Cypress pine moth recognition method based on image fusion

Abstract

The present application relates to the technical field of forestry pest remote sensing monitoring, and discloses a cypress pine moth identification method based on image fusion, which combines a wavelet fusion algorithm dynamically adjusted according to insect state and canopy density, introduces a special feature model for undergrowth interference and tree age correction, and a fusion weight dynamically optimized based on the analytic hierarchy process and measured data, to build a highly adaptive and accurate intelligent analysis core, which can ensure that the present application can still achieve a high-precision index of a larva identification rate of 90% or above and an insect state and damage level determination accuracy of 95% or above when facing different tree ages, different tree species and complex canopy structures, fundamentally solving the problems of early cypress pine moth pest misjudgment, misjudgment and poor scene adaptability, and finally realizing the whole-process accurate identification of early cypress pine moth pests in complex habitats.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology for forest pests and diseases, specifically to a method for identifying the Chinese cypress tussock moth based on image fusion. Background Technology

[0002] The Chinese cypress tussock moth (Parocneria furva), a specific leaf-eating pest of cypress trees, poses a devastating threat to precious tree species such as ancient cypress trees. During its voracious feeding phase, the larvae can consume over 90% of the leaves in a short period, leading to photosynthetic exhaustion and even death of the ancient cypress. Most of my country's existing ancient cypress trees are distributed in complex terrain areas such as the Qinling and Daba Mountains. Traditional manual monitoring suffers from three major drawbacks: ① It relies on visual inspection, with a single tree inspection taking more than 30 minutes, requiring several months to complete for thousands of acres of forest; ② It can only detect visible symptoms, missing early-stage Chinese cypress tussock moth infestations (L1-L2 larval stages); ③ Frequent human activity disrupts the habitat of ancient cypress trees. Although image fusion technology has been used for monitoring forest pests and diseases, existing solutions have significant limitations: (1) General algorithms are not adapted to the characteristics of high canopy closure (>0.8) and strong interference from understory vegetation in ancient cypress, resulting in a missed detection rate of up to 45% for pests in the lower canopy; (2) Fixed weighting mechanisms ignore the characteristics of insect evolution, with a misjudgment rate of over 30%; (3) Feature models lack tree age correction terms, using the same threshold for saplings under 100 years old and ancient trees over 1000 years old, resulting in poor adaptability; (4) Data processing links are fragmented, requiring 7-15 days from data collection to decision-making, missing the optimal window for prevention and control. Therefore, it is urgent to develop an intelligent monitoring system specifically for the ancient cypress ecosystem. Summary of the Invention

[0003] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an image fusion-based method for identifying the Chinese cypress tussock moth, which has the advantages of high recognition accuracy, strong scene adaptability, outstanding monitoring timeliness, minimal interference with ancient cypress trees, and strong decision support. It solves the problems of low early recognition accuracy of existing methods for the Chinese cypress tussock moth, poor adaptability to complex ancient cypress habitats, delayed monitoring and control decision-making cycles, and potential interference with precious ancient trees.

[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for identifying the Chinese pine tussock moth based on image fusion, comprising the following steps: Step 1: Data Acquisition: Based on the multi-source remote sensing collaborative observation strategy for ancient cypress ecological monitoring, drones equipped with various sensors are used to collect time-series data. Step 2, Preprocessing: The collected multi-source remote sensing data are sequentially subjected to radiometric calibration, atmospheric correction, image registration, outlier removal, and stratification to form a standardized dataset; Step 3: Image fusion specific to the Chinese cypress tussock moth: The db4 wavelet basis wavelet transform fusion algorithm is used to fuse information from multiple sources in a targeted manner; Step 4: Quantitative and Dynamic Adjustment of Fusion Weights: Construct a hierarchical analysis model to determine the basic weights, and dynamically adjust them in combination with field observation data to achieve accurate quantification and dynamic optimization of the fusion weights of multi-source data; Step 5: Image quality control: Establish a closed-loop quality control process to provide high-quality data sources for subsequent feature extraction and recognition; Step Six: Feature Extraction from Fuded Images: Based on the fused images, four core feature indices are calculated to construct a chlorophyll content inversion model and mine quantitative characteristics of pest damage. Step 7: Multi-dimensional feature comprehensive identification and hazard assessment: Based on the extracted features, a streamlined comprehensive judgment is performed to identify insect stage, instar, and hazard level; Step 8: Output Results and Platform Integration: Output the identification results in a formatted format and automatically synchronize them to the management platform to form a monitoring-management closed loop.

[0005] Preferably, the data acquisition in step one includes: (1) Collect multispectral data by using a camera mounted on a drone, controlling the flight altitude to be 80-85m and the flight speed to be 4-5m / s, and collect data 2-3 times per month; (2) Collect thermal infrared data using a FLIR Vue Pro R camera, control the flight altitude to 80-85m and the flight speed to 2.5-3.5m / s, and collect data 1-2 times per week during the active period of larvae; (3) Data was collected by using Velodyne VLP-16 to collect lidar data, with the flight altitude controlled at 80-85m and the speed at 3-4m / s, and data was collected 1-2 times per quarter; (4) Collect radar data through C / X dual-band equipment, control the flight altitude to 95-100m and the speed to 5-6m / s, and collect data synchronously with lidar.

