CN-122023662-A - Intelligent agricultural park scene simulation method and system based on digital twin technology

CN122023662ACN 122023662 ACN122023662 ACN 122023662ACN-122023662-A

Abstract

The invention discloses a digital twinning technology-based intelligent agricultural park scene simulation method and a digital twinning technology-based intelligent agricultural park scene simulation system, and relates to the technical field of intelligent agriculture, wherein the method is characterized in that an industrial camera is used for collecting crop top leaf edge image data and constructing a focal edge response image sequence, and a digital twinning model is built by combining a crop layout layer and camera parameters; and acquiring environment data, performing time alignment and dynamic time warping processing, constructing a focal edge response image sequence, performing joint modeling on the focal edge response image sequence and the environment data, generating a dynamic expansion prediction map, and constructing a focal Bian Tuxiang twin simulation model. The structural evolution index LEA is formed by calculating the area change rate Adif, the contour shrinkage index Csh and the gray scale offset response Gde, and the structural evolution trend quantity Etr is constructed by analyzing the continuous-moment structural evolution index LEA, so that the spraying and shading regulation strategy is triggered in a linkage mode. The method realizes the collaborative analysis of the image information, the environmental factors and the digital twin model.

Inventors

LI XIAORUI
LI YONGMEI
LU ZEYU
YANG LI
ZHANG XUEJIAN
HAN JIAHAO
LIU JIAN
HAN DONGDONG
ZHANG JIANHUA
LI FENG
CHEN XUEDONG
MA CONG
MA JING

