Intelligent monitoring method and system for tree flower bud maturity

Abstract

The application relates to the technical field of plant maturity intelligent monitoring, and discloses a tree flower bud maturity intelligent monitoring method and system, which comprises the following steps: according to a pre-acquired tree flower image, analyzing flower bud distribution density, collecting multi-angle images in a region with a flower bud distribution density greater than a preset density threshold value to obtain a first image set; extracting a flower bud shape parameter, identifying a flower bud contour, and calculating a flower bud volume; extracting a morphological feature of an inner region of the flower bud contour, combining the flower bud volume, calculating a probability value of the flower bud belonging to a preset maturity grade through feature fusion analysis, and obtaining the maturity of the flower bud; combining the maturity of the flower bud and historical growth data, simulating a growth process of the flower bud, predicting a flower bud maturity time, and intelligently monitoring the maturity of the tree flower bud; the application can improve the prediction accuracy of the flower bud maturity time, realizes accurate harvesting and yield increase of the tree flower.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring technology for plant maturity, and more specifically to a method and system for intelligent monitoring of the maturity of cauliflower buds. Background Technology

[0002] Currently, as an important specialty vegetable, the maturity of tree cauliflower buds directly affects its commercial value, harvesting timing, and subsequent processing quality. Traditionally, judging the maturity of tree cauliflower buds mainly relies on experienced farmers or quality inspectors to conduct manual observation and touch in the field, which has problems such as strong subjectivity, inconsistent standards, low efficiency, and inability to conduct large-scale real-time monitoring.

[0003] Existing technologies suffer from the following problems: they rely on image analysis from a single angle, and the dense growth, irregular shape, and severe mutual occlusion of cauliflower buds make it difficult to accurately analyze their true morphology from a single angle, leading to large errors in maturity assessment; they use only a few obvious features such as color and size for assessment, lacking analysis of subtle changes in the surface texture of the buds, and are insensitive to early maturity assessments; they employ static analysis, lacking dynamic integration with historical growth data, failing to simulate bud growth trends, and unable to provide predictive decision support for agricultural harvesting plans; to address at least one of the above problems, this application proposes an intelligent monitoring method and system for the maturity of cauliflower buds. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide an intelligent monitoring method and system for the maturity of cauliflower buds, which can effectively solve the problems in the background technology. The specific technical solution of this application is as follows:

[0005] A method for intelligent monitoring of the maturity of cauliflower buds includes:

[0006] Based on the pre-acquired images of tree cauliflower, the distribution density of flower buds is analyzed, and multi-angle image acquisition is performed on areas where the distribution density of flower buds is greater than a preset density threshold to obtain the first image set.

[0007] The images in the first image set are aligned temporally and spatially, and the morphological parameters of the flower buds are extracted, the flower bud outline is identified, and the flower bud volume is calculated.

[0008] The morphological features of the region within the flower bud outline are extracted and combined with the flower bud volume. The probability value of the flower bud belonging to the preset maturity level is calculated through feature fusion analysis to obtain the maturity of the flower bud.

[0009] By combining the maturity of flower buds with historical growth data, the growth process of flower buds is simulated to predict the maturity time of flower buds, so as to intelligently monitor the maturity of tree cauliflower flower buds.

[0010] Specifically, based on the pre-acquired images of the cauliflower, the distribution density of the flower buds is analyzed, and multi-angle image acquisition is performed on areas where the flower bud distribution density is greater than a preset density threshold to obtain a first image set, including:

[0011] Based on the pre-acquired images of tree cauliflower, the location of tree cauliflower is identified to obtain a bud distribution map;

[0012] Analyze the connected components in the flower bud distribution map, treat each connected component as a flower bud region, and construct a set of flower bud regions.

[0013] Based on the flower bud region set, multi-angle image acquisition is performed on regions where the flower bud distribution density is greater than a preset density threshold to obtain the first image set.