[0006] Preferably, the preprocessing procedure in step two is as follows: S1.1 Radiometric calibration: Radiometric calibration is performed using the MODTRAN 6.0 model, and the calibration error is ≤2.5% using an ASD FieldSpec 4 spectrometer. The raw signal received by the sensor is converted into radiance values. S1.2 Atmospheric Correction: Atmospheric correction for the FLAASH model, inputting aerosol optical thickness and water vapor content in the monitoring area, the difference between the corrected reflectance and the measured value is ≤3.5%; S1.3 Image registration: Based on ground control points, the quadratic polynomial registration method combined with the RANSAC algorithm is used to remove outliers, with a registration error of ≤0.32 pixels, achieving accurate spatial alignment of multi-source data; S1.4 Outlier Removal and Stratification: Outliers from multi-source data are removed according to corresponding standards; the LiDAR point cloud elevation threshold method is used to separate the understory vegetation and the ancient cypress canopy, with a separation accuracy of ≥93%, forming a standardized dataset.

[0007] Preferably, the specific image fusion process for the *Platycladus orientalis* moth in step three is as follows: S2.1 Optimization of Fusion Algorithm: The wavelet transform fusion algorithm based on db4 wavelet basis is adopted, and the number of decomposition layers is dynamically adjusted according to the state of the Chinese cypress tussock moth and the canopy closure of the ancient cypress. S2.2 Component Adaptive Fusion: Low-frequency components are weighted and averaged using dynamic weights based on insect state and canopy position; high-frequency components are fused using a combination of the absolute value maximization method and the gradient constraint method, and canopy position weights are introduced to enhance edge and texture information. S2.3 Image post-processing: The CLAHE algorithm is used to enhance the contrast of bands sensitive to pests and diseases after the fused image is processed, and then Gaussian filtering is used for smoothing and noise reduction.

[0008] Preferably, step four integrates weight quantification and dynamic adjustment: a three-level weight system is constructed—target layer, criterion layer, and scheme layer—and the basic weights are calculated using the analytic hierarchy process (AHP). =0.0381, where the criterion layer =0.545、 =0.297、 =0.158, Scheme Layer =0.42、 =0.28、 =0.20、 =0.10; combined with the insect stage adjustment coefficient Adjustment coefficient for canopy position The final weight is calculated using the following formula. : In the formula, For the final weight, Based on weights, This is the insect stage adjustment coefficient. This is the canopy position adjustment factor.

[0009] Preferably, the quality control process of the fused image in step five is executed in the following order: wavelet decomposition → component fusion → inverse transform → image enhancement → quality verification → result generation, forming a closed-loop control. When the quality standard is not met, the process returns to the front-end step for adjustment, and finally a fused image in TIFF format, WGS84 coordinate system, and pixels is generated. In the quality verification index, spectral fidelity is calculated by the difference in band reflectance between the fused image and the original multispectral image, so as to realize the quantitative monitoring of the degree of spectral distortion introduced by the fusion process.

[0010] Preferably, the spectral fidelity is calculated by the difference in band reflectance between the fused image and the original multispectral image, and the calculation formula is as follows: In the formula, Indicates spectral fidelity. Indicates the fused reflectivity. This represents the original reflectance.

[0011] Preferably, the calculation of the four core characteristic indices and the inversion of chlorophyll content in step six are as follows: (1) Vegetation index The calculation formula is: In the formula, For near-infrared reflectivity, Reflectivity in the red light band To determine the background reflectance of the understory vegetation, an understory interference correction factor of 0.1 is introduced. (2) Enhanced vegetation index The calculation formula is: In the formula For near-infrared reflectivity, For red light band reflectivity, Background reflectance of forest understory vegetation. For blue light band reflectivity; -7.5× in the denominator The design is optimized based on the physiological characteristic of a significant decrease in blue light reflectance after damage by the Chinese tussock moth, and its model fit is good. =0.91, optimized for the characteristic of reduced blue light reflectivity after damage; (3) Canopy temperature stress index The calculation formula is: In the formula, For the target canopy temperature, The average temperature of a healthy canopy. For the standard deviation of healthy canopy temperature, when >1.2; (4) Spectral index of specific diseases and pests The calculation formula is: In the formula, The reflectance of the damaged canopy, For healthy canopy reflectivity, This refers to the band weighting coefficient. This represents the percentage of the affected area. (5) Chlorophyll content The inversion model is as follows: In the formula, The vegetation index, To enhance the vegetation index, This is the tree age correction factor. It is an ancient cypress tree. The value was obtained by fitting measured data of ancient cypress trees aged 100-2000 years, with a range of 0.002-0.005. The value increases by 100 years for each additional 100 years of tree age. With the value increasing by 0.001, after introducing the tree age correction term, the model R²≥0.89 and RMSE≤1.25mg / g.

[0012] Preferably, step seven involves multi-dimensional feature comprehensive identification and hazard assessment: S3.1 Identification process control: The process is executed in the following order: temperature anomaly screening → interference removal → insect stage determination → instar determination → hazard level determination → manual verification of abnormal samples, forming a standardized identification link to ensure that each link is accurate and controllable; S3.2 Determination of Insect Stages and Instars: Insect stage determination is combined with changes in reflectance in sensitive wavelength bands. value, Values ​​and canopy location, multi-dimensional collaborative determination improves accuracy; age determination is based on... Based on the value and leaf damage rate Si, accurate differentiation of larvae at different ages can be achieved; S3.3 Hazard Level Classification: Hazard levels are classified according to chlorophyll content, The damage is classified into four levels based on the value and the percentage of affected area, which quantitatively distinguishes the degree of damage and provides a basis for targeted prevention and control.

