An intelligent identification system for apple tree growth status

Abstract

The present invention relates to the field of state recognition technology, specifically to an intelligent recognition system for the growth state of apple trees. The system includes a branch structure recognition module, a direction sequence calculation module, a growth stability judgment module, a fruit trend steering module, and a vector trajectory offset module. The present invention, by constructing an internode main axis direction vector sequence and quantifying its angle change trend, can identify the continuous change characteristics of the branch direction in the time series dimension, combines the synchronous detection of the fruit diameter growth rate and the color difference change trend, and introduces the linkage change of reflectivity and morphological symmetry as the judgment condition of the trend turning point, effectively locates the fruit growth mutation node, constructs a growth trajectory sequence through the three-dimensional vectors of multi-cycle branch angle increase, diameter growth rate and leaf area index, accurately identifies the individual growth direction offset area, enhances the accuracy and timeliness of growth anomaly identification, and effectively supports the implementation of intelligent management of fruit trees based on the linkage relationship between structure and fruit parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of state recognition, and in particular to an intelligent recognition system for the growth state of an apple tree. Background Art

[0002] The field of state recognition technology involves collecting, modeling, analyzing, and judging the state of a target object at a specific time or environment, encompassing methods such as image recognition, video analysis, sensor data analysis, edge computing modeling, and deep learning. The core of this field lies in the construction of mathematical or neural network models to classify, grade, or identify trends in the operational, growth, or behavioral states of entities. This field is widely used in scenarios such as intelligent agriculture, industrial automation, health monitoring, and traffic surveillance. It relies on a clear system of state characteristic parameters, such as morphology, color, texture, and dynamic changes, as well as highly interpretable model training schemes, to ensure accurate, robust, and real-time state recognition.

[0003] The Apple Tree Growth Status Intelligent Identification System utilizes image recognition and other methods to intelligently classify and determine the status of apple trees throughout their different growth cycles. By continuously collecting image information of apple trees during key growth stages and integrating it with an established growth status recognition model, the system enables intelligent monitoring and assessment of apple tree growth stages. This allows orchard managers to accurately formulate agricultural operation plans for irrigation, fertilization, pest control, and other areas based on the recognition results, thereby improving fruit tree management efficiency and fruit quality, and promoting refined agricultural management.

[0004] During the recognition process, traditional recognition systems mostly rely on static parameters such as morphology and color or overall state characteristics for model training. They lack quantitative analysis of the continuous change trend of branch internode direction and analysis of dynamic time-series angle differences. As a result, when faced with complex situations such as structural abnormalities or growth path deviations, they cannot fully reflect the changing characteristics of local growth trajectories. The judgment of fruit status is only based on the absolute values ​​of size and color indicators, and fails to integrate their change rate and the temporal correlation between them and the stability of branch structure, resulting in a lag in the recognition of sudden changes in fruit growth, affecting the early warning capabilities and accurate deployment of management plans for abnormal growth areas of fruit trees. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent identification system for the growth status of apple trees.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an intelligent identification system for the growth status of apple trees, the system comprising:

[0007] The branch structure recognition module obtains branch image data of the target individual apple tree during the observation period, extracts the contour boundary of the branch internode area frame by frame, extracts the coordinate values ​​of the starting and ending points of each internode axis, constructs the main axis direction vector of each internode, and generates a sequence of internode main axis direction vectors;

[0008] The direction sequence calculation module calls the internode main axis direction vector sequence, constructs the internode direction change angle sequence, marks the turning point, obtains the angle reversal frequency and the angle change average amplitude in the path, and generates the branch direction fluctuation characteristic index;

[0009] The growth stability judgment module sets a branch structure stability threshold interval based on the branch direction fluctuation characteristic index, performs dynamic time warping distance calculation on the direction change angle sequence and the corresponding sequence in the sample, marks the unstable branches, and obtains the branch structure growth stability judgment result;

[0010] The fruit trend turning module is based on the branch structure growth stability judgment result, collects the average diameter value and color difference value of the fruit during the observation period, selects the continuously increasing section of the average diameter change rate, locates the trend turning point, and obtains the mutation node distribution result of the fruit growth stage.

[0011] The present invention has the following improvements: the internode main axis direction vector sequence includes the internode starting coordinate value, the internode ending coordinate value and the internode direction unit vector; the branch direction fluctuation characteristic index includes the direction change amplitude sequence, the angle reversal frequency value and the angle change amplitude mean; the branch structure growth stability judgment result is specifically a branch stability state label, a direction trend matching label and a sample interval attribution label; the fruit growth stage mutation node distribution result includes the trend turning time point, the fruit reflectivity change point and the fruit morphology symmetry variation point.

[0012] The present invention is improved in that the branch structure identification module includes:

[0013] The image boundary extraction submodule obtains the branch image data of the target individual apple tree during the observation period, collects the pixel area information of the branch part in the image, and extracts the boundary pixel values ​​in the grayscale gradient direction within the corresponding pixel distribution range. Based on the connectivity between the boundary pixels, the contour boundary box of each internode area is constructed, and the coordinate set of the contour boundary in the image coordinate system is established to generate the internode area boundary coordinate set;

[0014] The internode axis construction submodule extracts the farthest point pair in the boundary set based on the internode region boundary coordinate set and calculates the Euclidean distance of the farthest point pair, selects the connecting line corresponding to the maximum distance as the longest internode boundary line, and calibrates the starting point and the ending point of the line in the image coordinate system, records the connecting line as the internode axis, and generates an internode axis coordinate sequence;

[0015] The direction vector generation submodule calculates the difference between the starting point and the ending point coordinate values ​​of each internode axis according to the internode axis coordinate sequence and constructs a two-dimensional vector, calculates the unit direction of the vector and standardizes it into a unit vector, arranges all unit direction vectors in sequence according to the internode order, establishes the internode direction change sequence, and generates the internode main axis direction vector sequence.

[0016] The present invention is improved in that the direction sequence calculation module includes:

[0017] The angle calculation submodule calls the internode main axis direction vector sequence, extracts the vector component values ​​of the direction vectors of two adjacent internodes and calculates the cosine value of the angle, reversely deduces the angle angle based on the cosine value and records the positive and negative change direction of the angle, and sequentially arranges the angle change values ​​of each internode to generate an internode angle change value sequence;

[0018] The angle change extraction submodule performs a first-order difference operation on two adjacent angle change values ​​according to the internode angle change value sequence and records the difference direction, compares the positive and negative directions between the differences item by item and marks the direction turning position, records the direction turning point as the segment node position in the sequence, and generates a direction turning position sequence;

[0019] The fluctuation feature extraction submodule calls the direction turning position sequence, counts the number of consecutive direction reversals in the sequence as the angle reversal frequency value, and calculates the arithmetic mean of each absolute value in the original internode angle change value sequence as the angle change amplitude, establishes a joint structural fluctuation feature including the reversal frequency and the change mean, and generates a branch direction fluctuation feature index.