[0014] Specifically, according to the flower bud region set, multi-angle image acquisition is performed on regions where the flower bud distribution density is greater than a preset density threshold to obtain a first image set, including:

[0015] Based on the flower bud region set, calculate the flower bud distribution density of each region, and filter out regions whose flower bud distribution density is greater than a preset density threshold to obtain the first region set;

[0016] Analyze the boundary range of each region in the first region set and select the corresponding boundary midpoints;

[0017] Based on the positional relationship between the midpoint of the boundary and the preset physical coordinate system, the corresponding acquisition angle is calculated, and multi-angle image acquisition is performed according to the acquisition angle to obtain the first image set.

[0018] Specifically, the steps of performing temporal and spatial alignment on the images in the first image set, extracting the morphological parameters of the flower buds, identifying the flower bud outline, and calculating the flower bud volume include:

[0019] The images in the first image set are aligned temporally and spatially, and the morphological parameters of the flower buds are extracted to identify the flower bud outlines and construct flower bud distribution data.

[0020] Based on the bud distribution data, the curvature of the bud outline at different positions is calculated to obtain the outline features, and the individual bud data is separated and the corresponding bud volume is calculated.

[0021] Specifically, the step of performing temporal and spatial alignment on the images in the first image set, extracting the morphological parameters of the flower buds, identifying the flower bud outlines, and constructing flower bud distribution data includes:

[0022] The images in the first image set are time-aligned and spatially aligned to obtain the corrected image set;

[0023] From the set of corrected images, extract the key points and morphological parameters of each image to construct a set of key point parameters;

[0024] Calculate the matching degree of key point parameter sets between images, and select the two images with the highest matching degree as the image benchmark;

[0025] Identify the flower bud outlines in the image baseline, and sequentially add the first image set to the corresponding distribution locations to construct flower bud distribution data.

[0026] Specifically, the step of calculating the curvature of the flower bud contour at different positions based on the flower bud distribution data to obtain contour features, separating individual flower bud data, and calculating the corresponding flower bud volume includes:

[0027] Based on the flower bud distribution data, the curvature of the flower bud outline at different positions is calculated to obtain the outline features. Clustering is then performed according to the outline features to separate individual flower bud data.

[0028] The data of each individual flower bud is projected onto a vertical plane to identify the corresponding flower bud edge;

[0029] Based on the flower bud edge, calculate the edge length and edge curvature, and calculate the corresponding flower bud volume by integration.

[0030] Specifically, the extraction of morphological features within the flower bud outline, combined with the flower bud volume, and the calculation of the probability value of the flower bud belonging to a preset maturity level through feature fusion analysis, yields the maturity of the flower bud, including:

[0031] The morphological features of the region within the flower bud outline are extracted and combined with the flower bud volume. Through feature fusion analysis, a flower bud feature vector is constructed.

[0032] Based on the flower bud feature vector, the probability value of the flower bud belonging to the preset maturity level is calculated, and the maturity of the flower bud is obtained.

[0033] Specifically, the extraction of morphological features within the flower bud contour area, combined with the flower bud volume, and the construction of a flower bud feature vector through feature fusion analysis include:

[0034] Extract the morphological features of the region within the flower bud outline. The morphological features include the curvature of each vertex in the flower bud outline, the area ratio and distribution dispersion of the positive Gaussian curvature region and the negative Gaussian curvature region on the flower bud surface, and the axial asymmetry of the two sides of the flower bud outline.

[0035] The pixel value distribution skewness and kurtosis of the flower bud outline are analyzed, and the spectral reflectance is simulated for the accumulation process of target pigments during the maturation process to calculate the corresponding spectral index.

[0036] Extract texture features from the region within the flower bud outline;

[0037] The morphological features, spectral indices, texture features, and bud volume are fused to construct a bud feature vector.

[0038] Specifically, the step of calculating the probability value of a flower bud belonging to a preset maturity level based on the flower bud feature vector to obtain the maturity of the flower bud includes:

[0039] Based on the flower bud feature vector, the maturity of the flower bud is analyzed through a preset neural network model, and the probability value of the flower bud belonging to the preset maturity level is calculated.