[0013] Preferably, the output of the results in step eight is integrated with the platform: the hierarchical pest and disease location map, standardized monitoring report and visualization results are output in the ancient cypress status visualization platform, so that the location accuracy of the location map is ≤3m, and the distribution and damage of pests are presented intuitively; and the data is synchronized to the ancient cypress health management platform through the MQTT communication protocol, with the data format being JSON, the transmission delay being ≤3.2 seconds, so as to achieve fast data synchronization; the data processing cycle of the whole process is controlled to be ≤36 hours, and the single batch processing efficiency is ≥120 plants / minute.

[0014] Compared with existing technologies, this invention provides a method for identifying the Chinese pine tussock moth based on image fusion, which has the following beneficial effects: 1. This invention constructs a highly adaptive and accurate analysis core by combining a wavelet fusion algorithm that dynamically adjusts insect life stages and canopy closure, introducing a dedicated feature model for forest understory interference and tree age correction, and using fusion weights dynamically optimized based on the analytic hierarchy process and measured data. This core ensures that the method of this invention can still stably achieve high-precision indicators such as a ≥90% identification rate of young larvae and an ≥95% accuracy rate in determining insect life stages and damage levels when facing different tree ages (100-2000 years), different tree species (Ancient Cypress, Chinese Arborvitae, Juniperus chinensis), and complex canopy structures. This fundamentally solves the problems of missed detection, misjudgment, and poor scene adaptability of early-stage Cypress tussock moth infestation, and achieves accurate identification of early-stage Cypress tussock moth infestation in complex habitats throughout the entire process.

[0015] 2. This invention designs a closed-loop processing chain from multi-source data collaborative acquisition, automated preprocessing, specific fusion, quality control to feature extraction and judgment. This chain improves the single-batch processing efficiency to ≥120 plants / minute through algorithm optimization, and compresses the entire cycle to ≤36 hours, significantly improving the timeliness of the method. At the same time, the terminal of the method process automatically outputs standardized and visualized spatial distribution maps and reports, and seamlessly connects with the business platform through the MQTT protocol, directly transforming high-precision identification results into actionable prevention and control basis, realizing rapid transformation and strong support from data to decision-making, thereby constructing an efficient closed-loop integrated intelligent process of monitoring-identification-decision.

[0016] 3. The entire implementation process of the method of this invention relies entirely on the UAV remote sensing platform for contactless data collection, completely avoiding the physical contact and potential interference of traditional manual survey methods on precious ancient cypress trees. This remote sensing perception + cloud analysis mode can ensure monitoring accuracy and timeliness while minimizing the impact on the monitored objects and their ecological environment. It fully meets the special ethical and environmental protection requirements for the protection of ancient and famous trees, and provides a reliable technical paradigm for the sustainable and low-interference health monitoring of precious ecological and cultural heritage. Ultimately, it establishes a completely non-invasive health monitoring paradigm for ancient and famous trees. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 The image fusion-based method for identifying the Chinese pine tussock moth includes the following steps: Step 1: Data Acquisition: Based on the multi-source remote sensing collaborative observation strategy for ancient cypress ecological monitoring, drones equipped with various sensors are used to collect time-series data. Step 2, Preprocessing: The collected multi-source remote sensing data are sequentially subjected to radiometric calibration, atmospheric correction, image registration, outlier removal, and stratification to eliminate data interference and form a standardized dataset. Step 3: Image fusion of *Symplocos spp.* specificity: The db4 wavelet basis wavelet transform fusion algorithm is used to fuse information from multiple sources in a targeted manner to enhance the identification of *Symplocos spp.* damage characteristics, improve the practicality of the fused images, and thus highlight the damage characteristics of *Symplocos spp.* Step 4: Quantitative and dynamic adjustment of fusion weights: Construct a hierarchical analysis model to determine the basic weights, and dynamically correct them in combination with field observation data to achieve accurate quantification and dynamic optimization of the fusion weights of multi-source data, thereby making the method of this invention adaptable to different monitoring scenarios; Step 5: Quality control of fused images: Establish a closed-loop quality control process, and ensure that the quality of fused images meets the standards through multi-dimensional verification, so as to provide a high-quality data source for subsequent feature extraction and recognition; Step Six: Feature Extraction from Fuded Images: Based on the fused images, four core feature indices are calculated to construct a chlorophyll content inversion model, mining quantitative characteristics of pest damage to provide data support for accurate identification. Step 7: Multi-dimensional Feature Comprehensive Identification and Hazard Assessment: Based on the extracted features, a streamlined comprehensive judgment is performed to achieve accurate identification of insect stage, instar, and hazard level; Step 8: Output Results and Platform Integration: Output the identification results in a formatted format and automatically synchronize them to the management platform to form a monitoring-management closed loop.