[0020] The present invention is improved in that the growth stability judgment module includes:

[0021] The sample index extraction submodule obtains healthy apple tree samples at the same growth stage based on the branch direction fluctuation characteristic index, extracts the direction reversal frequency value and the mean direction change amplitude of the branches in the samples, organizes the two types of index values ​​into independent data sequences according to the sample numbers, extracts the boundaries based on the minimum and maximum values ​​of each type of data in the sample sequence, and generates the branch structure characteristic interval range;

[0022] The threshold interval construction submodule calls the branch structure feature interval range, sets the upper and lower limits of the reversal frequency interval and the amplitude mean interval according to the two types of interval boundary values, obtains the direction reversal frequency and the change mean amplitude of the target individual, calculates the difference between the target value and the sample boundary value, and calculates the branch structure stability deviation degree value to obtain the branch structure stability deviation information;

[0023] The stability state judgment submodule compares the stability deviation value of each branch with the set stability judgment threshold based on the branch structure stability deviation information, and obtains the dynamic time regularization distance between the target branch direction change angle sequence and the healthy sample sequence. It judges whether any of the two judgment criteria exceeds the limit value. If any condition is met, it is marked as an unstable state to obtain the branch structure growth stability judgment result.

[0024] The present invention is improved in that the fruit trend steering module includes:

[0025] The diameter change extraction submodule collects the average diameter data and color difference value data of the fruit at the corresponding fruit-bearing position under multiple observation cycles based on the branch structure growth stability determination result and branches marked as unstable, calculates the average diameter change rate between each two cycles, and selects the continuously increasing time intervals to obtain the sequence change value of the average diameter change rate within the segment, thereby generating a fruit diameter change trend sequence;

[0026] The trend synchronization detection submodule calls the fruit diameter change trend sequence, performs first-order difference processing on the two sequences according to the color difference value change sequence of the corresponding period, calculates the difference in the difference direction of each period, marks the position where the difference direction is reversed at the same time as the trend intersection position, calculates the change aggregation intensity value within the turning interval, screens candidate nodes according to the set intensity threshold, and obtains the trend intersection identification result;

[0027] The mutation node identification submodule performs threshold judgment on the change intensity and trend reversal direction of the candidate nodes in each cycle based on the trend intersection identification results, aggregates and outputs the node positions that meet the two criteria, and screens the nodes whose reflectivity change values ​​and symmetry change values ​​simultaneously reach the set sensitivity level to obtain the mutation node distribution results of the fruit growth stage.

[0028] The present invention is improved in that the system further comprises:

[0029] The vector trajectory offset module extracts the average diameter growth rate, branch extension angle increase, and leaf area index change of each mutation node in the period before and after the mutation node based on the distribution results of the mutation nodes in the fruit growth stage, continuously constructs time series vector trajectories of multiple periods, records the direction offset segments, and generates the identification results of the growth direction offset areas of individual apple trees;

[0030] The identification results of the individual growth direction deviation areas of the apple trees specifically refer to the deviation duration period segments, angle change trend records and direction consistency abnormal segments.

[0031] The present invention is improved in that the vector trajectory offset module includes:

[0032] The three-parameter extraction submodule extracts the average fruit diameter growth rate, branch extension angle increase, and leaf area index change in the consecutive periods before and after each mutation node based on the distribution results of the mutation nodes in the fruit growth stage, and normalizes the three parameters in the same time period, generates a standardized vector for each set of data, and obtains a three-parameter standard vector sequence;

[0033] The vector trajectory generation submodule calls the three-parameter standard vector sequence, calculates the angle and cosine similarity between the standard vector of each cycle and the set target growth direction reference vector, extracts the angle change rate and similarity decrease trend in consecutive cycles, and calculates the direction deviation trend value;

[0034] The offset segment identification submodule compares the change value in the continuous cycle with the direction offset threshold according to the direction offset trend value, marks the period segment with continuously increasing trend value, and identifies whether the offset angle continues to increase in adjacent multiple cycles. The continuous segments that meet the conditions are marked to establish the individual growth direction offset area identification results of apple trees.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are:

[0036] In the present invention, by constructing a sequence of internode main axis direction vectors and quantifying their angle change trends, the continuous change characteristics of branch direction in the time series dimension can be identified, and a structural judgment of the branch growth trend is achieved based on the first-order difference and angle reversal frequency indicators. Combined with the synchronous detection of the fruit diameter growth rate and the color difference change trend, and the introduction of the linkage change of reflectivity and morphological symmetry as the judgment condition of the trend turning point, the fruit growth mutation node is effectively located, and a growth trajectory sequence is constructed through the three-dimensional vectors of multi-cycle branch angle increase, diameter growth rate and leaf area index. The individual growth direction deviation area is accurately identified, the accuracy and timeliness of growth abnormality identification are enhanced, and the implementation of intelligent management of fruit trees based on the linkage relationship between structure and fruit parameters is effectively supported. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system flow chart of the present invention;

[0038] Figure 2 This is a flow chart of the branch structure identification module of the present invention;

[0039] Figure 3 This is a flow chart of the direction sequence calculation module of the present invention;

[0040] Figure 4 This is a flow chart of the growth stability judgment module of the present invention;

[0041] Figure 5 This is a flow chart of the fruit trend steering module of the present invention;

[0042] Figure 6 This is a flow chart of the vector trajectory offset module of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0045] See also Figure 1 ,The present invention provides a technical solution: an intelligent identification system for the growth status of apple trees, the system includes a branch structure identification module, a direction sequence calculation module, a growth stability judgment module, a fruit trend steering module, and a vector trajectory offset module;

[0046] The branch structure recognition module obtains branch image data of the target individual apple tree during the observation period, extracts the contour boundary of the branch internode area frame by frame, establishes the internode axis according to the direction of the longest connecting line of the contour boundary, and extracts the coordinate values ​​of the starting and ending points of each internode axis. The coordinate difference is used to construct the direction vector of each internode main axis, generating a sequence of internode main axis direction vectors.

