CN-121705908-B - Ancient tree famous tree health assessment method and system

CN121705908BCN 121705908 BCN121705908 BCN 121705908BCN-121705908-B

Abstract

The invention discloses a method and a system for evaluating health of ancient tree names, wherein the method comprises the steps of collecting multidimensional health index data of target ancient tree names, carrying out cleaning, normalization and missing value processing on the index data to form a standardized data set, extracting feature vectors, inputting the standardized data set into a pre-trained gradient lifting decision tree model, outputting a health grade prediction result of the target ancient tree names, carrying out iterative training by taking a minimized objective function as a training target, wherein the objective function consists of a loss function and a regularization term, and comprises a weighted cross entropy loss term and a time sequence smoothing regularization term. According to the invention, the health assessment of ancient tree famous trees is changed from qualitative judgment depending on subjective experience to objective diagnosis based on multidimensional quantitative data and algorithms, and the scientificity and decision support value of the assessment result are remarkably improved.

Inventors

XU LIANGYI
YUE PEIHAO
YAN BEI
WEI HUIYU
LIANG CHENG
Xiang Guanyun

Assignees

湖南省森防科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260214

Claims (8)

1. A method for evaluating the health of ancient and famous trees, comprising: s1, collecting multidimensional health index data of a target ancient tree name, wherein the multidimensional health index data comprise an overground morphological index, an underground structure and soil environment index and a physiological function index; s2, cleaning, normalizing and missing value processing are carried out on the index data to form a standardized data set, and feature vectors are extracted from the standardized data set; S3, inputting the feature vector into a pre-trained gradient lifting decision tree model, and outputting a health evaluation index grade score of the target ancient tree name by the model so as to obtain a health grade prediction result; The gradient lifting decision tree model is obtained through training in the following mode: Using a historical ancient tree name data set, taking a minimized objective function as a training target, and carrying out iterative training on the gradient lifting decision tree model; the gradient lifting decision tree model adopts a grouping integration strategy for training and prediction, and comprises the following steps: The training stage comprises the steps of dividing training data into a plurality of subsets according to tree species or ground types of ancient trees, and respectively training a sub-gradient lifting decision tree model for each subset; The method comprises the steps of firstly dividing a target ancient tree to be evaluated into corresponding groups according to metadata of the target ancient tree, respectively predicting the target ancient tree by using one or more sub-models of the corresponding groups, and synthesizing the prediction results of the sub-models through a meta classifier to obtain a final health grade; In the training process of the gradient lifting decision tree model, a multi-mode fusion strategy is also adopted, and the method specifically comprises the steps of dividing health indexes with the same measurement sources into the same data modes, carrying out node splitting in the characteristics of the same data modes preferentially when the decision tree grows, carrying out weighted fusion on the representation of each data mode, and dynamically learning the weight of the representation by an attention mechanism; the objective function is composed of a loss function and a regularization term, comprising: (1) A weighted cross entropy loss term that introduces adaptive weights for different health class categories, the weights being related to the inverse of the number of category samples to promote the degree of interest for the rare category of samples; (2) And the time sequence smooth regular term is used for punishing severe fluctuation of the health scoring result of the same ancient tree at adjacent time points so as to enable the health evolution track output by the model to accord with the slow gradual change characteristic of the health of the ancient tree.
2. The method for evaluating the health of ancient tree names according to claim 1, wherein the weighted cross entropy loss term is specifically defined as: Wherein, the Is the true label of sample i and, Is the probability that model predictive sample i belongs to category c, For each category The dynamic weight of (2) is calculated by the following formula: where N is the total number of samples, nc is the number of samples of class c, Is a smoothing factor.
3. The method for evaluating the health of an ancient tree name according to claim 1, wherein the time-series smoothing regularization term is defined as: wherein M is the number of the ancient trees, For the time-series regularization strength parameter, 、 Is the health score calculated by the model for the ith ancient tree in the t-th and t-1 th years.
4. The method for evaluating the health of the ancient tree name according to claim 1 is characterized in that before model training, a training data set is subjected to layering and oversampling treatment, specifically, layering is carried out firstly according to tree species and tree ages of the ancient tree, and based on neighbor distribution of samples in a feature space, a boundary sample synthesis technology is adopted for oversampling for further improving the learning ability of the model on rare class samples aiming at the sparse number of the layers.
5. The method for evaluating the health of ancient tree names according to claim 1, wherein in the training process of the gradient-lifting decision tree model, dynamic weights are allocated to different health index features according to tree species metadata of the sample ancient tree when each decision tree is generated, and split nodes are selected according to weighted information gains so as to adapt to the difference of sensitivity of different tree species to health indexes.
6. The method for evaluating the health of ancient famous tree according to claim 1, wherein the index data comprises a normal leaf rate, a multispectral index NDVI, a branch tip loss degree, a trunk inclination, a bark damage degree, a trunk internal rot, a trunk liquid flow strength, a root system health degree, a soil coverage degree, a crown below-hardening degree, a soil pH value, a soil temperature, a soil humidity, a soil nutrient content, a surface water accumulation duration time and a plant disease and insect pest hazard level.
7. The method according to claim 1, wherein in step S2, the extracting the feature vector includes constructing a time-series derivative feature of the index data having continuous time-series data, the time-series derivative feature including a linear trend slope calculated based on a sliding time window, a coefficient of variation, and a difference between adjacent time points.
8. An ancient-tree-name-wood health evaluation system for implementing the ancient-tree-name-wood health evaluation method according to any one of claims 1 to 7, comprising: The data acquisition module is configured for acquiring multidimensional health index data of the target ancient tree famous tree; The preprocessing and characteristic engineering module is configured for preprocessing data and extracting characteristic vectors; The model prediction module comprises a pre-trained gradient lifting decision tree model and is used for receiving the feature vector and outputting a health grade prediction result; and the result output module is configured to display and store the health grade prediction result.