[0020] Specifically, in step one, data acquisition is as follows: multispectral data is collected using a Headwall Nano-Hyperspec camera and a DJI Motrice 300RTK drone, flying at an altitude of 80m and a speed of 5m / s, twice a month; thermal infrared data is collected using a FLIR VuePro R camera, flying at an altitude of 80m and a speed of 3m / s, once a week during the larval active period; lidar data is collected using a Velodyne VLP-16, flying at an altitude of 80m and a speed of 4m / s, once a quarter; and radar data is collected synchronously with lidar using a C / X dual-band device, flying at an altitude of 100m and a speed of 6m / s.

[0021] The advantages are: the above steps, by formulating a multi-source remote sensing collaborative acquisition strategy for monitoring the ecology of ancient cypress trees, can obtain raw data reflecting the multi-dimensional characteristics (spectral, temperature, structure, dielectric properties) of the damage caused by the tussock moth in the Chinese cypress, laying a data foundation for subsequent high-precision identification. At the same time, the non-contact remote sensing mode can significantly reduce the interference to the ancient cypress trees. Specifically, the preprocessing procedure in step two is as follows: S1.1 Radiometric calibration: Radiometric calibration is performed using the MODTRAN 6.0 model. The calibration error is ≤2.5% using an ASD FieldSpec 4 spectrometer. This converts the raw signal received by the sensor into a physically meaningful radiance value, eliminating the sensor's own error. S1.2 Atmospheric Correction: FLAASH model atmospheric correction, input aerosol optical thickness and water vapor content of the monitoring area, after correction the difference between reflectance and measured value ≤3.5%, eliminate interference factors such as atmospheric scattering and absorption, restore the true reflectance of the surface; S1.3 Image Registration: Based on ground control points, the quadratic polynomial registration method combined with the RANSAC algorithm is used to remove outliers. The registration error is ≤0.32 pixels, which realizes accurate spatial alignment of multi-source data and ensures the subsequent fusion effect. S1.4 Outlier Removal and Stratification: Outliers from multiple data sources are removed according to the corresponding standards; the lidar point cloud elevation threshold method (the lowest elevation value of the ancient cypress canopy is statistically analyzed from lidar point cloud data (after removing the understory vegetation point cloud), and 1.5m is subtracted from this as the elevation threshold for separating understory vegetation) is used to separate understory vegetation from the ancient cypress canopy, with a separation accuracy of ≥93%, forming a standardized dataset, effectively removing understory vegetation interference, and focusing on the ancient cypress canopy target.

[0022] The advantages are: through precise preprocessing and error control in multiple stages, the data quality is greatly improved, and the registration error and separation accuracy both reach high standards, providing reliable data support for subsequent fusion and recognition; the stratified processing of forest vegetation can reduce interference factors and improve the accuracy of subsequent recognition, while the standardized processing process lays the foundation for batch data processing and helps to improve the overall timeliness. Specifically, the image fusion process for the *Symplocos spp.* specific to *Symplocos spp.* in step three: S2.1 Optimization of Fusion Algorithm: The wavelet transform fusion algorithm based on db4 wavelet basis is adopted, and the number of decomposition layers is dynamically adjusted according to the stages of the Chinese cypress tussock moth (egg, larva, pupa, adult) and the canopy closure of the ancient cypress. S2.2 Component Adaptive Fusion: Low-frequency components are weighted and averaged using dynamic weights based on insect state and canopy position (sunny side / shady side / top canopy); High-frequency components are combined with the absolute value maximization method and gradient constraint method, and canopy position weights are introduced to enhance edge and texture information. S2.3 Image post-processing: The CLAHE algorithm is used to enhance the contrast of bands sensitive to pests and diseases after the fused image is processed, and then Gaussian filtering is used for smoothing and noise reduction. The advantages are: the above algorithms and component processing are dynamically adjusted by combining insect state and canopy features, and are highly targeted, which can accurately enhance the insect pest sensitive features and provide a guarantee for subsequent high-precision identification; through the synergistic processing of enhancement and noise reduction after fusion, it can not only improve the feature recognition, but also avoid misjudgment caused by noise interference, ultimately helping to reduce the misjudgment rate and missed judgment rate.

[0023] Specifically, step four integrates weight quantification and dynamic adjustment: a three-level weight system is constructed—target layer, criterion layer, and solution layer—and the basic weights are calculated using the analytic hierarchy process (AHP). =0.0381), where the criterion layer =0.545、 =0.297、 =0.158, Scheme Layer =0.42 (multispectral) =0.28 (thermal infrared) =0.20 (LiDAR) =0.10 (radar); combined with insect state adjustment coefficient Adjustment coefficient for canopy position (All data are based on 12 months of measured data (Jianmenguan Cuiyun Corridor Experimental Area), obtained through multiple linear regression analysis, with R² ≥ 0.92, and...) , The four values ​​each correspond to the scheme layer in turn. - ), through the formula: In the formula, For the final weight, Based on weights, This is the insect stage adjustment coefficient. This is the canopy position adjustment factor; The advantages are: by combining a three-level weighting system with the analytic hierarchy process, the basic weight allocation is ensured to be scientific and reasonable; among them, the dynamic adjustment coefficient is based on long-term measured data fitting to adapt to different insect stages and canopy positions, improving the adaptability of the method to complex scenarios. At the same time, the weights can be accurately quantified to help improve the recognition accuracy. By introducing the above dynamic adjustment mechanism, the fusion strategy can adapt to different insect stages and different growth environments (such as tree age and canopy position) of ancient cypress, solving the problem of poor adaptability of fixed parameters, thus supporting the monitoring of ancient cypress trees aged 100-2000 years and various cypress species.