[0047] The direction sequence calculation module calls the internode main axis direction vector sequence, calculates the angle cosine value of each adjacent internode main axis direction vector, extracts the positive and negative change trends of the angle, sequentially constructs the internode direction change angle sequence, performs first-order difference processing on the direction change angle sequence, marks the turning points, establishes the continuous angle change path of each branch, obtains the angle reversal frequency and angle change average amplitude in the path, and generates the branch direction fluctuation characteristic index;

[0048] The direction change angle sequence is the angle between adjacent vectors; the angle reversal frequency refers to the number of times the positive and negative directions of the angle change appear alternately; the angle change amplitude is the arithmetic mean of the absolute values ​​of the direction change angle sequence;

[0049] The growth stability judgment module obtains a set of branch direction fluctuation characteristic indicators from healthy apple tree samples at the same growth stage based on the branch direction fluctuation characteristic indicators. The maximum and minimum value intervals of the angle reversal frequency and the angle change amplitude in the samples are calculated respectively, and set as the branch structure stability threshold interval. The dynamic time warping distance is calculated between the direction change angle sequence and the corresponding sequence in the sample. The corresponding indicator value of the target individual is judged to belong to the branch structure stability threshold interval. If it is outside the stability threshold interval or the dynamic time warping distance exceeds the set threshold, it is marked as an unstable branch, and the branch structure growth stability judgment result is obtained;

[0050] The fruit trend turning module is based on the results of the branch structure growth stability judgment. According to the fruit-bearing positions corresponding to unstable branches, the average diameter and color difference values ​​of the fruit during the observation period are collected. The segments with continuously increasing average diameter change rates are selected, and the first-order differences are performed on each segment. The change trend and the color difference value change trend are synchronously tested, and the trend turning time points of the two are extracted. If the surface reflectance and morphological symmetry index of the period near the turning point change together, it is located as a trend turning point, and the distribution results of mutation nodes in the fruit growth stage are obtained.

[0051] The average diameter change rate is the diameter increment per unit time, usually obtained through multi-cycle measurement; the color difference value uses the a or b dimension of the internationally commonly used CIELAB color space indicator; the symmetry axis projection error represents the mean deviation distance of the image morphology from the central axis;

[0052] The vector trajectory offset module extracts the average diameter growth rate, branch extension angle increase, and leaf area index change in the cycles before and after each mutation node based on the distribution of mutation nodes during the fruit growth stage. These three parameters are standardized within the same time period to form a three-dimensional directional vector. Time series vector trajectories for multiple cycles are continuously constructed and compared with the set target growth direction reference vector for each cycle's cosine similarity and angle change trend. If the cosine value decreases and the angle increases in consecutive cycles, it is recorded as a direction offset segment, generating the individual apple tree growth direction offset area identification result.

[0053] The target growth direction reference vector was constructed based on the average of three indicators in healthy samples; the branch extension angle increase was obtained by the node angle difference; the leaf area index is a standard physiological and ecological parameter that represents the total leaf area per unit surface area;

[0054] The internode main axis direction vector sequence includes the internode starting coordinate value, the internode ending coordinate value and the internode direction unit vector. The branch direction fluctuation characteristic indicators include the direction change amplitude sequence, the angle reversal frequency value and the angle change amplitude mean. The branch structure growth stability judgment results are specifically the branch stability state label, the direction trend matching label and the sample interval attribution label. The mutation node distribution results of the fruit growth stage include the trend turning time point, the fruit reflectivity change point and the fruit morphological symmetry variation point. The individual growth direction deviation area identification results of apple trees specifically refer to the deviation continuous period segment, the angle change trend record and the direction consistency abnormal segment.

[0055] See also Figure 2 , the branch structure recognition module includes:

[0056] The image boundary extraction submodule obtains the branch image data of the target individual apple tree during the observation period, collects the pixel area information of the branch part in the image, and extracts the boundary pixel values ​​in the grayscale gradient direction within the corresponding pixel distribution range. Based on the connectivity between the boundary pixels, the contour boundary box of each internode area is constructed, and the coordinate set of the contour boundary in the image coordinate system is established to generate the internode area boundary coordinate set;

[0057] The branch image data of the target individual apple tree is obtained during the observation period. It is usually captured once an hour based on a fixed-point acquisition device. The captured images are stored in the form of standard RGB images. After selecting the branch area in the image, the pixel grayscale difference method is used to calculate the grayscale gradient value of each pixel in the horizontal and vertical directions. For any pixel, a difference operator matrix is ​​constructed based on its upper, lower, left, and right adjacent points, and its directional maximum and amplitude values ​​are recorded. For example, the grayscale gradient value of the pixel at the 150th row and 120th column in image IMG001 is 22, and its boundary direction is horizontal. Subsequently, all pixels whose gradient amplitude values ​​exceed the set threshold are marked as boundary candidate points. The boundary recognition threshold of the grayscale gradient value is set to 20. According to the standard under the image resolution of 720×480, if the grayscale difference between the branch and the background is less than 15, it will cause false boundary overlap, and if it exceeds 25, it may introduce interference from the leaf area. Therefore, the median value of 20 is taken as the stable extraction benchmark. This value is constructed by the mean weighted deviation of the gradient statistical distribution and is set to τ = μ G +0.5σ G , μ G is the boundary gradient mean, σ Gis the standard deviation of the boundary gradient, τ is the boundary recognition threshold. When the mean boundary gradient of a single image is 18.2 and the standard deviation is 3.5, the calculated threshold is τ = 18.2 + 0.5 × 3.5 = 19.95, rounded to 20. According to the grayscale distribution connectivity rule, these pixels are connected according to the 8-neighborhood structure to form several closed pixel sets. Each closed set is converted into a minimum circumscribed rectangular box, which is further projected into the image coordinate system to construct the four-point coordinate information of the bounding box. The projection coordinate value of each bounding box is marked according to the internode number, and finally the internode region boundary coordinate set is formed.

[0058] The internode axis construction submodule extracts the farthest point pair in the boundary set based on the internode region boundary coordinate set and calculates the Euclidean distance of the farthest point pair. It selects the line corresponding to the maximum distance as the longest internode boundary line, calibrates the starting and ending points of the line in the image coordinate system, records the line as the internode axis, and generates the internode axis coordinate sequence.