Description

Ancient tree famous tree health assessment method and system Technical Field The invention relates to the field of protection and monitoring of ancient and famous trees, in particular to a method and a system for evaluating health of the ancient and famous trees. Background The ancient tree is taken as a precious natural and cultural heritage, and has extremely high ecological, historical, scientific research and ornamental values. The accurate and timely evaluation of the health condition is the basis for implementing scientific protection and rejuvenation. The life cycle of ancient trees and famous trees is hundreds of years or even thousands of years, and the health state is the result of complex interaction of multidimensional factors such as ground morphology, underground root systems, physiological functions, soil environment and the like, so that the health assessment becomes a typical technical problem of multiple variables, nonlinearity and high complexity. The existing ancient and famous tree health assessment method mainly has the following three technical limitations: 1. subjectivity and unilaterality of the traditional manual evaluation method: Currently, the widely adopted evaluation method of the basic layer mainly depends on field visual inspection and experience judgment of forestry specialists. The method has the remarkable defects that (1) the subjectivity is strong, the evaluation result is easily influenced by personal experience and subjective judgment of an evaluator, the health grade judgment of different evaluators on the same ancient tree is possibly huge and lacks objective and unified quantification standards, (2) the evaluation is carried out on one-sided, wherein the evaluation is concentrated on overground parts which are easy to observe (such as yellow branches and leaves, trunk cavities and the like), and key hidden factors (such as root system distribution, physical and chemical properties of soil, trunk liquid flow and other physiological activities) for determining the vitality of the tree are difficult to effectively diagnose, the root cause of health problems cannot be systematically revealed, and (3) the efficiency is low, the time and the labor are consumed, and the requirements of large-scale and large-batch healthy general investigation and dynamic monitoring of the ancient tree are difficult to deal with. 2. Surfacing and data island problems for modern sensing technology applications: With the development of technology, some researches start to try to introduce various sensors (such as multispectral cameras, soil sensors and the like) for data acquisition. However, such technical applications tend to stay at the level of simple data listing and single-index threshold alarms. The method mainly aims at solving the problems that (1) multi-source data fusion is insufficient, acquired multi-dimensional index data such as ground morphology, root system structure, soil environment, physiological function and the like are mutually fractured, an effective fusion analysis model is lacking, correlation diagnosis between apparent symptoms and internal physiological and environmental stress cannot be formed, and (2) an evaluation model is lacking, namely an intelligent evaluation model which has data but is capable of digesting the multi-dimensional heterogeneous data and outputting comprehensive and quantifiable health grade is lacking, so that a technical means is disjointed with a final decision. 3. "Soil and water disuse" directly applied by general machine learning model: To solve the problem of the evaluation model, there have been studies attempting to apply machine learning algorithms (e.g., decision trees, support vector machines, etc.). However, when the general models are directly applied, the unique properties of the ancient tree health data are seriously ignored, so that the evaluation effect is poor, and the method is particularly characterized in that (1) the data are extremely unbalanced and insensitive, the natural distribution of the health state of the ancient tree is extremely unbalanced, the healthy and sub-healthy samples account for the majority, and the critical endangered samples are extremely few. The loss function of the general model tends to be of most types, so that the prediction omission rate of the endangered ancient tree which needs to be concerned is high, and (2) the time sequence dynamic perception capability is lacking, namely the healthy evolution of the ancient tree is a slow continuous process, and annual monitoring data has strong time correlation. The general model regards annual data as independent samples, can not capture the trend of health decline by utilizing history information, can only carry out static judgment and early warning hysteresis, and (3) ignores the heterogeneity of characteristic importance, namely that under different tree species (such as camptothectic acid and camptothecins) and different habitats (such as citie