Assignees

宁夏农林科学院农业经济与信息技术研究所

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. The intelligent agricultural park scene simulation method based on the digital twin technology is characterized by comprising the following steps of: S1, acquiring image data of a top leaf edge area of a crop through an industrial camera and forming an image frame sequence, establishing a digital twin model by combining a park crop layout layer, extracting an image of the top leaf area of the crop and performing edge enhancement processing to generate a focal edge response image sequence; S2, extracting environment data, preprocessing to obtain an environment factor sequence, carrying out dynamic time warping matching on the environment factor sequence and the focal edge response image sequence, establishing a corresponding relation between environment change and focal edge response, and correcting to form an input sample pair; S3, carrying out sliding window combination on the input sample pair to construct a sample unit, carrying out image prediction, generating a dynamic expansion prediction map, carrying out structural fusion processing on the dynamic expansion prediction map and a focal edge response image, synchronously loading environmental factors, and constructing a focal Bian Tuxiang twin simulation model; s4, calculating an area change rate Adif, a contour shrinkage index Csh and a gray scale offset response Gde in a focal edge image twinning model, and fusing to form a structure evolution index LEA to evaluate a focal edge structure; s5, carrying out trend analysis on the structure evolution index LEA at continuous moments, and constructing a structure evolution trend quantity Etr to carry out focal edge trend assessment.
2. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 1, wherein S1 includes S11 and S12; S11, acquiring image data of a top leaf edge area of crops in an agricultural park in real time according to an industrial camera arranged in the agricultural park, and arranging the image data according to an acquisition time stamp and an inter-frame interval strategy set by the industrial camera to form an image frame sequence; S12, a crop layout layer is extracted from an agricultural park geographic information system, standardized treatment is carried out, the crop layout layer is mapped to a park coordinate system, each crop plant is used as a twin object entity, and a unique identification code UID is given to each entity; the crop layout layer comprises the types of crops in each land, sowing time, planting density and distribution area boundaries; Converting three-dimensional space coordinates into pixel coordinates on an image plane through perspective projection technology according to internal and external parameters of an industrial camera, constructing a digital twin model of an industrial park, and establishing a mapping relation between an image frame sequence and the digital twin space according to a coordinate mapping function; the internal and external parameters comprise the installation position, the orientation, the focal length and the distortion of the camera.
3. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 2, wherein S1 further comprises S13 and S14; S13, converting the spatial position of the crop in the digital twin model into two-dimensional coordinates in an image plane by utilizing a projection relation between camera parameters of an industrial camera and a spatial coordinate system based on the three-dimensional spatial position of the crop instance recorded by the digital twin model, extracting a color histogram and a texture direction field of the target crop according to an image analysis algorithm, combining the two-dimensional coordinates, the color histogram and the texture direction field, and identifying and positioning an image area where an individual target crop is located in an image frame sequence by utilizing a similarity matching algorithm to obtain an image area of the target crop; A color histogram, namely acquiring color distribution information by calculating a color histogram of a target crop area in the image, and converting the color distribution information into a color feature vector; The grain direction field is used for extracting grain characteristics of the crop leaves through a gray level symbiotic matrix method to form a directional characteristic diagram related to a crop growth mode; Extracting 2 blade area images at the top of the target crop as focal response observation objects according to the growth direction and the relative height model of the plant principal axis in the target crop image area, performing image cutting and rotation normalization operation to output top area images, then applying a morphological gradient enhancement algorithm to strengthen blade edge profile characteristics in the top area images, then carrying out edge detection on the top area images in each frame image frame sequence according to an edge detection algorithm, and extracting edge profiles of the blades to obtain a blade edge area pixel set; S14, calculating the gray scale change rate of each edge pixel in the pixel set of the leaf edge region by utilizing local gray scale differentiation, judging that the current region is a focal edge region when the gray scale change rate of the pixel point between continuous time frames is more than 0.2, and performing image processing on the focal edge region by a second-order edge enhancement algorithm to construct a focal edge response image; and arranging the focal edge response images generated at each moment in time sequence according to the image acquisition time stamp to form a focal edge response image sequence.
4. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 3, wherein S2 includes S21, S22 and S23; s21, extracting environmental data from a historical database in an agricultural park, and acquiring an environmental factor sequence after data preprocessing; the environmental data comprises an environmental temperature Tw, an environmental radiation Lf and an environmental humidity Hs; The data preprocessing comprises data alignment, data smoothing and denoising; S22, performing time stamp standardization processing on the focal edge response image sequence and the environment factor sequence, recursively calculating the similarity between the focal edge response image sequence and the environment factor sequence by adopting a dynamic time warping algorithm, constructing a similarity measurement matrix between the change trend of the focal edge response image and the change track of the environment factor, traversing the similarity measurement matrix between different time points, minimizing the deviation value of the focal edge response image sequence and the environment factor sequence in the accumulated difference dimension, and outputting a corresponding mapping index; Recombining the time points of the environmental factor sequences based on the mapping relation of the corresponding mapping indexes, so that each focal edge response image corresponds to the environmental factors which change with the focal edge response image; S23, matching continuous change points in the environment factor sequence with actual response starting points in the focal edge response image sequence to obtain response lag time differences, establishing a mathematical mapping relation between a plurality of response lag time differences and corresponding environment factor change amplitudes, constructing response lag expression values by using a multiple linear regression mode, correcting image response at the current time point based on the response lag expression values, and pairing the focal edge response image with the environmental factors which are advanced and lagged to form an input sample pair after delay correction.
5. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 4, wherein S3 includes S31 and S32; S31, sliding and slicing input sample pairs in a focal edge response image sequence with a fixed window length, constructing a plurality of groups of structure sample units, wherein each group of sample units comprises continuous focal edge response images and corresponding environmental factor data, in each sample unit, performing image prediction recursion operation by taking environmental factor trend corresponding to the last frame of the focal edge response image sequence as input, outputting future simulation images, and splicing the predicted future simulation images with an original focal edge response image sequence to form a dynamic expansion prediction map; S32, carrying out structure fusion processing on the focal edge response image at the current moment and the dynamic expansion prediction map, synchronously loading environmental factors in the current period, forming a simulation correlation image of an image state and environmental state correlation structure, carrying out region comparison and edge contour fusion processing on the simulation correlation image through an image structure similarity analysis algorithm, constructing an image-level twin body covering a leaf edge region, and injecting the image-level twin body into the digital twin model to construct a focal Bian Tuxiang twin simulation model.
6. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 5, wherein S4 includes S41; S41, under a unified space coordinate system in a focal edge image twin simulation model, extracting a focal edge pixel set corresponding to a leaf edge region in a continuous time window focal edge response image, identifying a pixel contour of the focal edge region, counting a coverage area, performing difference value on the areas of the focal edge region at adjacent moments, and then comparing with the coverage area at the previous moment to obtain an area change rate Adif; Extracting edge contours of the focusing edge areas, calculating differences of contour lengths of adjacent moments under the same spatial scale, and comparing the calculated differences with the contour length of the previous moment to obtain a contour shrinkage index Csh; And carrying out gradient analysis on the image gray distribution in the focusing edge area, calculating gradient variation of the current moment and the previous moment at the same position, and constructing gray offset response Gde.
7. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 6, wherein S4 further includes S42 and S43; S42, calling a fusion processing function, namely selecting one of the surface variation rate Adif and the contour shrinkage index Csh, which has a large value, as a dominant structure variation factor max { Adif (t), csh (t) }, and performing compression ln [1+Gde (t) ] on the gray scale offset response Gde by using a natural logarithmic function; Adding the dominant structural change factor and the gray level compression term to construct a structural evolution index LEA, reflecting the overall change amplitude and direction of the structural state of the top leaf margin area of the plant, wherein the method is as follows; LEA(t)=max{Adif(t),Csh(t)}+ln[1+Gde(t)]; Wherein LEA (t) represents a structural evolution index at a time t, and ln represents a logarithmic function; S43, evaluating the focal edge structure of the plant leaf edge structure in the agricultural park according to the structure evolution index LEA, wherein the specific evaluation scheme is as follows; when the structural evolution index LEA is less than 1, the current plant leaf margin structure is indicated to be in a normal state, and the current plant is marked as a green mark and is kept to be monitored normally; when the structural evolution index LEA is more than or equal to 1, the current plant leaf margin structure is in a stress state, the current plant is marked as a red mark at the moment, spraying water supplementing and a window shade opening are activated through a focal margin image twin simulation model, and leaf margin trend prediction is activated.
8. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 7, wherein S5 includes S51; s51, when the focal edge state is evaluated to be in the focal edge critical state, carrying out leaf edge trend prediction on the structure evolution indexes LEA at a plurality of continuous moments according to a time sequence, and constructing a structure evolution trend quantity Etr to represent the accumulated trend of the change of the focal edge structure of the blade in a continuous time window, wherein the method is as follows; ; wherein, etr (t) represents the structure evolution trend quantity at the time t, and LEA (k) and LEA (k+1) represent the structure evolution indexes at the time k and the time k+1, respectively.
9. The intelligent agricultural park scene simulation method based on the digital twin technology as set forth in claim 8, wherein S5 further comprises S52; S52, collecting structure evolution trend quantity Etr of a plurality of crop samples in a historical focal edge image twin simulation model, classifying the structural states of the focal edges of the historical blades, setting the peak value of the structure evolution trend quantity Etr of a historical back-off state interval as a trend sensitive threshold Te, setting the peak value of the structure evolution trend quantity Etr of a historical steady state interval as a trend confirmation threshold Tt, and carrying out focal edge trend assessment with the structure evolution trend quantity Etr acquired in real time, wherein the steps are as follows; When the structure evolution trend quantity Etr is smaller than the trend sensitivity threshold Te, the current blade focal edge structure is in a retreating state, and the existing management strategy is maintained; When the trend sensitivity threshold Te is less than or equal to the structure evolution trend quantity Etr and is less than the trend confirmation threshold Tt, the current blade focal edge structure is in a stable state, and a focal edge image twin simulation model recording flow is entered at the moment, so that the current shading state and the water spraying state are maintained; When the structural evolution trend quantity Etr is more than or equal to the trend confirmation threshold value Tt, the current blade focal edge structure is in a continuously aggravated state, the unfolding area of the window shade is increased to the maximum adjustable range of the current area, and the top water spraying device is started.
10. The intelligent agricultural park scene simulation system based on the digital twin technology comprises the intelligent agricultural park scene simulation method based on the digital twin technology as set forth in any one of claims 1-9, and is characterized by comprising an image extraction module, a time lag alignment module, a simulation model construction module, a structure evolution analysis module and a trend discrimination module; The image extraction module is used for acquiring image data of a top leaf edge area of a crop according to an industrial camera and forming an image frame sequence, establishing a digital twin model by combining a park crop layout layer, extracting an image of the top leaf area of the crop and carrying out edge enhancement processing to generate a focal edge response image sequence; The time lag alignment module is used for extracting environment data and preprocessing to obtain an environment factor sequence, carrying out dynamic time alignment matching on the environment factor sequence and the focal edge response image sequence, establishing a corresponding relation between environment change and focal edge response, and forming an input sample pair after correction; The simulation model construction module is used for carrying out sliding window combination on input sample pairs to construct a sample unit and carrying out image prediction to generate a dynamic expansion prediction map, carrying out structural fusion processing on the dynamic expansion prediction map and a focal edge response image and synchronously loading environmental factors to construct a focal Bian Tuxiang twin simulation model; the structure evolution analysis module is used for calculating an area change rate Adif, a contour shrinkage index Csh and a gray scale offset response Gde in a focal edge image twin model, and fusing to form a structure evolution index LEA for focal edge structure evaluation; the trend discrimination module is used for carrying out trend analysis on the structure evolution index LEA at continuous moments, and constructing a structure evolution trend quantity Etr to carry out focal edge trend assessment.