[0059] Based on the internode region boundary coordinate set, the Euclidean distance between the coordinates of the four points of each bounding box is extracted, and 6 groups of point pair calculation results are constructed for each internode region. For example, the four points in the region with internode number 1 are A(102,220), B(160,260), C(102,260), and D(160,220), then there are 6 groups of distances AB, AC, AD, BC, BD, and CD. The Euclidean distance formula is used to calculate the distance between the four points. Calculation, (x1, y1) and (x2, y2) are the coordinates of two points respectively, D is the Euclidean distance between the two points, such as AB corresponding value is 72.11px, select the largest one as the longest boundary line of the internode, and mark the corresponding point pair as the starting and ending points of the axis of the internode. After repeating the operation of each internode, establish the axis coordinate pair sequence corresponding to the internode number, and store it uniformly in the internode axis coordinate sequence. The direction vector generation submodule calculates the difference between the starting point and the ending point coordinate values ​​of each internode axis according to the internode axis coordinate sequence and constructs a two-dimensional vector. The vector unit direction is calculated and standardized as a unit vector. All unit direction vectors are arranged in sequence according to the internode order, and the direction change sequence of the internode is established to generate the internode main axis direction vector sequence;

[0060] According to the internode axis coordinate sequence, for each axis vector, perform vector difference construction based on the starting coordinates (x1, y1) and the ending coordinates (x2, y2) If the starting and ending points in interval 1 are (102, 220) and (160, 260), then the vector is (58, 40), and the calculated unit vector direction is:

[0061]

[0062] Combine all internode unit direction vectors into a matrix sequence in image order In this way, the direction change path of the main axis structure of the apple tree branches during the observation period is established, and finally the internode main axis direction vector sequence is generated, in which the direction vector normalization processing standard is the unit norm normalization method. According to the definition That is, unit length projection, which avoids angle distortion caused by different scales in subsequent angle calculations, is the core processing basis for direction standardization.

[0063] See also Figure 3 , the direction sequence calculation module includes:

[0064] The angle calculation submodule calls the internode main axis direction vector sequence, extracts the vector component values ​​of the direction vectors of two adjacent internodes and calculates the cosine value of the angle. It reversely deduces the angle angle based on the cosine value and records the positive and negative change direction of the angle. It sorts the angle change values ​​of each internode in sequence to generate a sequence of internode angle change values.

[0065] Call the internode main axis direction vector sequence, extract the component values ​​of the direction vectors of any two adjacent internodes respectively, and use vector and Represents the direction vector between internode t and internode t+1, by calculating its dot product and modulus product, using the formula Get the cosine value of the included angle, and use this value to infer the included angle as θ t =arccos(cosθ t ) and judge the angle increase or decrease trend, define the angle positive change as clockwise offset, negative change as counterclockwise offset, and use the starting vector as the reference system for judgment. If θ t >0 and the front-back vector product is less than zero, it is a reverse offset, otherwise it is a forward offset. Taking image IMG001 as an example, the direction vector of internode 1 is (0.80, 0.55), and that of internode 2 is (0.71, 0.71). The dot product is 0.80·0.71+0.55·0.71=0.9925, the modulus-length product is 1, the angle cosine value is 0.9925, and the corresponding angle is about 7.1°. The recorded change direction is positive. The angle difference of each pair of adjacent internodes is sequentially added to the queue to finally generate a sequence of internode angle change values.

[0066] The angle change extraction submodule performs a first-order difference operation on two adjacent angle change values ​​based on the internode angle change value sequence and records the difference direction. It compares the positive and negative directions of the differences item by item and marks the direction turning position. The direction turning point is recorded as the segment node position in the sequence to generate a direction turning position sequence.

[0067] According to the sequence of internode angle changes, θ t Represents the angle difference between nodes t. Difference is performed on every two consecutive angle values ​​to construct the sequence Δθ t =θ t+1 -θt , if the internode sequence angle value is [7.1°, 12.5°, 9.2°, 11.4°], the difference result is [5.4°,

[0068] -3.3°, 2.2°], determine the positive and negative direction of the difference value and record the direction changes in sequence. The identification rule of the direction turning point is: the current difference direction is opposite to the direction of the previous cycle, such as changing from positive to negative or negative to positive is considered a turning point. This rule is based on the sign difference calculation method, that is, if sign(Δθ t )≠sign(Δθ t-1 ), then the t position is recorded as the turning point. Taking the above sequence as an example, the second difference is -3.3°, and the previous term is 5.4°, the direction turning is established, and the recording position is t=2. The time index position of each turning point is recorded in the direction turning position sequence, and finally the sequence is formed;

[0069] The fluctuation feature extraction submodule calls the direction turning position sequence, counts the number of consecutive direction reversals in the sequence as the angle reversal frequency value, and calculates the arithmetic mean of each absolute value in the original internode angle change value sequence as the angle change amplitude. It establishes a joint structural fluctuation feature that includes the reversal frequency and the change mean, and generates a branch direction fluctuation feature index.

[0070] Call the direction turning position sequence, count the number of consecutive direction reversals as the angle reversal frequency value, if the turning position sequence is [2,4,6,7], then the reversal frequency is 4 times, and the corresponding frequency value can be standardized to the number of occurrences per 10 internode units, that is, the frequency value is 0.4, then take the absolute value of each item in the internode angle change value sequence and calculate its arithmetic mean. If the angle value is [7.1°, 12.5°, 9.2°, 11.4°], then The mean is used as the representative value of the angle variation, where the average value is set as the reference value of the angle variation. Its value reflects the overall fluctuation level of the structure in the process of directional evolution. It is standardized for subsequent cross-image comparison. The normalization method is the relative maximum angle ratio μ θ / θ max , μ θ is the reference value of the angle change amplitude, θ max is the maximum angle value. Taking the maximum angle in the current figure as 14.2°, the normalized value is 10.05 / 14.2=0.708. Finally, the reversal frequency value and the average amplitude of the angle change are combined as feature pairs to form a structural fluctuation characteristic index and generate a branch direction fluctuation characteristic index.