[0040] The maturity of the flower buds is identified based on the probability values.

[0041] A smart monitoring system for the maturity of tree cauliflower buds, used to implement the aforementioned smart monitoring method for the maturity of tree cauliflower buds, includes:

[0042] The image acquisition module analyzes the bud distribution density based on the pre-acquired tree cauliflower images, and performs multi-angle image acquisition on areas where the bud distribution density is greater than a preset density threshold to obtain a first image set.

[0043] The flower bud analysis module performs temporal and spatial alignment on the images in the first image set, extracts the morphological parameters of the flower buds, identifies the flower bud outline, and calculates the flower bud volume.

[0044] The fusion calculation module extracts the morphological features of the region within the flower bud outline, combines them with the flower bud volume, and calculates the probability value of the flower bud belonging to the preset maturity level through feature fusion analysis to obtain the maturity of the flower bud.

[0045] The maturity monitoring module combines the maturity of flower buds with historical growth data to simulate the growth process of flower buds and predict the maturity time of flower buds, so as to intelligently monitor the maturity of tree cauliflower flower buds.

[0046] The beneficial effects of this application are as follows: By analyzing the distribution density of flower buds, dense areas of flower buds can be identified and multi-angle images can be acquired, which can improve the efficiency of acquisition and processing while ensuring data quality; by performing temporal and spatial alignment on multi-angle images, analyzing the curvature features at different positions of the contour, and identifying individual flower buds to calculate their corresponding volumes, flower buds can be accurately separated and their volumes calculated in densely distributed scenes, providing accurate morphological parameters for maturity judgment; by combining morphological features, spectral indices, and texture features to construct flower bud feature vectors, the growth process of flower buds can be simulated, and the maturity of tree cauliflower flower buds can be intelligently monitored, which can improve the accuracy of flower bud maturity prediction and achieve precise harvesting and increased yield and efficiency of tree cauliflower. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the intelligent monitoring method for the maturity of tree cauliflower buds according to Embodiment 1 of this application.

[0048] Figure 2 This is a schematic diagram of the flower bud distribution in Embodiment 1 of this application;

[0049] Figure 3 This is a flowchart illustrating the process of constructing flower bud distribution data in Embodiment 1 of this application;

[0050] Figure 4 This is a schematic diagram of the structure of an intelligent monitoring system for the maturity of tree cauliflower buds according to Embodiment 1 of this application. Detailed Implementation

[0051] The present application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0052] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0053] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0054] Example 1:

[0055] refer to Figure 1 The image shows a specific implementation of a method for intelligent monitoring of the maturity of cauliflower buds according to this application, comprising:

[0056] S101. Based on the pre-acquired images of tree cauliflower, analyze the distribution density of flower buds, and perform multi-angle image acquisition on areas where the distribution density of flower buds is greater than a preset density threshold to obtain a first image set.

[0057] S102. Perform time and space alignment on the images in the first image set, extract the morphological parameters of the flower buds, identify the flower bud outline, and calculate the flower bud volume.

[0058] S103. Extract the morphological features of the region within the flower bud outline, combine them with the flower bud volume, and calculate the probability value of the flower bud belonging to the preset maturity level through feature fusion analysis to obtain the maturity of the flower bud.

[0059] S104. By combining the maturity of flower buds and historical growth data, the growth process of flower buds is simulated and the maturity time of flower buds is predicted, so as to intelligently monitor the maturity of tree cauliflower flower buds.

[0060] In this embodiment, a fixed pole station equipped with a visible light camera is used to acquire a two-dimensional image of the top of the cauliflower plants covering the entire monitoring area from a vertical top-down perspective, thus obtaining a pre-acquired cauliflower image. The image is then processed using a semantic segmentation model based on deep learning. The semantic segmentation model includes, but is not limited to, the U-Net network model. The training dataset of the model consists of manually annotated cauliflower images containing buds and background. The model is trained using a cross-entropy loss function with a learning rate of 0.001. Through numerous iterations, the model is able to accurately segment the bud pixel group from the complex background. The model output is a bud distribution map, in which each pixel is classified as either a bud or a non-bud.