[0024] Specifically, the quality control process for the fused image in step five follows a sequence of wavelet decomposition → component fusion → inverse transform → image enhancement → quality verification → result generation, forming a closed-loop control. If the quality standards are not met, the process returns to the previous steps for adjustment. The final result is a 0.8m × 0.8m pixel fused image (containing 6 bands) in TIFF format, WGS84 coordinate system, adapted to subsequent processing and application requirements. In the quality verification indicators, spectral fidelity is calculated using the difference in band reflectance between the fused image and the original multispectral image. This quantitatively monitors the degree of spectral distortion introduced during the fusion process, ensuring that the fused image maintains the spectral consistency of the original multispectral image, providing a reliable data foundation for subsequent pest feature inversion. This calculated value ensures that the fused image retains the original spectral features, guaranteeing the accuracy of feature extraction. The calculation formula is as follows: In the formula, Indicates spectral fidelity. Indicates the fused reflectivity. Indicates the original reflectivity; The advantages are: through the above closed-loop quality control process, unqualified images can be corrected in a timely manner, effectively avoiding the impact of low-quality images on subsequent results and ensuring recognition accuracy; by standardizing image formats and coordinate systems, the universality of the results can be improved, and spectral fidelity verification can ensure the authenticity of features, which can lay the foundation for accurately extracting pest and disease related indices and help improve the overall recognition reliability.

[0025] Specifically, step six involves the calculation of the four core characteristic indices and the inversion of chlorophyll content: (1) Vegetation index The calculation formula is: In the formula, For near-infrared band (800nm) reflectance, Reflectivity in the red light band (650nm) To adjust the background reflectance of the understory vegetation, an understory interference correction coefficient of 0.1 was introduced, which effectively reduced the interference from the understory vegetation. (2) Enhanced vegetation index The calculation formula is: In the formula For near-infrared band (800nm) reflectance, Reflectivity in the red light band (650nm) Background reflectance of forest understory vegetation. Reflectance in the blue light band (450nm); -7.5× in the denominator The design is optimized based on the physiological characteristic of a significant decrease in blue light reflectance after damage by the Chinese tussock moth, and its model fit is good. =0.91, optimized for the characteristic of reduced blue light reflectivity after damage; (3) Canopy temperature stress index The calculation formula is: In the formula, For the target canopy temperature, The average temperature of a healthy canopy. For the standard deviation of healthy canopy temperature, when When the value is greater than 1.2, it is considered an abnormal temperature and is used to detect abnormal canopy temperature caused by pests (PI>1.2 is the threshold). (4) Spectral index of specific diseases and pests The calculation formula is: In the formula, The reflectance of the damaged canopy, For healthy canopy reflectivity, This refers to the band weighting coefficient. The percentage of damaged area is calculated by combining the reflectance of damaged and healthy canopies and the percentage of damaged area, along with the spectral and area information of damaged and healthy pixels. (5) Chlorophyll content The inversion model is as follows: In the formula, The vegetation index, To enhance the vegetation index, This is the tree age correction factor. It is an ancient cypress tree. The value was obtained by fitting measured data of ancient cypress trees aged 100-2000 years, with a range of 0.002-0.005. The value increases by 100 years for each additional 100 years of tree age. With the value increasing by 0.001, after introducing the tree age correction term, the model's R² ≥ 0.89 and RMSE ≤ 1.25 mg / g; The advantages are: the calculation formulas for the above four types of core feature indices all incorporate targeted correction terms and optimized parameters, which can be adapted to the physiological characteristics of the tussock moth damage and the growth characteristics of ancient cypress, achieving accurate quantification of the physiological, biochemical and physical changes caused by pests and diseases; by introducing the tree age correction term, the method can be adapted to ancient cypresses of different ages, greatly improving adaptability; at the same time, the high-fit inversion model makes the insect stage and damage level determination reliable, directly supporting an accuracy rate of ≥95% for insect stage and damage level determination, thereby helping to improve the identification accuracy.

[0026] Specifically, step seven involves multi-dimensional feature comprehensive identification and hazard assessment: S3.1 Identification process control: The process is executed in the following order: temperature anomaly screening → interference removal → insect stage determination → instar determination → hazard level determination → manual verification of abnormal samples, forming a standardized identification link to ensure that each link is accurate and controllable; S3.2 Determination of Insect Stages and Instars: Insect stage determination is combined with changes in reflectance in sensitive wavelength bands. value, Values ​​and canopy location, multi-dimensional collaborative determination improves accuracy; age determination is based on... Based on the value and leaf damage rate Si, accurate differentiation of larvae at different ages can be achieved; S3.3 Hazard Level Classification: Hazard levels are classified according to chlorophyll content, The damage is classified into four levels based on the value and the percentage of affected area, which quantitatively distinguishes the degree of damage and provides a basis for targeted prevention and control.

[0027] The advantages are: through the standardized identification process and the synergy of multi-dimensional judgment indicators, the identification rate of young larvae is ≥90%, the accuracy rate of insect stage judgment is ≥95%, and the accuracy rate of hazard level judgment is ≥95%, which greatly reduces the false judgment rate and the missed judgment rate, and can accurately capture early-stage tussock moth pests; its manual verification process can further verify the results, thereby improving the reliability of identification and solving the problem of insufficient identification accuracy in existing technologies.