[0071] See also Figure 4 , the growth stability judgment module includes:

[0072] The sample index extraction submodule obtains healthy apple tree samples at the same growth stage based on the branch direction fluctuation characteristic index, extracts the direction reversal frequency value and the mean direction change amplitude of the branches in the samples, organizes the two types of index values ​​into independent data sequences according to the sample number, and extracts the boundaries based on the minimum and maximum values ​​of each type of data in the sample sequence to generate the branch structure characteristic interval range;

[0073] A total of 20 healthy apple trees at the same growth stage were collected, with each tree containing between 15 and 20 branches. The frequency of directional reversals and the mean amplitude of directional changes were extracted for all branches. The frequency of directional reversals was calculated by counting the number of reversals in the directional position sequence and normalizing it to a frequency per 10 internodes. The mean amplitude of directional changes was calculated by calculating the absolute value of each item in the internode angle change sequence and taking the arithmetic mean. The sample was assumed to have three reversals, 12 internodes, a reversal frequency normalization value of 0.25, and an angle change range of [8.2°, 9.7°, 7.5°, 10.1°], with a mean amplitude of 8.875°. The two types of indicators of each branch in each sample were summarized according to the sample number to organize two independent data series. The boundaries were extracted by comparing the minimum and maximum values ​​in each series. For example, the inversion frequency interval was [0.18, 0.43], and the amplitude mean interval was [6.8°, 11.4°], which were used as the upper and lower limits of the characteristic values, respectively. The setting of the boundary was based on the 95% coverage interval of the healthy sample range. The extraction was done after removing abnormal extreme values ​​from the original sample to ensure that the proposed interval was statistically representative and practically reasonable. As the number of samples increased, the boundary range tended to stabilize. The range was not treated as a constant value but was dynamically adjusted as the sample stage was updated, and finally the branch structure characteristic interval range was generated.

[0074] The threshold interval construction submodule calls the branch structure feature interval range, sets the upper and lower limits of the reversal frequency interval and the amplitude mean interval according to the two types of interval boundary values, obtains the direction reversal frequency and the change mean amplitude of the target individual, and calculates the difference between the target value and the sample boundary value using the formula:

[0075]

[0076] Calculate and obtain the branch structure stability deviation degree value to obtain the branch structure stability deviation information;

[0077] Where f represents the normalized value of the inversion frequency of the target branch, represents the normalized value of the average reversal frequency of the sample branches, m represents the normalized value of the average amplitude of the direction change of the target branches, Represents the normalized value of the average amplitude of the direction change of the sample branches, r j represents the normalized value of the jth item in the target branch direction change angle sequence, represents the normalized value of the average value of the jth item in the sample direction change angle sequence, n represents the total number of items in the direction change angle sequence, C represents the structural complexity ratio, which is the ratio of the number of target branch nodes to the average number of nodes, and S represents the degree of deviation of the branch structure stability;

[0078] Call the two types of interval boundary values ​​in the branch structure feature interval range, set the reversal frequency normalization value interval to [0.18, 0.43], the average amplitude interval to [6.8, 11.4], and obtain the direction reversal frequency value of the current target individual branch as 0.31, the direction change amplitude average is 9.5°, and the difference between the two items is calculated. For example, the frequency difference is |0.31-0.30| / 0.30=0.033, and the amplitude difference is |9.5-9.0| / 9.0=0.056, where the sample means are set as The target values ​​are f = 0.31 and m = 9.5 respectively. We further introduce the structural complexity ratio C. Assuming that the current branch contains 17 nodes and the average number of nodes in the sample is 15, then The target branch direction change angle sequence is called as [7.5, 9.2, 8.1, 10.4, 9.9], and the corresponding sample average sequence is [7.2, 9.0, 8.4, 10.1, 9.6]. The square of the third difference is (8.1-8.4) 2 =0.09, sum up all the squared terms of the sequence differences and divide by the number of terms, then take the square root to get:

[0079]

[0080] Substitute the above values ​​into the following formula to calculate the degree of offset:

[0081]

[0082] S=0.033+ln(1.06328)+0.224=0.033+0.0614+0.224=0.3184;

[0083] Where f represents the normalized value of the inversion frequency of the target branch, represents the normalized value of the average reversal frequency of the sample branches, m represents the normalized value of the average amplitude of the direction change of the target branches, represents the normalized value of the average amplitude of the direction change of the sample branches, C represents the structural complexity ratio, which is calculated as the ratio of the current branch node number to the sample average node number, r j Represents the normalized value of the jth item in the target direction change angle sequence, It represents the normalized mean of the jth item of the sample, n represents the total number of items in the sequence, and S represents the degree of structural stability deviation, which reflects the level of difference between the current branch and the sample interval.

[0084] The benefit of the formula is that by incorporating the differences in the distribution of frequency, amplitude and angle sequences into the same deviation calculation model and introducing a structural complexity weight factor, the ability to perceive the deformation of complex structures is enhanced, thereby accurately evaluating the deviation trend of structural stability.

[0085] The result shows that the degree of structural deviation is 0.3184, which is lower than the preset stability threshold of 0.45. Combined with the set judgment logic, the branch is judged to be in a structurally stable state. This value is directly used as the core judgment basis in the subsequent stability state judgment submodule.

[0086] The stability state judgment submodule compares the stability deviation value of each branch with the set stability judgment threshold based on the branch structure stability deviation information, and obtains the dynamic time warping distance between the target branch direction change angle sequence and the healthy sample sequence. It judges whether any of the two judgment criteria exceeds the limit value. If any condition is met, it is marked as unstable state, and the branch structure growth stability judgment result is obtained;

[0087] Based on the branch structure stability offset information, a stability judgment threshold of 0.45 is set. The current offset value of all branches is compared with the threshold. If the offset value is higher than the threshold, the structure is judged to be unstable. At the same time, the system obtains the dynamic time warping (DTW) distance between the target branch direction change angle sequence and the average sequence of healthy samples. A matrix is ​​established with the cycle length as the step size to calculate the cumulative cost path. The DTW distance threshold is set to 8.5. If the current calculated DTW is 9.2, it exceeds the judgment interval. The dual judgment logic of this step is: if either the offset value is greater than 0.45 or the DTW distance is greater than 8.5, it is unstable. The offset threshold in the judgment logic refers to the 90th percentile of the sample offset value distribution. The DTW distance is set to be obtained based on the average DTW distance within the sample + 2 times the standard deviation, which has a stable data foundation. For example, if the current branch offset value is 0.3184, which is lower than the threshold, but the DTW is 9.2, which is higher than the limit value, it is finally judged to be unstable. The result of the branch structure growth stability judgment is unstable.

[0088] See also Figure 5 , the fruit trend steering module includes:

[0089] The diameter change extraction submodule is based on the results of the branch structure growth stability determination. According to the branches marked as unstable, it collects the average fruit diameter data and color difference value data of the corresponding fruit-bearing position under multiple observation cycles, calculates the average diameter change rate between each two cycles, and selects the continuously increasing time intervals to obtain the sequence change value of the average diameter change rate within the segment, thereby generating a fruit diameter change trend sequence.