[0061] Preferably, the binarized bud distribution map is analyzed, and sets of bud pixels with connected pixel values ​​are selected. Each independent connected region is considered a bud region. The bud distribution density of a region is calculated by counting the number of bud pixels within each region and dividing by the area of ​​the bounding rectangle of that region. A density threshold is set based on the typical planting density and canopy structure of different cauliflower varieties. For example, the density threshold is set when the area of ​​bud pixels accounts for more than 60% of the region area. Regions with bud distribution densities higher than the density threshold are selected to obtain a first region set. For each region in the first region set, the midpoint of the boundary is calculated, and a multi-angle shooting path is planned based on the three-dimensional coordinates of the midpoint in a preset global physical coordinate system. For example, the camera-mounted gimbal is controlled to shoot around the center of the boundary at multiple preset angles, such as horizontal azimuth angles of 0 degrees, 90 degrees, 180 degrees, 270 degrees, and a certain pitch angle, to obtain a first image set including multi-view information for each region.

[0062] It should be noted that by analyzing flower bud density, areas requiring detailed three-dimensional observation can be located, reducing invalid shots and lowering the computational load and time cost of image transmission, storage, and processing. Dense areas of flower buds are prone to errors in maturity assessment. By acquiring images from multiple angles, it can be ensured that even in these complex scenarios, sufficient vantage point information can be obtained to reconstruct the three-dimensional morphology of flower buds. This provides an accurate data foundation for subsequent high-precision analysis of flower bud maturity and avoids the information loss problem caused by the single perspective in traditional two-dimensional methods.

[0063] Furthermore, the acquired multi-angle images are time-aligned, and pixel position deviations are corrected using linear interpolation to ensure that all images are at the same time. Spatial alignment is performed by feature point matching. Key points and corresponding descriptors are extracted from all images in the first image set, the Euclidean distance between descriptors is calculated, corresponding key points in different images are matched, and the camera parameters of each image and the three-dimensional spatial coordinates of all matched key points are analyzed to construct flower bud distribution data.

[0064] Specifically, clustering is performed on the flower bud distribution data. To separate buds that are stuck together, the curvature of each point on each contour is calculated to obtain contour features. For example, the neck region where two buds are stuck together will show a high Gaussian curvature change. By clustering, points with continuous smooth curvature changes are grouped into the same bud, thus achieving the separation of individual flower bud data. The separated three-dimensional point cloud of an individual flower bud is projected onto a plane perpendicular to the principal axis of the flower bud. The discrete edge points are connected to form a closed two-dimensional polygonal flower bud edge. The volume of the flower bud is obtained by integrating the polygonal flower bud edge.

[0065] It should be noted that spatiotemporal alignment can effectively eliminate data misalignment caused by the movement of the acquisition equipment and interference from the external environment, thereby improving the accuracy of the data. By analyzing curvature features to cluster and separate adhered flower buds, flower bud segmentation can be achieved in densely planted scenarios, avoiding volume calculation errors caused by misjudging multiple flower buds as a whole, and improving the robustness and accuracy of monitoring in complex scenarios.

[0066] Specifically, morphological features, spectral index features, and texture features are extracted from individual flower bud data and corresponding texture images. Morphological features include, but are not limited to, vertex curvature, the area ratio of positive to negative Gaussian curvature regions on the statistical surface and their distribution dispersion on the flower bud surface, and the calculation of the axial asymmetry of the flower bud's two-sided contours relative to the principal axis. Spectral index features include, but are not limited to, analyzing the distribution of pixel values ​​in each channel of the RGB image within the flower bud contour region, calculating skewness and kurtosis, establishing an empirical mapping relationship between the content of these pigments and the nonlinear combination of RGB channel values ​​based on known degradation and accumulation patterns of chlorophyll and carotenoids during maturation, and calculating spectral indices reflecting intrinsic biochemical changes. Texture features include, but are not limited to, using the gray-level co-occurrence matrix to extract parameters such as contrast, correlation, homogeneity, and energy from the flower bud texture image. All extracted feature values ​​and flower bud volume are combined to form a flower bud feature vector.

