large-scale survey scenarios of wild forest and grassland communities, thus limiting the scope of application of the technology. Therefore, this invention proposes a precise identification method for forest and grassland resources based on image analysis.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for accurate identification of forest and grassland resources based on image analysis includes the following steps: Step S1: Collect multi-view sequence images of the target forest and grassland plots, and simultaneously acquire real-time positioning data, shooting posture parameters and environmental site basic data corresponding to the plots; Step S2: Perform unified preprocessing on the acquired sequence images, sequentially completing illumination normalization correction, complex background suppression, vegetation occlusion area completion, and noise filtering to obtain standardized forest and grassland images; Step S3: Based on the matching feature points, localization and pose data of the sequence images, combined with the prior knowledge base of forest and grassland vegetation, establish a global mapping between pixels and geospatial coordinates to complete the targetless spatial scale calibration. Step S4: Perform multi-scale segmentation on the standardized image to extract the contours and feature parameters of different types of vegetation targets, including trees, shrubs, and herbs; Step S5: Based on the extracted feature parameters, complete the classification of forest and grassland species, inversion of resource parameters, and determination of health level, and output the forest and grassland resource survey results matching geographic coordinates.

[0007] Preferably, the complex background suppression and occlusion region completion in step S2 include: identifying non-vegetation background and vegetation occlusion regions in the image through semantic segmentation, performing mask suppression on the non-vegetation background, and using adjacent frame feature matching and interpolation algorithms on the occlusion regions to complete vegetation feature completion and repair.

[0008] Preferably, the targetless spatial scale calibration in step S3 includes: performing feature point matching and bundle adjustment on the sequence images, constructing a three-dimensional sparse point cloud of the sample plot and solving the image exterior orientation elements, combining the forestry and grassland prior knowledge base to complete the absolute scale calculation, establishing a rigid coordinate mapping, and generating global scale conversion parameters.

[0009] Preferably, the multi-scale segmentation in step S4 includes: using a multi-scale threshold segmentation and contour optimization algorithm, setting a hierarchical detection window, extracting and segmenting features for the corresponding structures of trees, shrubs and grasses respectively, performing edge optimization and target segmentation on overlapping vegetation targets, and outputting segmented regions and contour key points.

[0010] Preferably, the species classification in step S5 includes: extracting geometric morphology and texture spectrum from the segmented vegetation target, constructing a multi-dimensional feature parameter set, generating a comprehensive feature vector using a multi-feature weighted fusion algorithm, inputting it into a pre-constructed classification and discrimination model, and outputting the species classification result and confidence level.

[0011] Preferably, the resource parameter inversion in step S5 includes: calculating the core parameters of tree height, diameter at breast height, and volume for trees; calculating the parameters of clump width, clump height, and biomass for shrubs; calculating the parameters of cover, height, and frequency for herbs; identifying outliers based on features; and outputting the vegetation health stress level results.

[0012] Preferably, step S5 further includes an accuracy optimization step, which includes comparing the output classification results with resource parameters and field measured data of the sample plots. When the error exceeds a preset threshold, the corresponding data is added as an incremental sample to the algorithm training set for iterative optimization to improve the recognition accuracy.

[0013] Preferably, in step S1, the acquired multi-view sequence images, positioning data, attitude parameters and environmental ground data are synchronized at the millisecond level and spatially registered at the pixel level to ensure the spatiotemporal reference of the multi-source data is unified.

[0014] Preferably, in step S4, the detection threshold and window size of the segmentation algorithm are adaptively adjusted based on the vegetation type, shooting distance and ambient lighting conditions of the forest and grassland plot to be identified, so as to adapt to the identification needs of different field scenes.

[0015] Preferably, in step S5, the output forest and grassland resource survey results are standardized and coded according to the national forest and grassland resource inventory standards to generate spatial vector data and statistical reports that can be directly entered into the database, supporting batch archiving and summary statistics of multiple batches of sample plot data.

[0016] Compared with the prior art, the beneficial effects of this invention are as follows: 1. This invention, through targetless spatial scale calibration and multi-dimensional feature extraction, breaks free from the limitations of physical scales, adapts to the detection of various vegetation structures, realizes the calculation of all vegetation parameters and the accurate identification of forest and grassland resources without the constraints of close-range targets, breaks through detection limitations, and expands the overall applicability of the technology.

