Intelligent nested recognition and measurement method for forest biomass based on unmanned aerial vehicle remote sensing

Abstract

The application discloses a forest biomass intelligent nested recognition and measurement method based on unmanned aerial vehicle remote sensing, and specifically comprises the following steps: S1, forest biomass data collection; S2, data processing; S3, information extraction; S4, nested entity boundary detection: the text paragraph information generated in S3 is processed, word segmentation and word type annotation are carried out, and the nested boundary information mentioned by each text paragraph is detected; S5, entity multi-layer information annotation: after that, the data annotated in S4 is processed, and the entity type, entity subtype and mentioned role information of each group of text paragraphs are recognized; the application proposes an intelligent nested recognition and measurement method, uses a conditional random field model to fuse a plurality of features for statistical decision, secondly, a multi-layer information annotation problem is regarded as a classification problem, two classification engine parallel modes are designed from the realization angle, and a classifier is not designed for each layer of information annotation, so that the performance is obviously improved compared with the single classifier.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nested recognition measurement, and particularly relates to an intelligent nested recognition and measurement method for forest biomass based on unmanned aerial vehicle remote sensing. BACKGROUND

[0002] Unmanned aerial vehicle remote sensing refers to a modern scientific technology that uses sensors of unmanned aerial vehicles to collect electromagnetic wave information of target objects, and after processing and analysis, identifies the target objects, reveals the geometric, physical properties and mutual relationship and change rule thereof. In other words, it is "remote sensing", which monitors the quantity information of various organisms in the forest through the unmanned aerial vehicle, and then processes the biomass information. However, the existing biomass information processing method needs to increase a large number of processing sources for various characteristics, so the statistical decision-making ability of the data is poor, and when information labeling is performed, a classifier needs to be designed for each layer of information labeling, resulting in poor data processing performance of the classifier. SUMMARY

[0003] To achieve the above object, the present application is realized by the following technical scheme: an intelligent nested recognition and measurement method for forest biomass based on unmanned aerial vehicle remote sensing, specifically comprising the following steps:

[0004] S1, forest biomass data acquisition: first, remotely control the unmanned aerial vehicle to fly through a wireless remote control device, and use the remote sensing sensor on the unmanned aerial vehicle to acquire data information from the forest organisms in real time;

[0005] S2, data processing: then process the forest organism data information acquired in S1 by using a computer, and generate image information from the collected biological data;

[0006] S3, information extraction: extract the information data in the multiple groups of images in S2, and generate text paragraphs from the information data displayed in the images;

[0007] S4, nested entity boundary detection: then process the text paragraph information generated in S3, perform word segmentation and part-of-speech tagging, detect the nested boundary information mentioned in each text paragraph, specifically, label each word in the word segmented and part-of-speech tagged text with an encoding, and then obtain the nested boundary structure information that may exist between the entity mentions through a decoding algorithm;

[0008] S5, entity multi-layer information labeling: then process the labeled data in S4, and identify the entity type, entity subtype and mention role information of each group of text paragraphs.

[0009] Preferably, the data information of the forest organisms in S1 comes from large animals, plants, insects and rodents, etc.

[0010] Preferably, the images generated in S2 are classified and arranged according to the forest organism types in S1.

[0011] Preferably, the text paragraphs in S3 are collected by computer, and the collected data are arranged from largest to smallest according to the forest organism types in S2.

[0012] Preferably, in step S4, a code value is assigned to each word in a text segment and each sentence within it. This code value contains nesting information between text segments. By solving the annotation problem in the text segment sequence and then performing a decoding process, nested entity mention boundary detection is achieved. First, for each word, a position label is obtained based on the positions in all text segments containing it. A conditional random field model is defined for the position labels to realize the detection of a word sequence X = {X1, X2, ..., X...}. n To a labeled sequence Y = {Y1, Y2, ..., Y} n The mapping of ,} is defined by the conditional random field model, and the probability of obtaining a labeled sequence given a sequence can be expressed as:

[0013] Where Z(X) refers to the normalization factor, and Y... i This refers to the code value containing hierarchical structure information, where Y is the set of all token sequences, n represents the length of the given word sequence, and f k (y i-1 ,y i (x,i) is an eigenfunction used to represent any non-independent feature, λ k It is the weight coefficient assigned to the k-th feature function.

[0014] Preferably, step S5 adopts a compromise method to divide all text paragraph information into two major categories: entity category and mention category. Two classification engines are used to classify them in parallel. For the two classification engines, we define the feature representation of text paragraph mentions to be consistent, mainly divided into three types of features: external features of entity mentions, internal features of entity mentions, and interactive features of entity mentions.

[0015] Beneficial effects

[0016] This invention provides an intelligent nested identification and measurement method for forest biomass based on UAV remote sensing. Compared with existing technologies, it has the following advantages: This intelligent nested identification and measurement method for forest biomass based on UAV remote sensing utilizes a conditional random field model to fuse multiple features for statistical decision-making, thereby improving the statistical decision-making capability of this method. Secondly, it treats the multi-layer information labeling problem as a classification problem and designs two parallel classification engines from an implementation perspective, avoiding the need to design a classifier for each layer of information labeling, which significantly improves performance compared to using a single classifier. Attached Figure Description

[0017] Figure 1 The flow step block diagram of the intelligent nested recognition and measurement method for forest biomass based on unmanned aerial vehicle remote sensing is shown in the figure. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] Please refer to Figure 1 The embodiments of the present application provide a technical solution: an intelligent nested recognition and measurement method for forest biomass based on unmanned aerial vehicle remote sensing, which specifically includes the following steps:

[0020] S1, forest biomass data acquisition: first, remotely control the unmanned aerial vehicle to fly through the wireless remote control device, and use the remote sensing sensor on the unmanned aerial vehicle to obtain data information from the forest biomass in real time;

[0021] S2, data processing: then process the forest biomass data information obtained in S1 by using a computer, and generate image information from the collected biological data;

[0022] S3, information extraction: extract the information data in the multiple groups of images in S2, and generate text paragraphs from the information data displayed in the images;

[0023] S4, nested entity boundary detection: then process the text paragraph information generated in S3, perform word segmentation and part-of-speech tagging, detect the nested boundary information mentioned in each text paragraph, specifically, label each word in the segmented and tagged text with an encoding, and then obtain the nested boundary structure information that may exist between entity mentions through a decoding algorithm;

[0024] S5, entity multi-layer information labeling: then process the labeled data in S4, and identify the entity type, entity subtype and mention role information of each group of text paragraphs.

