CN-122024219-A - Three-dimensional planting growth state intelligent evaluation system based on image recognition

CN122024219ACN 122024219 ACN122024219 ACN 122024219ACN-122024219-A

Abstract

The invention relates to the technical field of intelligent agriculture and image processing, in particular to an intelligent evaluation system for a three-dimensional planting growth state based on image recognition. And positioning the plant individuals based on the map, backtracking and extracting the historical characteristics of the plant individuals to form a time sequence growth characteristic chain, identifying a growth abnormal stage by analyzing the characteristic chain change, and automatically taking out contemporaneous illumination and nutrient records for cross verification. And finally, calculating the deviation degree of the plant growth state by fusing the multi-source data, and outputting a hierarchical evaluation report. The system realizes the accurate positioning of the growth state of the individual plants in the stereoscopic planting environment, the tracking of the dynamic process and the auxiliary diagnosis of the abnormal root cause.

Inventors

Peng Nisong
LING XIAOMING
Wei Niankang

Assignees

深圳市海卓生物科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The intelligent evaluation system for the three-dimensional planting growth state based on image recognition is characterized by comprising: The acquisition and preprocessing module is used for carrying out dynamic scanning around the three-dimensional planting frame by the image acquisition device, generating an original image cluster containing space position information, and carrying out invalid region rejection on the original image cluster to obtain a purified image cluster; The feature decoupling and mapping module is used for performing growth feature decoupling on the purified image cluster, separating a structural feature subset representing plant morphology and a spectral feature subset reflecting plant physiology, constructing a spatial index map, and mapping and correlating the structural feature subset and the spectral feature subset according to the spatial position when the structural feature subset and the spectral feature subset are acquired; The feature extraction module is used for positioning plant individual coordinates to be evaluated in the growth state in the spatial index map according to the associated structural feature subsets and spectrum feature subsets, and extracting the structural feature subsets and spectrum feature subsets of the plant individual coordinates in a history acquisition period in a retrospective way to form a time sequence growth feature chain of the plant individual; The detection and verification module is used for analyzing the change mode of the time sequence growth characteristic chain, identifying an abnormal evolution stage with the characteristic evolution rate exceeding a preset normal range, and starting a cross verification process of the illumination and nutrient supply history record; and the evaluation output module is used for fusing the characteristic data of the abnormal evolution stage and the result data of the cross verification process, calculating to obtain the deviation degree of the growth state of the specific plant individual, integrating the deviation degree of all the plant individuals and the overall environmental parameters, and outputting a hierarchical evaluation report.
2. The intelligent evaluation system for the growth state of stereoscopic planting based on image recognition according to claim 1, wherein the step of generating an original image cluster containing spatial position information by dynamically scanning around the stereoscopic planting frame by the image acquisition device comprises: Controlling a mobile platform carrying a multispectral sensor to run along a preset three-dimensional track of a three-dimensional planting frame, and triggering an image acquisition event at fixed time intervals in the running process; When the image acquisition event is triggered each time, synchronously recording the instantaneous three-dimensional space coordinates and the attitude angles of the multispectral sensor, and generating a pose data packet; Binding the multispectral image collected each time with a corresponding pose data packet to form an initial image data unit; After completing one complete scanning period of the stereoscopic planting frame, summarizing all initial image data units to generate an original image cluster; and performing redundant frame screening on the original image clusters, and removing repeated image data units with spatial overlapping degree exceeding a threshold value based on the spatial coordinate similarity of adjacent initial image data units.
3. The intelligent evaluation system for three-dimensional planting growth state based on image recognition according to claim 1, wherein the step of removing an invalid region from the original image cluster to obtain a purified image cluster comprises the steps of: Carrying out foreground segmentation on each frame of image in the original image cluster by adopting a background modeling algorithm to distinguish plant areas from non-plant background areas; detecting regular occlusion contours formed by the fixed support structures and irregular occlusion spots formed by temporary debris in the non-plant background area; based on a preset three-dimensional model of the three-dimensional planting frame, calculating theoretical projection of the regular shielding outline in the image, and comparing and correcting the theoretical projection with the actually detected regular shielding outline; for irregularly blocked spots, analyzing the stability of the irregularly blocked spots through a multi-frame image sequence, marking the continuously existing stable spots as permanent shielding objects, and marking the temporarily occurring spots as temporary interference; And carrying out pixel filling or whole-frame elimination processing on the image area which is marked as the permanent shielding object and corresponds to the temporary interference, and generating a purified image cluster which does not contain the invalid image area.
4. The intelligent evaluation system for three-dimensional plant growth state based on image recognition according to claim 1, wherein the step of performing growth feature decoupling on the cleaned image clusters to separate a subset of structural features characterizing plant morphology from a subset of spectral features reflecting plant physiology comprises: carrying out channel separation on the purified image clusters, and extracting image data of a red channel, a green channel, a blue channel and a near infrared channel; Extracting plant height, leaf area, stem thickness and branch number of a plant from image data of a red channel, a green channel and a blue channel through an edge detection and contour tracking algorithm to form a structural feature subset; calculating normalized vegetation index and photochemical reflection index from image data of a near infrared channel and a red channel, and extracting a reflectivity curve of a specific wave band from multispectral data to form a spectral feature subset; And attaching a space-time tag derived from a specific image in the cleaned image cluster to each piece of characteristic data in the structural characteristic subset and the spectral characteristic subset.