[0017] 2. This invention adapts to complex field environments and covers a wide range of forest and grassland community survey scenarios through multi-source data spatiotemporal registration and multi-scale image segmentation. It enables efficient collection and standardized output of large-scale forest and grassland resources, breaks application scenario limitations, and improves the overall efficiency of the entire survey process. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a method for accurate identification of forest and grassland resources based on image analysis proposed in this invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Reference Figure 1 A method for accurate identification of forest and grassland resources based on image analysis includes the following steps: Step S1: Collect multi-view sequence images of the target forest and grassland plots, and simultaneously acquire real-time positioning data, shooting posture parameters and environmental site basic data corresponding to the plots; the image acquisition unit and the positioning posture acquisition unit operate in conjunction to ensure the temporal consistency of multi-source data acquisition.

[0021] Step S2: Perform unified preprocessing on the acquired sequence images, sequentially completing illumination normalization correction, complex background suppression, vegetation occlusion area completion, and noise filtering to obtain standardized forest and grassland images; the image processing unit receives the acquired raw data and completes all image preprocessing operations according to the preset process.

[0022] Step S3: Based on the matching feature points, localization and pose data of the sequence images, and combined with the prior knowledge base of forest and grassland vegetation, establish a global mapping between pixels and geospatial coordinates to complete the targetless spatial scale calibration; the scale calibration unit retrieves the prior knowledge base data in the storage unit to complete the calculation of the coordinate mapping relationship.

[0023] Step S4: Perform multi-scale segmentation on the standardized image to extract the contours and feature parameters of different types of vegetation targets, such as trees, shrubs, and herbs; the image segmentation unit receives the standardized image data and completes the feature extraction and segmentation operations for different vegetation targets.

[0024] Step S5: Based on the extracted feature parameters, complete the classification of forest and grassland species, inversion of resource parameters, and determination of health level, and output the forest and grassland resource survey results matching geographic coordinates. The identification and calculation unit receives the feature parameters, completes the full-dimensional calculation, and outputs the final result through the output unit.

[0025] Step S2, involving complex background suppression and occlusion region completion, includes: identifying non-vegetation backgrounds and occluded areas in the image through semantic segmentation; the semantic segmentation module performs pixel-level classification of the input image to distinguish different types of image regions; masking suppression is applied to the non-vegetation background, with the masking module zeroing out pixel values ​​in non-vegetation areas to reduce interference from invalid regions; adjacent frame feature matching and interpolation algorithms are used for occluded areas, with the feature matching module retrieving image data from adjacent frames to complete the feature matching operation for the occluded areas; and vegetation feature completion and repair are then completed. The interpolation module fills in pixels based on the matching results, restoring the vegetation feature information of the occluded areas.

[0026] Step S3, targetless spatial scale calibration, includes: feature point matching and bundle adjustment of the image sequence; the feature point calculation module performs corresponding point matching on the image sequence and completes bundle adjustment calculation; constructing a 3D sparse point cloud of the sample plot and calculating the exterior orientation elements of the image; the 3D reconstruction module generates point cloud data based on the matching results and calculates the exterior orientation parameters of the image; combining the forestry and grassland prior knowledge base to complete the absolute scale calculation; the scale calculation module retrieves prior knowledge base data and completes the absolute scale assignment of the point cloud data; establishing a rigid coordinate mapping; the coordinate mapping module combines positioning data to establish the transformation relationship between multiple coordinate systems; and generating global scale conversion parameters. The parameter generation module outputs the scale conversion parameters for subsequent size calculation steps.

[0027] Step S4, multi-scale segmentation, includes: employing a multi-scale threshold segmentation and contour optimization algorithm; the segmentation algorithm module loads a preset algorithm model and performs multi-scale segmentation operations on the input image; setting a hierarchical detection window, with the window configuration module pre-setting hierarchical window parameters according to vegetation type to adapt to vegetation targets at different scales; performing feature extraction and segmentation on the corresponding structures of trees, shrubs, and grasses; the feature extraction module completes feature extraction and segmentation operations for the corresponding structures according to vegetation type; performing edge optimization and target separation on overlapping vegetation targets; the edge optimization module optimizes the contours of overlapping areas to separate different targets; and outputting the segmented region and contour key points. The data output module outputs the segmented region data and contour key point data for subsequent steps.