[0028] Specifically, in step eight, the results are output and integrated with the platform: the tiered pest and disease location map (with four colors indicating different levels of damage), standardized monitoring report, and visualization results (feature comparison map, identification heat map) are output to the cypress status visualization platform, ensuring that the location accuracy of the location map is ≤3m, and intuitively presenting the distribution and damage of pests; and the data is synchronized to the cypress health management platform via the MQTT communication protocol, with the data format being JSON and the transmission delay being ≤3.2 seconds, achieving rapid data synchronization; the entire data processing cycle is ≤36 hours, and the processing efficiency of a single batch (1000 plants) is ≥120 plants / minute, greatly improving the monitoring timeliness.

[0029] The advantages are: the visualized and standardized results make it easy to intuitively understand the pest situation; the connection with the management platform enables rapid data flow, providing precise technical support for targeted prevention and control, reducing pest losses and control costs; the efficient data processing and transmission shortens the entire process cycle, avoids missing the best prevention and control window, solves the problem of insufficient timeliness of traditional monitoring, and the non-contact monitoring mode has minimal impact on ancient cypress trees throughout the process, meeting the low-interference and environmental protection requirements for the protection of ancient and famous trees.

[0030] To verify the effectiveness and superiority of the method of the present invention, three representative experimental areas were selected in the ancient cypress grove of Cuiyun Corridor in Jianmenguan, Sichuan Province for implementation and verification. Three control groups were set up in the same experimental area to implement existing typical technology comparison schemes, as detailed below: Example 1 (Early pest monitoring in areas with scattered ancient cypress trees) Scenario: Monitoring the initial damage caused by young larvae (1st to 3rd instar) on ancient cypress trees aged 100-500 years and with relatively sparse distribution.

[0031] Application of this invention: Employing dynamically adjusted fusion weights (emphasizing multispectral and thermal infrared data), and enabling understory disturbance correction terms and tree age correction coefficients (…). =0.002-0.005), execute the complete closed-loop process.

[0032] Objective: To verify the ability of this invention to capture early and subtle insect pest characteristics and its high-precision identification effect under complex backgrounds.

[0033] Example 2 (Monitoring of pest outbreaks in densely populated areas of ancient cypress) Scenario: Monitoring for outbreaks of damage caused by middle- to high-age larvae (4th to 6th instar) in ancient cypress forests with trees aged 500-1000 years and high canopy closure.

[0034] Application of this invention: The fusion algorithm is dynamically adjusted to a deeper wavelet decomposition, with weights tilted towards thermal infrared and lidar (canopy structure), and the tree age correction coefficient... Take the median value and amplify temperature anomalies. ) and specific pest and disease indices ( The judgment weight of ).

[0035] Objective: To verify the stability and accuracy of the identification and level determination of the present invention in high canopy density and severely hazardous scenarios.

[0036] Example 3 (Monitoring of a mixed forest of ultra-old ancient cypress and Chinese arborvitae) Scenario: Includes ancient cypress trees over 1500 years old and adjacent cypress groves, with mixed pest infestations.

[0037] Application of this invention: Employing a full-parameter dynamic adjustment mode, tree age correction coefficient The upper limit (0.005) is used to verify the adaptability of the feature index model to different cypress species.

[0038] Objective: To verify the adaptability of this invention to ancient cypress trees with extremely long lifespans and its ability to be extended to monitoring other cypress species.

[0039] Comparative Example 1 (Fixed Weight Fusion Method) Methods: Conventional IHS or PCA fusion methods were used with fixed fusion weights (e.g., multispectral: thermal infrared: lidar = 0.5:0.3:0.2), without dynamic adjustments, and without introducing understory disturbances or tree age correction.

[0040] Drawbacks: The fusion process lacks specificity, the weights are not scientifically sound, and the model has poor adaptability.

[0041] Comparative Example 2 (General Machine Learning Recognition Method) Methods: After standard image fusion using commercial software, the images were directly imported into a general deep learning model (such as YOLOv5) for pest detection. The model was pre-trained using a publicly available pest dataset and was not optimized for the features of the Chinese cypress tussock moth and ancient cypress.

[0042] Limitations: The algorithm lacks specificity, feature extraction relies on large-scale training, it is not suitable for ancient cypress scenarios with scarce samples, and it ignores factors such as tree age.

[0043] Comparative Example 3 (Manual Field Survey Method) Method: Relying entirely on professional technicians to conduct regular on-site sampling and visual assessment using tools such as high-branch pruning shears and binoculars.

[0044] Disadvantages: Manual field surveys are inefficient, have a narrow coverage, are highly subjective, and cause physical interference to ancient cypress trees.

[0045] The key performance indicators of the examples and comparative examples are compared in Table 1 below: Table 1 Note: In Example 2, the recognition rate was slightly affected by the high canopy closure, which partially obscured the young larvae, but it was still significantly higher than that of the control example.

[0046] (1)† The data in Comparative Example 3 are based on sampling extrapolation, resulting in low coverage and consistency.

[0047] (2) Comparative Example 2 model has a long training and tuning time and requires GPU inference.

[0048] (3) § Comparative Example 1 has a simple process but low accuracy, which limits its practical application value.