[0090] According to the results of the branch structure growth stability determination, the branches marked as unstable, such as No. B-12, B-14, etc., collected the average diameter data and color difference value data of the fruit corresponding to the fruit-bearing position of these branches in the observation period T1 to T8. The average diameter unit is mm, and the color difference value is normalized after conversion using the CIE76 index standardization. Taking No. B-12 as an example, the average diameter values ​​of its fruit-bearing area in each period are [61.5, 62.3, 63.0, 64.1, 65.5, 66.8, 67.2, 67.3], using the formula Calculate the average diameter change rate, d i+1 and d i are the average diameter values ​​of the two cycles, Δd i The average diameter change rate is obtained, and the change rates of adjacent periods are [0.013, 0.011, 0.017, 0.022, 0.020, 0.006, 0.001]. The continuous increasing segments are screened and T2 to T5 are obtained as the segment with increasing change rate. The trend sequence of the average diameter change rate sequence in this segment is constructed to obtain:

[0091] [0.011, 0.017, 0.022]. The screening criterion is that the change rate values ​​within three consecutive cycles meet the increasing relationship of the previous one being less than the next one, reflecting the continuous expansion state of the fruit growth stage, and finally generating the fruit diameter change trend sequence.

[0092] The trend synchronization detection submodule calls the fruit diameter change trend sequence, performs first-order difference processing on the two sequences according to the color difference value change sequence of the corresponding period, and calculates the difference in the difference direction of each period. The position where the difference direction is reversed at the same time is marked as the trend intersection position, using the formula:

[0093]

[0094] The calculation obtains the change aggregation intensity value within the turning interval, filters the candidate nodes according to the set intensity threshold, and obtains the trend intersection identification result;

[0095] Among them, d1 represents the normalized value of the average diameter of the cycle before the turn, d2 represents the normalized value of the average diameter of the cycle after the turn, c1 and c2 represent the normalized values ​​of the color difference before and after the turn, respectively, s i represents the normalized value of the fruit surface reflectance at the i-th time point, represents the normalized value of the sample mean reflectance at the i-th time point, N represents the total number of reflectance measurement points, ε represents the minimum perturbation constant to avoid the denominator approaching zero and causing extreme value instability, and P represents the change aggregation intensity value;

[0096] Call the fruit diameter change trend sequence generated by the previous module and the color difference value change sequence observed in the segment, such as [2.3, 2.8, 3.5], perform first-order difference processing, for example, the diameter difference sequence is [0.006, 0.005], and the color difference difference sequence is [0.5, 0.7]. Calculate the difference direction change respectively. If the two types of sequences are synchronously reversed in the difference direction of a certain cycle, it is recorded as the trend intersection position. For example, the T4 cycle corresponds to the trend reversal point. Further introduce the fruit morphological change index before and after the cycle. Set the T3 morphological index to 0.72, T4 to 0.81, the reflectivity difference T3 to 0.63, T4 to 0.71, and calculate the index change value. Use the formula:

[0097]

[0098] Substituting the parameter values, let d1 = 65.5, d2 = 66.8, c1 = 2.3, c2 = 2.8. After normalization, d1 = 0.982, d2 = 1.001, c1 = 0.91, c2 = 0.93, and the difference terms are |d1-d2| = 0.019, |c1-c2| = 0.02. The morphological reflectance term is:

[0099]

[0100] Finally, put it into the formula:

[0101]

[0102] The intensity threshold is set to 0.01. According to the threshold standard, the P value is greater than the threshold, so it is marked as a candidate trend intersection point, and the trend intersection point identification result is obtained.

[0103] Table 1 Fruit characteristic change parameters

[0104] cycle Average diameter (mm) Color difference value (NBS) Reflectivity Morphological indicators T3 65.5 2.3 0.63 0.72 T4 66.8 2.8 0.71 0.81

[0105] As shown in Table 1, the T3 to T4 period shows trend changes in diameter, color difference, reflectivity and morphology, and the synchronous reversal point has the characteristics of indicator aggregation.

[0106] Among them, d1, d2, c1, and c2 represent the average diameter and normalized color difference of the period before and after the turn, respectively. i is the normalized reflectivity value at time point i, is the sample mean normalized value, N is the number of reflectance measurement points, ε is set to 0.001 to avoid the denominator being 0, and P is the change aggregation intensity value, which is used to comprehensively evaluate the joint intensity of trend changes.

[0107] The benefit of the formula is that by jointly introducing the standard deviation and amplitude of three different types of data from different sources, namely diameter, color difference and reflectivity, it measures the synchronization of multiple changes at trend turning points, enhances the sensitivity of turning detection, and reduces the risk of misjudgment of a single indicator.

[0108] The result shows that the change aggregation intensity is 0.01206, which exceeds the set threshold of 0.01, indicating that there is a real trend intersection in this cycle, with significant synchronous changes in indicators, providing basic candidate points for subsequent mutation node screening.

[0109] The mutation node identification submodule uses the trend intersection identification results to perform threshold judgment on the change intensity and trend reversal direction of the candidate nodes in each cycle, aggregates and outputs the node positions that meet the two criteria, and selects nodes whose reflectivity change values ​​and symmetry change values ​​simultaneously reach the set sensitivity level to obtain the mutation node distribution results during the fruit growth stage;

[0110] According to the candidate node positions screened out from the trend intersection identification results, a comprehensive judgment is made on the change intensity and trend reversal direction of each node in its previous and next cycles. The judgment criteria for change intensity are set as whether the P value is higher than 0.01 and whether the trend direction satisfies the double differential sign transformation of color difference and diameter. Further screening is carried out to see whether the reflectivity change of the point exceeds the set reflectivity sensitivity threshold of 0.07 and whether the symmetry index change exceeds the symmetry difference threshold of 0.08. Taking T4 as an example, the reflectivity change in the previous and next cycles is 0.08, and the symmetry change is 0.09, both exceeding the preset thresholds. Therefore, T4 is judged to be a true mutation node. Finally, all points that meet the conditions are integrated into a mutation node group, and the distribution results of mutation nodes in the fruit growth stage are obtained, which serves as a reference for subsequent fruit state structure transition analysis.

[0111] See also Figure 6 , the vector trajectory offset module includes:

[0112] The three-parameter extraction submodule extracts the average fruit diameter growth rate, branch extension angle increase, and leaf area index change in the consecutive periods before and after each mutation node based on the distribution results of the mutation nodes in the fruit growth stage. The three parameters are normalized in the same time period, and a standardized vector is generated for each set of data to obtain a three-parameter standard vector sequence.