[0028] Step S5, species classification, includes: extracting geometric morphology and texture spectrum from the segmented vegetation targets; the feature extraction module extracting feature information for the corresponding dimensions of the segmented vegetation targets; constructing a multi-dimensional feature parameter set; the parameter set construction module integrating the extracted features according to preset rules to generate a standardized feature parameter set; using a multi-feature weighted fusion algorithm to generate a comprehensive feature vector; the feature fusion module performing multi-feature fusion according to preset weights to generate a comprehensive feature vector; inputting a pre-constructed classification model; the model operation module loading the pre-trained classification model to complete the input and operation of the feature vector; and outputting the species classification result and confidence score. The result output module outputs the classification result and corresponding confidence score for use in subsequent inversion steps.

[0029] Step S5, resource parameter inversion, includes: calculating core parameters such as tree height, diameter at breast height (DBH), and volume for trees; the tree parameter calculation module, combined with scale conversion parameters, completes the calculation of corresponding tree parameters. For shrubs, calculating parameters such as clump width, clump height, and biomass; the shrub parameter calculation module, combined with scale conversion parameters, completes the calculation of corresponding shrub parameters. For herbs, calculating parameters such as canopy coverage, height, and frequency; the herb parameter calculation module, combined with scale conversion parameters, completes the calculation of corresponding herb parameters. Based on feature outlier identification, the outlier identification module compares with a standard feature library to complete the identification calculation of vegetation feature outliers. Finally, it outputs the vegetation health stress level results. The health level output module, based on the outlier identification results, outputs the corresponding vegetation health stress level.

[0030] Step S5 also includes an accuracy optimization step, which involves comparing the output classification results with resource parameters and field measurement data from sample plots. The accuracy comparison module retrieves the measured verification data and completes the error comparison calculation of the output results. When the error exceeds a preset threshold, the corresponding data is used as an incremental sample, and the sample selection module completes the selection and organization of incremental samples according to the error threshold. The algorithm training set is then added for iterative optimization. The model iteration module adds the incremental samples to the training set to complete the iterative training and optimization of the model, thereby improving recognition accuracy. The model update module loads the optimized model to replace the original model, ensuring a stable improvement in recognition accuracy throughout the entire process.

[0031] In step S1, the data preprocessing module receives the acquired multi-view sequence images, positioning data, attitude parameters, and environmental terrain data, and completes the data format unification and organization. Millisecond-level time synchronization and pixel-level spatial registration are performed, with the spatiotemporal registration module completing the time synchronization and spatial registration of the multi-source data according to preset rules. This ensures the consistency of the spatiotemporal reference of the multi-source data. The registration verification module verifies the registration results, ensuring the consistency and accuracy of the spatiotemporal reference of the multi-source data.

[0032] In step S4, based on the vegetation type, shooting distance, and ambient lighting conditions of the forest and grassland plots to be identified, the scene parameter acquisition module receives the scene parameter data of the plots and completes the parameter processing and analysis. The detection threshold and window size of the segmentation algorithm are adaptively adjusted. The parameter adaptive adjustment module adjusts the algorithm threshold and window size based on the scene parameters to adapt to the recognition needs of different field scenes. The adjusted algorithm parameters are loaded into the segmentation module in real time to ensure segmentation results under different scenarios.

[0033] In step S5, the output forestry and grassland resource survey results are standardized and coded according to the national forestry and grassland resource inventory standards. The standardization coding module loads the national standard coding rules and completes the standardization coding of the survey results. Spatial vector data and statistical reports that can be directly entered into the database are generated. The data generation module generates corresponding spatial vector data and statistical reports based on the coding results. Batch archiving and summary statistics of multiple batches of sample plot data are supported. The data management module completes the batch archiving, summary statistics, and storage of multiple batches of data.

[0034] The functional principle of this invention can be explained through the following operational methods: In use, the image acquisition unit first acquires multi-view sequence images of the target forest and grassland plots. The positioning attitude acquisition unit simultaneously acquires the real-time positioning data, shooting attitude parameters and environmental site basic data of the plots. The acquired multi-source data is synchronized at the millisecond level and spatially registered at the pixel level. Then, the image processing unit performs unified preprocessing on the sequence images to complete illumination correction, background suppression, occlusion completion and noise filtering to obtain standardized forest and grassland images.

[0035] Secondly, based on the matching feature points, localization and pose data of the sequence images, the scale calibration unit establishes a global mapping between pixels and geospatial coordinates by combining the prior knowledge base of forest and grassland vegetation, and completes the targetless spatial scale calibration. Then, the standardized image is segmented at multiple scales by the image segmentation unit, and the algorithm threshold and window size are adaptively adjusted to extract the contours and feature parameters of different types of vegetation targets. The forest and grassland species classification and resource parameter inversion are completed by the identification and calculation unit.