5. The intelligent evaluation system for three-dimensional planting growth state based on image recognition according to claim 1, wherein the step of constructing a spatial index map, mapping and associating the subset of structural features and the subset of spectral features according to spatial positions at the time of acquisition, comprises: establishing a three-dimensional virtual grid coordinate system by taking the actual physical size of the three-dimensional planting frame as a reference, wherein the three-dimensional virtual grid coordinate system divides a planting space into unit grids with equal volumes; according to the space coordinates in the space-time tag, distributing the feature data in the structural feature subset and the spectrum feature subset to the unit grid to which the structural feature subset and the spectrum feature subset belong; The method comprises the steps of sequencing and linking structural feature data and spectral feature data which are distributed in the same unit grid and come from different acquisition moments according to acquisition time, so as to form a feature evolution log in the unit grid; And establishing topological connection relations among the unit grids to form a spatial index map reflecting the adjacent relation of plants in the three-dimensional space.
6. The intelligent evaluation system for three-dimensional plant growth state based on image recognition according to claim 5, wherein the step of locating individual coordinates of plants to be evaluated for growth state in the spatial index map according to the associated structural feature subset and spectral feature subset comprises: traversing each unit grid in the spatial index map, and checking the data integrity of a characteristic evolution log in the unit grid; screening out cell grids in which the structural characteristic data and the spectral characteristic data continuously exist in a plurality of continuous acquisition periods, and marking the cell grids as active plant cells; performing cluster analysis on a plurality of adjacent active plant units in space positions, and combining a plurality of active plant units belonging to the same plant into a plant individual cluster; and calculating the three-dimensional geometric center of each plant individual cluster, and defining the coordinates of the three-dimensional geometric center as plant individual coordinates of the plant individuals.
7. The intelligent evaluation system for three-dimensional planting growth state based on image recognition according to claim 1, wherein the step of retrospectively extracting a structural feature subset and a spectral feature subset of individual plant coordinates in a history acquisition period to form a time sequence growth feature chain of the individual plants comprises the steps of: inquiring the spatial index map according to the plant individual coordinates, and determining all active plant units corresponding to the plant individual clusters to which the plant individual coordinates belong; Extracting structural feature data and spectral feature data recorded at all historical acquisition moments from feature evolution logs of each active plant unit in time sequence; arranging the extracted structural feature data according to a time line to generate a structural feature evolution sequence; The extracted spectral feature data are arranged according to the same time line, and a spectral feature evolution sequence is generated; The structural characteristic evolution sequence data and the spectral characteristic evolution sequence data at the same time point are paired to form a time-ordered characteristic data pair sequence, so as to form a time-sequence growth characteristic chain.
8. The intelligent evaluation system for three-dimensional planting growth state based on image recognition according to claim 1, wherein the step of analyzing the change pattern of the time-series growth characteristic chain to identify an abnormal evolution phase in which the characteristic evolution rate exceeds a preset normal range comprises: Calculating the characteristic value change rate between adjacent time points for the structural characteristic evolution sequence in the time sequence growth characteristic chain to obtain a structural change rate sequence; calculating the characteristic value change rate between adjacent time points for the spectrum characteristic evolution sequence in the time sequence growth characteristic chain to obtain a spectrum change rate sequence; setting normal fluctuation thresholds of the structure change rate and the spectrum change rate respectively; marking a period of which the absolute value of the change rate continuously exceeds a structural normal fluctuation threshold value in the structural change rate sequence as a structural abnormality candidate period; In the spectrum change rate sequence, marking a period of which the absolute value of the change rate continuously exceeds a spectrum normal fluctuation threshold value as a spectrum abnormal candidate period; and taking the intersection of the structural anomaly candidate period and the spectral anomaly candidate period, and confirming the intersection as an anomaly evolution stage.
9. The intelligent evaluation system for the growth state of stereoscopic planting based on image recognition according to claim 1, wherein the step of starting the cross-validation process of the illumination and nutrient supply history record comprises the steps of: The illumination intensity historical record and the nutrient solution concentration historical record which completely correspond to the abnormal evolution stage in time are called from an environment monitoring database; Comparing the illumination intensity historical record with a preset optimal illumination intensity curve required by the current plant type to be evaluated, and calculating an illumination deviation index; comparing the nutrient solution concentration history record with a preset optimal nutrient solution concentration curve required by the type of the plant to be evaluated at present, and calculating a nutrient supply deviation index; And analyzing the cooperative change relation of the illumination deviation index and the nutrient supply deviation index in the abnormal evolution stage to generate an environmental factor influence analysis report.
10. The intelligent evaluation system for three-dimensional planting growth state based on image recognition according to claim 1, wherein the step of merging the characteristic data of the abnormal evolution stage and the result data of the cross-validation process to calculate the degree of deviation of the growth state for a specific plant individual comprises: Extracting actual observation values of structural features and spectral features from a time sequence growth feature chain corresponding to the abnormal evolution stage; obtaining standard reference values of structural features and spectral features corresponding to plant types and growth periods; Respectively calculating the difference between the actual observed value of the structural feature and the standard reference value and the difference between the actual observed value of the spectral feature and the standard reference value; Weighting the illumination deviation index and the nutrient supply deviation index according to the environmental factor influence analysis report to obtain a comprehensive environmental stress coefficient; And correcting the structural characteristic difference degree and the spectral characteristic difference degree by using the comprehensive environmental stress coefficient, and fusing the corrected difference degree to generate a standard growth state deviation degree value.