[0067] Preferably, the flower bud feature vector is input into a pre-trained multilayer perceptron neural network model. The number of nodes in the input layer of this model is equal to the dimension of the feature vector, and the number of nodes in the output layer is equal to the preset number of maturity levels. For example, the three nodes in the output layer correspond to immature, moderately mature, and overripe, respectively. The training dataset comes from historical monitoring data, consisting of flower bud samples manually judged by agricultural experts according to standards, with the aforementioned feature vectors already extracted. The backpropagation algorithm is used, with cross-entropy as the loss function, in conjunction with the Adam optimizer, and the initial learning rate is set to 0.0005 for training until the model's classification accuracy on the validation set tends to stabilize. The trained model calculates a probability distribution vector for each input flower bud feature vector, outputting the probability value of belonging to each preset maturity level. The level with the highest probability value is taken as the final maturity judgment result for that flower bud.

[0068] It should be noted that by integrating multi-dimensional features such as bud morphology, simulated spectrum, and texture, the constructed feature vector can comprehensively reflect the coordinated changes in the external geometric shape and internal biochemical state of the bud, avoiding the shortcomings of single features being easily affected by light and having a single discrimination dimension. For example, the positive and negative Gaussian curvature distribution can sensitively reflect the details of the swelling and relaxation of the bud surface structure, and the spectral index reflects the difficult-to-detect early pigment changes. By using a neural network model for probability classification, the output results are more accurate, improving the accuracy of the bud maturity judgment results.

[0069] Specifically, the system continuously records the maturity level and corresponding feature vector of each flower bud during the monitoring period. The monitoring period is set according to the system's monitoring accuracy requirements, for example, once a day. Once the maturity level at the current moment is obtained, historical growth data sequences over a past period are retrieved based on the flower bud ID or location information. A growth model is then used to fit this historical data. The growth model includes, but is not limited to, a modified logistic growth curve model. This model expresses the key indicators of the flower bud as a function of time, and the model parameters reflect the specific growth rate, maximum potential size, and other characteristics of the flower bud. Through fitting, the specific position and growth rate of the flower bud on the current growth curve are analyzed. Combined with the currently determined maturity level threshold, for example, setting a spectral index of 0.8 as optimal maturity, the fitted growth model is used for extrapolation calculations to predict the future time required for the key indicator to reach the optimal maturity threshold. This time point is the predicted flower bud maturity time.

[0070] It should be noted that predicting the bud maturity time provides an accurate basis for decision-making in arranging the optimal harvest time, which can optimize agricultural production planning and management processes, reduce economic losses caused by harvesting too early or too late, and improve the response efficiency and accuracy of the bud maturity monitoring process.

[0071] This application analyzes the distribution density of flower buds to identify densely populated areas and acquires images from multiple angles, improving acquisition and processing efficiency while ensuring data quality. It performs temporal and spatial alignment on the multi-angle images, analyzes the curvature features at different locations of the contour, and identifies individual flower buds to calculate their corresponding volumes. This allows for accurate flower bud separation and volume calculation even in densely distributed scenes, providing accurate morphological parameters for maturity assessment. Furthermore, by combining morphological features, spectral indices, and texture features to construct flower bud feature vectors, it simulates the flower bud growth process and intelligently monitors the maturity of tree cauliflower flower buds. This improves the accuracy of predicting flower bud maturity time, enabling precise harvesting and increased yield and efficiency for tree cauliflower.

[0072] Furthermore, based on the pre-acquired images of the cauliflower, the distribution density of flower buds is analyzed. Multi-angle image acquisition is performed on areas where the flower bud distribution density is greater than a preset density threshold, resulting in a first image set, including:

[0073] S201. Based on the pre-acquired images of tree cauliflower, identify the location of tree cauliflower to obtain a bud distribution map;

[0074] S202. Analyze the connected components in the flower bud distribution map, treat each connected component as a flower bud region, and construct a set of flower bud regions.