[0049] (4) ¶ The adaptability of Comparative Example 3 depends on personnel experience and is difficult to standardize and promote.

[0050] From Table 1, we can obtain: (1) The three embodiments are superior to all comparative examples in terms of early larval identification rate (≥90%), insect stage determination accuracy (≥95%), and hazard level determination accuracy (≥95%). This verifies that the present invention can effectively solve the core defects of the prior art (comparative examples 1 and 2) in terms of insufficient identification accuracy and serious misjudgment of early cypress tussock moth infestation by using specific fusion, dynamic weighting, and optimized feature index and correction terms.

[0051] (2) The embodiments were successfully applied to mixed forests of ancient cypress and cypress trees ranging from scattered to dense and from hundreds to thousands of years old, verifying that the present invention has excellent generalization ability and adaptability through dynamic adjustment mechanism and tree age correction, and solved the problem of poor adaptability of fixed parameters or general models in comparative examples 1 and 2.

[0052] (3) The closed-loop process of this invention realizes sub-second processing of a single plant and a cycle of ≤36 hours for the whole process. The efficiency is far superior to manual investigation (Comparative Example 3) and also superior to machine learning methods that require a lot of training (Comparative Example 2). It can respond quickly, avoid missing the prevention and control window period, and solve the pain point of time delay.

[0053] (4) All remote sensing implementation examples (1,2,3) achieve contactless monitoring, fully meeting the strict requirements for low-interference protection of ancient and famous trees. At the same time, the standardization results and platform docking capabilities (which Comparative Example 3 lacks, and Comparative Examples 1 and 2 are incomplete) provide strong direct support for prevention and control decisions.

[0054] In summary, this invention, through systematic innovation from data acquisition, specific fusion, dynamic weight quantification, feature optimization extraction to streamlined identification, forms a complete technical solution with high precision, strong adaptability, high efficiency, low interference, and strong support. The comparative data in Table 1 fully demonstrates that this invention successfully solves the shortcomings of traditional monitoring technologies, has significant technological advancements and practical application value in the field of precise monitoring of tussock moths in ancient and Sichuan cypress trees, and can be extended to a wider range of monitoring of leaf-eating pests in cypress species.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying the Chinese pine tussock moth based on image fusion, characterized in that, Includes the following steps: Step 1: Data Acquisition: Based on the multi-source remote sensing collaborative observation strategy for ancient cypress ecological monitoring, drones equipped with various sensors are used to collect time-series data. Step 2, Preprocessing: The collected multi-source remote sensing data are sequentially subjected to radiometric calibration, atmospheric correction, image registration, outlier removal, and stratification to form a standardized dataset; Step 3: Image fusion specific to the Chinese cypress tussock moth: The db4 wavelet basis wavelet transform fusion algorithm is used to fuse information from multiple sources in a targeted manner; Step 4: Quantitative and Dynamic Adjustment of Fusion Weights: Construct a hierarchical analysis model to determine the basic weights, and dynamically adjust them in combination with field observation data to achieve accurate quantification and dynamic optimization of the fusion weights of multi-source data; Step 5: Image quality control: Establish a closed-loop quality control process to provide high-quality data sources for subsequent feature extraction and recognition; Step Six: Feature Extraction from Fuded Images: Based on the fused images, four core feature indices are calculated to construct a chlorophyll content inversion model and mine quantitative characteristics of pest damage. Step 7: Multi-dimensional feature comprehensive identification and hazard assessment: Based on the extracted features, a streamlined comprehensive judgment is performed to identify insect stage, instar, and hazard level; Step 8: Output Results and Platform Integration: Output the identification results in a formatted format and automatically synchronize them to the management platform to form a monitoring-management closed loop.

2. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: Data collection in step one: (1) Collect multispectral data by using a camera mounted on a drone, controlling the flight altitude to be 80-85m and the flight speed to be 4-5m / s, and collect data 2-3 times per month; (2) Collect thermal infrared data using a FLIR Vue Pro R camera, control the flight altitude to 80-85m and the flight speed to 2.5-3.5m / s, and collect data 1-2 times per week during the active period of larvae; (3) Data was collected by using Velodyne VLP-16 to collect lidar data, with the flight altitude controlled at 80-85m and the speed at 3-4m / s, and data was collected 1-2 times per quarter; (4) Collect radar data through C / X dual-band equipment, control the flight altitude to 95-100m and the speed to 5-6m / s, and collect data synchronously with lidar.

3. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: The preprocessing procedure in step two is as follows: S1.1 Radiometric calibration: Radiometric calibration is performed using the MODTRAN 6.0 model, and the calibration error is ≤2.5% using an ASD FieldSpec 4 spectrometer. The raw signal received by the sensor is converted into radiance values. S1.2 Atmospheric Correction: Atmospheric correction for the FLAASH model, inputting aerosol optical thickness and water vapor content in the monitoring area, the difference between the corrected reflectance and the measured value is ≤3.5%; S1.3 Image registration: Based on ground control points, the quadratic polynomial registration method combined with the RANSAC algorithm is used to remove outliers, with a registration error of ≤0.32 pixels, achieving accurate spatial alignment of multi-source data; S1.4 Outlier Removal and Stratification: Outliers from multi-source data are removed according to corresponding standards; the LiDAR point cloud elevation threshold method is used to separate the understory vegetation and the ancient cypress canopy, with a separation accuracy of ≥93%, forming a standardized dataset.

4. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: The specific image fusion process of the *Sophora spp.* moth in step three: S2.1 Optimization of Fusion Algorithm: The wavelet transform fusion algorithm based on db4 wavelet basis is adopted, and the number of decomposition layers is dynamically adjusted according to the state of the Chinese cypress tussock moth and the canopy closure of the ancient cypress. S2.2 Component Adaptive Fusion: Low-frequency components are weighted and averaged using dynamic weights based on insect state and canopy position; high-frequency components are fused using a combination of the absolute value maximization method and the gradient constraint method, and canopy position weights are introduced to enhance edge and texture information. S2.3 Image post-processing: The CLAHE algorithm is used to enhance the contrast of bands sensitive to pests and diseases after the fused image is processed, and then Gaussian filtering is used for smoothing and noise reduction.

5. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: Step four integrates weight quantification and dynamic adjustment: a three-level weight system is constructed—target layer, criterion layer, and scheme layer—and the basic weights are calculated using the analytic hierarchy process (AHP). =0.0381, where the criterion layer =0.545、 =0.297、 =0.158, Scheme Layer =0.42、 =0.28、 =0.20、 =0.10; combined with the insect stage adjustment coefficient Adjustment coefficient for canopy position The final weight is calculated using the following formula. : In the formula, For the final weight, Based on the weights, This is the insect stage adjustment coefficient. This is the canopy position adjustment factor.

6. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: The quality control process of the fused image in step five is executed in the following order: wavelet decomposition → component fusion → inverse transform → image enhancement → quality verification → result generation, forming a closed-loop control. When the quality standard is not met, the process returns to the front-end step for adjustment. Finally, a fused image in TIFF format, WGS84 coordinate system, and pixel is generated. In the quality verification index, spectral fidelity is calculated by the difference in band reflectance between the fused image and the original multispectral image, so as to realize the quantitative monitoring of the degree of spectral distortion introduced by the fusion process.

7. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 6, characterized in that: The spectral fidelity is calculated by the difference in band reflectance between the fused image and the original multispectral image, and the calculation formula is as follows: In the formula, Indicates spectral fidelity. Indicates the fused reflectivity. This represents the original reflectance.

8. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: The calculation of the four core characteristic indices and the inversion of chlorophyll content in step six are as follows: (1) Vegetation index The calculation formula is: In the formula, For near-infrared reflectivity, For red light band reflectivity, To determine the background reflectance of the understory vegetation, an understory interference correction factor of 0.1 is introduced. (2) Enhanced vegetation index The calculation formula is: In the formula For near-infrared reflectivity, For red light band reflectivity, Background reflectance of forest understory vegetation. For blue light band reflectivity; -7.5× in the denominator The design is optimized based on the physiological characteristic of a significant decrease in blue light reflectance after damage by the Chinese tussock moth, and its model fit is good. =0.91, optimized for the characteristic of reduced blue light reflectivity after damage; (3) Canopy temperature stress index The calculation formula is: In the formula, For the target canopy temperature, The average temperature of a healthy canopy. For the standard deviation of healthy canopy temperature, when >1.2; (4) Spectral index of specific diseases and pests The calculation formula is: In the formula, The reflectance of the damaged canopy, For healthy canopy reflectivity, This refers to the band weighting coefficient. This represents the percentage of the affected area. (5) Chlorophyll content The inversion model is as follows: In the formula, The vegetation index, To enhance the vegetation index, This is the tree age correction factor. It is an ancient cypress tree. The value was obtained by fitting measured data of ancient cypress trees aged 100-2000 years, with a range of 0.002-0.

005. The value increases by 100 years for each additional 100 years of tree age. With the value increasing by 0.001, after introducing the tree age correction term, the model R²≥0.89 and RMSE≤1.25mg / g.

9. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: Step seven involves multi-dimensional feature comprehensive identification and hazard assessment. S3.1 Identification process control: The process is executed in the following order: temperature anomaly screening → interference removal → insect stage determination → instar determination → hazard level determination → manual verification of abnormal samples, forming a standardized identification link to ensure that each link is accurate and controllable; S3.2 Determination of Insect Stages and Instars: Insect stage determination is combined with changes in reflectance in sensitive wavelength bands. value, Values ​​and canopy location, multi-dimensional collaborative determination improves accuracy; age determination is based on... Based on the value and leaf damage rate Si, accurate differentiation of larvae at different ages can be achieved; S3.3 Hazard Level Classification: Hazard levels are classified according to chlorophyll content, The damage is classified into four levels based on the value and the percentage of affected area, which quantitatively distinguishes the degree of damage and provides a basis for targeted prevention and control.

10. The image fusion-based method for identifying the Chinese pine tussock moth according to claim 1, characterized in that: In step eight, the results are output and integrated with the platform: the hierarchical pest and disease location map, standardized monitoring report and visualization results are output to the ancient cypress status visualization platform, so that the location accuracy of the location map is ≤3m, and the distribution and damage of pests are presented intuitively; and the data is synchronized to the ancient cypress health management platform through the MQTT communication protocol, with the data format being JSON and the transmission delay being ≤3.2 seconds, so as to achieve fast data synchronization; the entire process data processing cycle is controlled to be ≤36 hours, and the single batch processing efficiency is ≥120 plants / minute.

Images