Description

Three-dimensional planting growth state intelligent evaluation system based on image recognition Technical Field The invention relates to the technical field of intelligent agriculture and image processing, in particular to an intelligent evaluation system for a three-dimensional planting growth state based on image recognition. Background Currently, in stereoscopic planting scenes, monitoring of crop growth states generally relies on image recognition techniques. In the prior art, a fixed point camera or a regular manual shooting mode is adopted to acquire plant images, and then a computer vision algorithm is utilized to extract and analyze characteristics of single or batch two-dimensional images so as to judge the color, the shape or the estimated biomass of the plants. Such methods treat each image as an independent analysis unit, lack systematic recording and utilization of spatial positional relationships between images, and the analysis results are essentially directed to a two-dimensional image plane, rather than to a specific individual plant in stereo space. Due to the lack of accurate spatial position correlation, the state of the same plant individual in different periods is difficult to stably and accurately track and position in a multi-layer high-density three-dimensional planting frame in the prior art. This results in analysis that is often based on "picture areas" rather than "physical plants", and the analysis results are highly confusing when there is occlusion or positional movement between plants. Meanwhile, the state evaluation is mainly carried out by relying on the characteristics of a single time point in the prior art, the static judgment is realized, dynamic processes such as plant growth rate, characteristic change trend and the like cannot be effectively quantized and analyzed, and automatic association verification of abnormal fluctuation in the growth process and contemporaneous environment management operation is more difficult. Therefore, how to realize the accurate continuous tracking of the individual plant identity in the three-dimensional space and how to perform abnormal diagnosis based on multi-source data verification on the growth dynamic process of the plant individual plant identity becomes a key problem for improving the intelligent management precision of the three-dimensional planting. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an intelligent evaluation system for the growth state of three-dimensional planting based on image recognition. In order to achieve the purpose, the invention adopts the following technical scheme that the three-dimensional planting growth state intelligent evaluation system based on image recognition comprises: The acquisition and preprocessing module is used for carrying out dynamic scanning around the three-dimensional planting frame by the image acquisition device, generating an original image cluster containing space position information, and carrying out invalid region rejection on the original image cluster to obtain a purified image cluster; The feature decoupling and mapping module is used for performing growth feature decoupling on the purified image cluster, separating a structural feature subset representing plant morphology and a spectral feature subset reflecting plant physiology, constructing a spatial index map, and mapping and correlating the structural feature subset and the spectral feature subset according to the spatial position when the structural feature subset and the spectral feature subset are acquired; The feature extraction module is used for positioning plant individual coordinates to be evaluated in the growth state in the spatial index map according to the associated structural feature subsets and spectrum feature subsets, and extracting the structural feature subsets and spectrum feature subsets of the plant individual coordinates in a history acquisition period in a retrospective way to form a time sequence growth feature chain of the plant individual; The detection and verification module is used for analyzing the change mode of the time sequence growth characteristic chain, identifying an abnormal evolution stage with the characteristic evolution rate exceeding a preset normal range, and starting a cross verification process of the illumination and nutrient supply history record; and the evaluation output module is used for fusing the characteristic data of the abnormal evolution stage and the result data of the cross verification process, calculating to obtain the deviation degree of the growth state of the specific plant individual, integrating the deviation degree of all the plant individuals and the overall environmental parameters, and outputting a hierarchical evaluation report. Preferably, the step of dynamically scanning by the image acquisition device around the stereoscopic planting frame to generate an original image cluster containing spati