[0025] In the present application, the data information of the forest biomass in S1 comes from large animals, plants, insects and rodents, so that a large amount of biological information in the forest can be comprehensively collected, and the comprehensiveness of the collected data is ensured.

[0026] In the present application, the images generated in S2 are classified and arranged according to the forest biomass types in S1, which facilitates data viewing, and the data storage is more convenient, facilitating subsequent statistical decision-making.

[0027] In this invention, the text paragraphs in S3 are collected by computer, and the collected data are arranged from largest to smallest according to the forest organism types in S2.

[0028] In this invention, step S4 assigns a code value to each word in a text segment and its internal sentences. This code value contains nesting information between text segments. By solving the annotation problem in the text segment sequence and then implementing the nested entity mention boundary detection through the decoding process, firstly, for each word, a position label is obtained based on the positions in all text segments containing it. A conditional random field model is defined for the position labels to realize the detection of nested entity mention boundaries from a word sequence X = {X1, X2, ..., X...}. n To a labeled sequence Y = {Y1, Y2, ..., Y} n The mapping of ,} is defined by the conditional random field model, and the probability of obtaining a labeled sequence given a sequence can be expressed as:

[0029] Where Z(X) refers to the normalization factor, and Y... i This refers to the code value containing hierarchical structure information, where Y is the set of all token sequences, n represents the length of the given word sequence, and f k (y i-1 ,y i (x,i) is an eigenfunction used to represent any non-independent feature, λ k It is the weight coefficient assigned to the k-th feature function. By using the conditional random field model to fuse multiple features for statistical decision-making, the statistical decision-making ability of this method is improved.

[0030] In this invention, S5 adopts a compromise method to divide all text paragraph information into two main categories: entity class and mention class. Two classification engines are used to classify them in parallel. For the two classification engines, we define the feature representation of text paragraph mentions as consistent, which is mainly divided into three types of features: external features of entity mentions, internal features of entity mentions, and interactive features of entity mentions. The multi-level information annotation problem is regarded as a classification problem. From the implementation perspective, a parallel approach of two classification engines is designed to avoid designing a classifier for each level of information annotation, which has a significant performance improvement compared to using a single classifier.

[0031] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0032] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it should be taken in a descriptive sense and not a limiting sense.

[0033] While the embodiments of the application have been shown and described herein, it is to be understood that the application is not limited to these embodiments. Rather, it is to be understood that various modifications, changes, substitutions and alterations can be made to the embodiments without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents.

Claims

1. A method for intelligent nested identification and measurement of forest biomass based on UAV remote sensing, characterized in that, Specifically, the following steps are included: S1. Forest biomass data collection: First, the drone is remotely flown by radio remote control equipment, and the remote sensing sensor on the drone is used to acquire data information from forest organisms in real time. S2, Data Processing: The forest biological data obtained in S1 is then processed by computer, and the collected biological data is used to generate image information. S3. Information Extraction: Extract information data from multiple sets of images in S2, and generate text paragraphs from the information data displayed in the images; S4. Nested Entity Boundary Detection: The text paragraph information generated in S3 is then processed, and word segmentation and part-of-speech tagging are performed to detect the nested boundary information mentioned in each text paragraph. S5, Entity Multi-layer Information Labeling: Then, information processing is performed on the data labeled in S4 to identify the entity type and mentioned role information for each group of text paragraphs.

2. The method for intelligent nested identification and calculation of forest biomass based on UAV remote sensing according to claim 1, characterized in that, The data on forest organisms in S1 comes from macrofolk animals, plants, insects, and rodents.

3. The method for intelligent nested identification and measurement of forest biomass based on UAV remote sensing according to claim 1, characterized in that, The images generated in S2 are classified and arranged according to the forest organism types in S1.

4. The method for intelligent nested identification and calculation of forest biomass based on UAV remote sensing according to claim 1, characterized in that, The text paragraphs in S3 are collected by computer, and the collected data are arranged from largest to smallest according to the forest organism types in S2.

5. The method for intelligent nested identification and calculation of forest biomass based on UAV remote sensing according to claim 1, characterized in that, In S4, a code value is assigned to each word in a sentence within a text segment. This code value contains the nesting information between text segments. By solving the annotation problem in the text segment sequence, and then realizing the nested entity mention boundary detection through the decoding process, firstly, for each word, a position label is obtained according to the position of mentioning it in all text segments containing it, and a conditional random field model is defined for the position label.

6. The method for intelligent nested identification and calculation of forest biomass based on UAV remote sensing according to claim 1, characterized in that, The S5 adopts a compromise method to divide all text paragraph information into two major categories: entity class and mention class. Two classification engines are used to classify them in parallel. For the two classification engines, we define the feature representation of text paragraph mentions to be consistent, which is mainly divided into three types of features: external features of entity mentions, internal features of entity mentions, and interactive features of entity mentions.

Images