[0075] S203. Based on the flower bud region set, perform multi-angle image acquisition on the regions where the flower bud distribution density is greater than the preset density threshold to obtain the first image set.

[0076] In this embodiment, the pre-acquired images of tree cauliflower are captured by high-definition cameras deployed in the field at a vertical angle. The images are then identified using a pre-trained convolutional neural network (CNN) model, specifically the U-Net model. The model's input is the original RGB three-channel image, and its output is a binarized image of the same size as the input image. The output value of each pixel represents the probability that it belongs to the bud category. The training data includes images of tree cauliflower taken from a large number of historical images, from which agricultural experts manually and precisely delineate the outlines of each visible bud, labeling the area within the outline as a bud and the area outside the outline as the background. Using the training data, the model is trained using the Adam optimizer with the cross-entropy loss function as the optimization objective. The initial learning rate can be set to 0.001. Through multiple iterations, the model learns the mapping relationship from the original pixels to the category labels, resulting in the pre-trained U-Net model.

[0077] Preferably, the acquired images of cauliflower are input into a pre-trained U-Net model. The model calculates the probability of each pixel being a flower bud and outputs the probability of whether it is a flower bud. A probability threshold is set according to the system's monitoring accuracy requirements, for example, a probability threshold of 0.5. Figure 2As shown, pixels with a probability greater than the probability threshold are identified as flower bud pixels and assigned a value of 1, while the remaining pixels are assigned a value of 0, resulting in a flower bud distribution map. Model-based segmentation effectively overcomes interference from variations in natural lighting, leaf shading, and the similarity in color between flower buds and the background, improving the accuracy and completeness of flower bud identification in complex farmland environments. The generated flower bud distribution map provides accurate data support for subsequent spatial analysis, enhancing the accuracy and reliability of the monitoring system.

[0078] Furthermore, a dilation operation is performed on the bud distribution map using tiny circular structural elements. This bridges the small gaps between adjacent bud pixels caused by recognition errors or image noise, allowing pixels belonging to the same physical bud to be better connected. A subsequent erosion operation of the same size restores the original size of the region, improving region integrity without altering its area. The entire bud distribution map is scanned, and region growing is performed using the 8-connectivity criterion. A unique label is assigned to each interconnected group of bud pixels, marking all independent pixel sets. Each independent connected pixel set constitutes a bud region. A bud region corresponds to a single, independent bud or multiple bud groups that cannot be separated in a 2D image due to imaging angle or excessively dense growth. The coordinates of all pixels in each bud region are recorded to obtain the bud region set. Through connected component analysis, the number of buds in the image can be automatically counted, and the spatial occupancy of each bud can be defined, reflecting the actual bud distribution. This allows the system to perform feature calculations and logical judgments on a region-by-region basis, improving analysis efficiency and accuracy.

[0079] Specifically, based on the bud region set, multi-angle image acquisition is performed on areas where the bud distribution density exceeds a preset density threshold to obtain the first image set. Density filtering reduces the number of areas requiring multi-angle acquisition, decreasing the total data volume, shortening acquisition time, and alleviating the burden of subsequent image storage and processing. Providing multi-angle images ensures sufficient information for bud morphology analysis in complex scenarios, avoiding analysis failures or accuracy reduction due to insufficient information, thus improving the system's practicality and accuracy while maintaining monitoring precision.

[0080] Furthermore, based on the flower bud region set, multi-angle image acquisition is performed on regions where the flower bud distribution density is greater than a preset density threshold to obtain a first image set, including:

[0081] S301. According to the flower bud region set, calculate the flower bud distribution density of each region, and filter out the regions whose flower bud distribution density is greater than the preset density threshold to obtain the first region set.

[0082] S302. Analyze the boundary range of each region in the first region set and select the corresponding boundary midpoints;

[0083] S303. Based on the positional relationship between the midpoint of the boundary and the preset physical coordinate system, calculate the corresponding acquisition angle, and perform multi-angle image acquisition according to the acquisition angle to obtain the first image set.