[0036] Finally, the model iteration unit compares the output classification results with the resource parameters and the actual measured data of the sample plots to complete the iterative optimization of the algorithm model. The data management unit standardizes and encodes the output forest and grassland resource survey results according to the national forest and grassland resource inventory standards, generating corresponding spatial vector data and statistical reports. This enables large-scale and efficient surveys of forest and grassland resources in the field, expands the application scenarios of image analysis technology, and improves the overall efficiency of forest and grassland resource survey operations.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for accurate identification of forest and grassland resources based on image analysis, characterized in that, Includes the following steps: Step S1: Collect multi-view sequence images of the target forest and grassland plots, and simultaneously acquire real-time positioning data, shooting posture parameters and environmental site basic data corresponding to the plots; Step S2: Perform unified preprocessing on the acquired sequence images, sequentially completing illumination normalization correction, complex background suppression, vegetation occlusion area completion, and noise filtering to obtain standardized forest and grassland images; Step S3: Based on the matching feature points, localization and pose data of the sequence images, combined with the prior knowledge base of forest and grassland vegetation, establish a global mapping between pixels and geospatial coordinates to complete the targetless spatial scale calibration. Step S4: Perform multi-scale segmentation on the standardized image to extract the contours and feature parameters of different types of vegetation targets, including trees, shrubs, and herbs; Step S5: Based on the extracted feature parameters, complete the classification of forest and grassland species, inversion of resource parameters, and determination of health level, and output the forest and grassland resource survey results matching geographic coordinates.

2. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, The complex background suppression and occlusion region completion in step S2 include: identifying non-vegetation backgrounds and vegetation occlusion regions in the image through semantic segmentation, performing mask suppression on the non-vegetation background, and using adjacent frame feature matching and interpolation algorithms on the occlusion regions to complete vegetation feature completion and repair.

3. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, The targetless spatial scale calibration in step S3 includes: performing feature point matching and bundle adjustment on the sequence images, constructing a three-dimensional sparse point cloud of the sample plot and solving the image exterior orientation elements, combining the forestry and grassland prior knowledge base to complete the absolute scale calculation, establishing a coordinate rigid mapping, and generating global scale conversion parameters.

4. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, The multi-scale segmentation in step S4 includes: using a multi-scale threshold segmentation and contour optimization algorithm, setting a hierarchical detection window, extracting and segmenting features for the corresponding structures of trees, shrubs and grasses respectively, performing edge optimization and target segmentation for overlapping vegetation targets, and outputting segmented regions and contour key points.

5. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, The species classification in step S5 includes: extracting geometric morphology and texture spectrum from the segmented vegetation targets, constructing a multi-dimensional feature parameter set, generating a comprehensive feature vector using a multi-feature weighted fusion algorithm, inputting it into a pre-constructed classification and discrimination model, and outputting the species classification result and confidence level.

6. The method for accurate identification of forest and grassland resources based on image analysis according to claim 5, characterized in that, The resource parameter inversion in step S5 includes: calculating the core parameters of tree height, diameter at breast height, and volume for trees; calculating the parameters of clump width, clump height, and biomass for shrubs; calculating the parameters of cover, height, and frequency for herbs; identifying outliers based on features; and outputting the vegetation health stress level results.

7. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, Step S5 further includes an accuracy optimization step, which involves comparing the output classification results with resource parameters and field measured data of the sample plots. When the error exceeds a preset threshold, the corresponding data is used as an incremental sample and added to the algorithm training set for iterative optimization to improve the recognition accuracy.

8. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, In step S1, the collected multi-view sequence images, positioning data, attitude parameters and environmental ground data are synchronized at the millisecond level and spatially registered at the pixel level to ensure the spatiotemporal reference of the multi-source data is unified.

9. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, In step S4, based on the vegetation type, shooting distance and ambient lighting conditions of the forest and grassland plot to be identified, the detection threshold and window size of the segmentation algorithm are adaptively adjusted to meet the identification needs of different field scenes.

10. The method for accurate identification of forest and grassland resources based on image analysis according to claim 1, characterized in that, In step S5, the output forest and grassland resource survey results are standardized and coded according to the national forest and grassland resource inventory standards to generate spatial vector data and statistical reports that can be directly entered into the database, supporting batch archiving and summary statistics of multiple batches of sample plot data.

Images