Description

Intelligent agricultural park scene simulation method and system based on digital twin technology Technical Field The invention relates to the technical field of intelligent agriculture, in particular to a digital twinning technology-based intelligent agricultural park scene simulation method and system. Background With the continuous fusion of information physical systems, internet of things and artificial intelligence technologies, digital twinning gradually develops from single equipment modeling to a dynamic mapping technology oriented to a complex system. The method is characterized in that a virtual space is driven by real-time data, so that a physical object has synchronous mapping, state deduction and behavior prediction capabilities in a digital space. In the agricultural field, the crop growth process has strong time sequence, environment coupling and space heterogeneity, and fine cognition and dynamic regulation are difficult to realize only by relying on the traditional monitoring means. Therefore, digital twin technology is gradually introduced into intelligent agricultural park construction for constructing virtual-real mapping relations between crop-environment-management behaviors. On the basis, an intelligent agricultural park scene simulation mode which takes the physiological state of crops as a core, takes environmental change as a drive and takes images and multisource perception data as a support is formed, so that agricultural production is changed from static experience judgment to a computable and deductible digital operation mode. Although the existing intelligent agricultural system has a certain progress in the aspects of environment monitoring, data acquisition and remote control, most of the intelligent agricultural system still stays in a parameter level monitoring or threshold control stage, has limited capability of describing the actual physiological state of crops, and especially lacks continuous and quantifiable dynamic expression means in the aspect of identifying early stress characteristics such as leaf edge pyroedge, dry out and the like of crops. On one hand, the traditional image analysis method usually only focuses on single-frame image characteristics, lacks time correlation modeling with environmental change, and on the other hand, lacks an effective coupling mechanism between environmental data and crop phenotype data, so that the inherent logic relationship among environmental change, physiological response and morphological evolution is difficult to be described, and early warning lag and rough intervention are caused. In addition, the existing system takes an experience threshold value or a single index as a judgment basis, so that the structure evolution trend in the coke edge development process is not easily reflected, and the requirements of fine regulation and control and simulation deduction are difficult to support. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a digital twinning technology-based intelligent agricultural park scene simulation method and system, which solve the problems in the background technology. The intelligent agricultural park scene simulation method based on the digital twin technology comprises the following steps: S1, acquiring image data of a top leaf edge area of a crop through an industrial camera and forming an image frame sequence, establishing a digital twin model by combining a park crop layout layer, extracting an image of the top leaf area of the crop and performing edge enhancement processing to generate a focal edge response image sequence; S2, extracting environment data, preprocessing to obtain an environment factor sequence, carrying out dynamic time warping matching on the environment factor sequence and the focal edge response image sequence, establishing a corresponding relation between environment change and focal edge response, and correcting to form an input sample pair; S3, carrying out sliding window combination on the input sample pair to construct a sample unit, carrying out image prediction, generating a dynamic expansion prediction map, carrying out structural fusion processing on the dynamic expansion prediction map and a focal edge response image, synchronously loading environmental factors, and constructing a focal Bian Tuxiang twin simulation model; s4, calculating an area change rate Adif, a contour shrinkage index Csh and a gray scale offset response Gde in a focal edge image twinning model, and fusing to form a structure evolution index LEA to evaluate a focal edge structure; s5, carrying out trend analysis on the structure evolution index LEA at continuous moments, and constructing a structure evolution trend quantity Etr to carry out focal edge trend assessment. Preferably, the S1 includes S11 and S12; S11, acquiring image data of a top leaf edge area of crops in an agricultural park in real time according to an industrial camera arranged in the agricul