[0113] According to the distribution results of mutation nodes during the fruit growth stage, the observation data within four cycles before and after each mutation node were selected, and three parameters, namely, the average fruit diameter growth rate, the branch extension angle increase, and the leaf area index change in the corresponding time period, were extracted. Taking a mutation node T5 as an example, the previous and subsequent cycles are T3, T4, T5, and T6, respectively. The observed values ​​are as follows: the average fruit diameters are 60.1mm, 61.3mm, 62.8mm, and 63.7mm, respectively, with corresponding growth rates of [0.02, 0.024, and 0.014]; the branch extension angles are 29.5°, 33.8°, 36.1°, and 39.4°, respectively, with increases of [4.3, 2.3, and 3.3]; the leaf area index is 2.15, 2.43, 2.66, and 2.88, respectively, with changes of [0.28, 0.23, and 0.22]. Min-Max normalization is used for each group of parameters, that is, each data item is mapped to the [0, 1] interval according to its interval range. For example, the normalized calculation method for fruit growth rate is: x is the original value, x min is the minimum value, x max The three types of data are normalized to construct three-dimensional standard vectors. For example, the normalized value corresponding to the T4 period is [0.6, 0.35, 0.54], and the T5 period is [1.0, 0.58, 0.64]. These three-parameter standard vector sequences of the periods near each mutation node are arranged in this order for subsequent directionality detection.

[0114] The vector trajectory generation submodule calls the three-parameter standard vector sequence, calculates the angle and cosine similarity between the standard vector of each cycle and the set target growth direction reference vector, and extracts the angle change rate and similarity decrease trend in consecutive cycles using the formula:

[0115]

[0116] Calculate and obtain the direction deviation trend value;

[0117] Among them, θ t Represents the normalized value of the rate of change of the angle between the standard vector and the reference vector in the tth period, φ t Represents the normalized value of the angle between the standard vector and the reference vector in the tth period, cosφ t represents the cosine similarity value between direction vectors, T represents the total number of cycles, δ represents the perturbation constant of the angle denominator, which is used to avoid numerical instability when the denominator is zero, and Q represents the direction deviation trend value, which is used to measure the deviation degree of directional evolution in the time series;

[0118] Call the three-parameter standard vector sequence obtained in the previous module, set the target growth direction reference vector to the unit vector [0.577, 0.577, 0.577], and calculate the angle and cosine similarity between the standard vector of each period and the reference vector. Taking the T4 to T6 period as an example, the T4 standard vector is [0.6, 0.35, 0.54], and the cosine value is:

[0119]

[0120] Since the cosine value exceeds 1 in the calculation, it is necessary to correct the cosine value. The normalized value should be limited to [0,1]. After adjustment, cosφ4=0.99. Calculate θ in sequence. t =arccos(cosφ t ), and get the angle value, which is used to calculate the rate of change of the angle in consecutive cycles and the decreasing trend of the cosine similarity, and finally the following formula is used:

[0121]

[0122] Assume that the total number of cycles is T = 4, the cycle angles are [7.2°, 10.5°, 13.8°, 15.1°], converted to radians are [0.126, 0.183, 0.241, 0.264], the angle change rates are [0.057, 0.058, 0.023], the cosine similarity values ​​are [0.99, 0.982, 0.974, 0.969], and the difference is [0.01, 0.018, 0.026, 0.031]. Substituting into the formula and setting δ = 0.001, the calculation is as follows:

[0123]

[0124]

[0125] Q=0.8598+0.02125=0.88105;

[0126] Finally, the direction deviation trend value Q = 0.88105;

[0127] Among them, θ t represents the normalized value of the angle between the standard vector and the reference vector in the tth period, φ t Indicates the angle value of the tth cycle, cosφ t is the corresponding cosine similarity, δ is the disturbance term to prevent division by zero error, T is the total number of cycles, and Q is the direction deviation trend value, which measures the intensity of periodic deviation.

[0128] The benefit of the formula is that by introducing the relative ratio of the angle change rate and the average amount of cosine similarity attenuation, the Q value can simultaneously reflect the continuity and accumulation of the directional deviation, thereby improving the robustness and sensitivity of the detection of abnormal growth direction changes.

[0129] The results show that the directional deviation trend in the current observation section is significant, and the Q value is much higher than the set deviation benchmark value of 0.4, indicating that the periodic growth trend deviates from the target vector strongly and should be recorded as an abnormal growth section.

[0130] The offset segment identification submodule compares the change value within consecutive cycles with the directional offset threshold based on the directional offset trend value, marks the periodic segments where the trend value continues to increase, and identifies whether the offset angle continues to increase in adjacent multiple cycles. It marks the continuous segments that meet the conditions and establishes the identification results of the growth direction offset area of ​​individual apple trees.

[0131] According to the direction shift trend value Q = 0.88105 obtained in the previous module, the shift threshold is set to 0.4, and the direction shift trend of this periodic segment is judged to be significant. Further statistics show that the angle change sequence from T3 to T6 is [7.2°, 10.5°, 13.8°, 15.1°], with an obvious continuous increase trend. The angle growth rate is always positive, and the continuity is established. Therefore, this four-periodic segment is marked as a direction shift area, and the identification result of the growth direction shift area of ​​individual apple trees is established for structural pruning and growth guidance applications.

[0132] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent identification system for apple tree growth status, characterized in that: The system comprises: The branch structure recognition module obtains branch image data of the target individual apple tree during the observation period, extracts the contour boundary of the branch internode area frame by frame, extracts the coordinate values ​​of the starting and ending points of each internode axis, constructs the main axis direction vector of each internode, and generates a sequence of internode main axis direction vectors; The direction sequence calculation module calls the internode main axis direction vector sequence, constructs the internode direction change angle sequence, marks the turning point, obtains the angle reversal frequency and the angle change average amplitude in the path, and generates the branch direction fluctuation characteristic index; The growth stability judgment module sets a branch structure stability threshold interval based on the branch direction fluctuation characteristic index, performs dynamic time warping distance calculation on the direction change angle sequence and the corresponding sequence in the sample, marks the unstable branches, and obtains the branch structure growth stability judgment result; The fruit trend turning module collects the average diameter and color difference values ​​of the fruit during the observation period based on the branch structure growth stability judgment result, selects the section with the continuously increasing average diameter change rate, locates the trend turning point, and obtains the distribution results of the mutation nodes in the fruit growth stage; The fruit trend steering module includes: The diameter change extraction submodule collects the average diameter data and color difference value data of the fruit at the corresponding fruit-bearing position under multiple observation cycles based on the branch structure growth stability determination result and branches marked as unstable, calculates the average diameter change rate between each two cycles, and selects the continuously increasing time intervals to obtain the sequence change value of the average diameter change rate within the segment, thereby generating a fruit diameter change trend sequence; The trend synchronization detection submodule calls the fruit diameter change trend sequence, performs first-order difference processing on the two sequences according to the color difference value change sequence of the corresponding period, calculates the difference in the difference direction of each period, marks the position where the difference direction is reversed at the same time as the trend intersection position, calculates the change aggregation intensity value within the turning interval, screens candidate nodes according to the set intensity threshold, and obtains the trend intersection identification result; The mutation node identification submodule performs threshold judgment on the change intensity and trend reversal direction of the candidate nodes in each cycle based on the trend intersection identification results, aggregates and outputs the node positions that meet the two criteria, and screens the nodes whose reflectivity change values ​​and symmetry change values ​​simultaneously reach the set sensitivity level to obtain the mutation node distribution results of the fruit growth stage.

2. The apple tree growth status intelligent identification system according to claim 1, characterized in that: The internode main axis direction vector sequence includes the internode starting coordinate value, the internode ending coordinate value and the internode direction unit vector; the branch direction fluctuation characteristic index includes the direction change amplitude sequence, the angle reversal frequency value and the angle change amplitude mean; the branch structure growth stability judgment result is specifically the branch stability state label, the direction trend matching label and the sample interval attribution label; the fruit growth stage mutation node distribution result includes the trend turning time point, the fruit reflectance change point and the fruit morphology symmetry variation point.

3. The apple tree growth status intelligent identification system according to claim 2, characterized in that: The branch structure recognition module includes: The image boundary extraction submodule obtains the branch image data of the target individual apple tree during the observation period, collects the pixel area information of the branch part in the image, and extracts the boundary pixel values ​​in the grayscale gradient direction within the corresponding pixel distribution range. Based on the connectivity between the boundary pixels, the contour boundary box of each internode area is constructed, and the coordinate set of the contour boundary in the image coordinate system is established to generate the internode area boundary coordinate set; The internode axis construction submodule extracts the farthest point pair in the boundary set based on the internode region boundary coordinate set and calculates the Euclidean distance of the farthest point pair, selects the connecting line corresponding to the maximum distance as the longest internode boundary line, and calibrates the starting point and the ending point of the line in the image coordinate system, records the connecting line as the internode axis, and generates an internode axis coordinate sequence; The direction vector generation submodule calculates the difference between the starting point and the ending point coordinate values ​​of each internode axis according to the internode axis coordinate sequence and constructs a two-dimensional vector, calculates the unit direction of the vector and standardizes it into a unit vector, arranges all unit direction vectors in sequence according to the internode order, establishes the internode direction change sequence, and generates the internode main axis direction vector sequence.

4. The apple tree growth status intelligent identification system according to claim 3, characterized in that: The direction sequence calculation module includes: The angle calculation submodule calls the internode main axis direction vector sequence, extracts the vector component values ​​of the direction vectors of two adjacent internodes and calculates the cosine value of the angle, reversely deduces the angle angle based on the cosine value and records the positive and negative change direction of the angle, and sequentially arranges the angle change values ​​of each internode to generate an internode angle change value sequence; The angle change extraction submodule performs a first-order difference operation on two adjacent angle change values ​​according to the internode angle change value sequence and records the difference direction, compares the positive and negative directions between the differences item by item and marks the direction turning position, records the direction turning point as the segment node position in the sequence, and generates a direction turning position sequence; The fluctuation feature extraction submodule calls the direction turning position sequence, counts the number of consecutive direction reversals in the sequence as the angle reversal frequency value, and calculates the arithmetic mean of each absolute value in the original internode angle change value sequence as the angle change amplitude, establishes a joint structural fluctuation feature including the reversal frequency and the change mean, and generates a branch direction fluctuation feature index.

5. The apple tree growth status intelligent identification system according to claim 4, characterized in that: The growth stability judgment module includes: The sample index extraction submodule obtains healthy apple tree samples at the same growth stage based on the branch direction fluctuation characteristic index, extracts the direction reversal frequency value and the mean direction change amplitude of the branches in the samples, organizes the two types of index values ​​into independent data sequences according to the sample numbers, extracts the boundaries based on the minimum and maximum values ​​of each type of data in the sample sequence, and generates the branch structure characteristic interval range; The threshold interval construction submodule calls the branch structure feature interval range, sets the upper and lower limits of the reversal frequency interval and the amplitude mean interval according to the two types of interval boundary values, obtains the direction reversal frequency and the change mean amplitude of the target individual, calculates the difference between the target value and the sample boundary value, and calculates the branch structure stability deviation degree value to obtain the branch structure stability deviation information; The stability state judgment submodule compares the stability deviation value of each branch with the set stability judgment threshold based on the branch structure stability deviation information, and obtains the dynamic time regularization distance between the target branch direction change angle sequence and the healthy sample sequence. It judges whether any of the two judgment criteria exceeds the limit value. If any condition is met, it is marked as an unstable state to obtain the branch structure growth stability judgment result.

6. The apple tree growth status intelligent identification system according to claim 5, characterized in that: The system further comprises: The vector trajectory offset module extracts the average diameter growth rate, branch extension angle increase, and leaf area index change of each mutation node in the period before and after the mutation node based on the distribution results of the mutation nodes in the fruit growth stage, continuously constructs time series vector trajectories of multiple periods, records the direction offset segments, and generates the identification results of the growth direction offset areas of individual apple trees; The identification results of the individual growth direction deviation areas of the apple trees specifically refer to the deviation duration period segments, angle change trend records and direction consistency abnormal segments.

7. The apple tree growth status intelligent identification system according to claim 6, characterized in that: The vector trajectory offset module includes: The three-parameter extraction submodule extracts the average fruit diameter growth rate, branch extension angle increase, and leaf area index change in the consecutive periods before and after each mutation node based on the distribution results of the mutation nodes in the fruit growth stage, and normalizes the three parameters in the same time period, generates a standardized vector for each set of data, and obtains a three-parameter standard vector sequence; The vector trajectory generation submodule calls the three-parameter standard vector sequence, calculates the angle and cosine similarity between the standard vector of each cycle and the set target growth direction reference vector, extracts the angle change rate and similarity decrease trend in consecutive cycles, and calculates the direction deviation trend value; The offset segment identification submodule compares the change value in the continuous cycle with the direction offset threshold according to the direction offset trend value, marks the period segment with continuously increasing trend value, and identifies whether the offset angle continues to increase in adjacent multiple cycles. The continuous segments that meet the conditions are marked to establish the individual growth direction offset area identification results of apple trees.

Images