[0084] In this embodiment, the bud region set includes the coordinate set of all bud pixels in each region. For each region in the bud region set, the minimum bounding rectangle of that region is calculated. The minimum bounding rectangle is the smallest regular rectangle that can completely enclose all pixels in that region. The bud distribution density of that region is calculated by dividing the total number of bud pixels contained in that region by the area of ​​the minimum bounding rectangle. This bud distribution density reflects the density of the area filled with buds within the rectangle. For example, if a region has 1500 pixels and its bounding rectangle has an area of ​​2500 pixels, then the density value is 0.6.

[0085] Preferably, a density threshold is set based on the typical planting density and canopy structure of different cauliflower varieties. For example, the density threshold is set to 0.55. All regions are traversed, and the bud distribution density is compared with the density threshold one by one. Regions with bud distribution densities greater than the density threshold are selected to obtain the first region set. By using the density threshold for selection, the region with the densest bud growth and the greatest likelihood of shading can be focused on.

[0086] Specifically, for each region in the first set of regions, calculate the minimum and maximum coordinates of that region in the image pixel coordinate system along the horizontal direction and the minimum and maximum coordinates along the vertical direction to determine the boundary range. Calculate the midpoint of the boundary range; the horizontal coordinate of the midpoint is equal to the average of the minimum and maximum coordinates of that region in the image pixel coordinate system along the horizontal direction, and the vertical coordinate is equal to the average of the minimum and maximum coordinates of that region in the vertical direction. Calculate the corresponding midpoint coordinates for each region in the first set of regions.

[0087] It should be noted that by calculating the midpoint of the boundary, the corresponding spatial coordinate point is determined for each high-density area as the focus of multi-angle surround shooting. This ensures that the entire target area remains in the center of the image or within the main field of view when shooting from different perspectives. This maximizes the capture of complete or main information of the area in each shot, providing accurate input data for the next step of precise calculation of the acquisition angle. It avoids shooting blind spots or incomplete information caused by improper selection of reference points, and improves the data quality and consistency of multi-view image sets.

[0088] Specifically, the coordinates of the boundary midpoints are transformed into a preset physical coordinate system using camera calibration parameters and known shooting height information. This coordinate system can be established by setting the origin of the camera's optical center position during the initial pre-collection. The three-dimensional coordinates of each boundary midpoint in this physical coordinate system are calculated, and a set of acquisition angles is planned, defining the ideal line-of-sight directions for observing flower buds from multiple directions. For example, four uniform azimuth angles of 0°, 90°, 180°, and 270° and a fixed pitch angle of -20° can be planned. The image acquisition device is controlled to adjust the angles sequentially according to the set acquisition angles and perform high-definition shooting. All images taken from different planned angles are compiled to obtain the first image set. By setting multi-angle shooting, sufficient multi-view images for high-quality 3D reconstruction can be acquired for each high-density area. Multi-angle image acquisition of a limited selected area can reduce the total time and data volume, improving the efficiency and accuracy of the flower bud maturity monitoring process.

[0089] Furthermore, the images in the first image set are time-aligned and spatially aligned, and the morphological parameters of the flower buds are extracted, the flower bud contours are identified, and the flower bud volume is calculated, including:

[0090] S401. Perform temporal and spatial alignment on the images in the first image set, extract the morphological parameters of the flower buds, identify the flower bud outlines, and construct flower bud distribution data.

[0091] S402. Calculate the curvature of the flower bud outline at different positions based on the flower bud distribution data to obtain the outline features, separate the individual flower bud data and calculate the corresponding flower bud volume.

[0092] In this embodiment, images in the first image set are aligned temporally and spatially, and morphological parameters of flower buds are extracted to identify flower bud contours and construct flower bud distribution data. Temporal alignment can effectively compensate for target micro-movements caused by natural wind or equipment movement, avoiding the damage of motion artifacts to subsequent feature matching and 3D reconstruction accuracy, and ensuring the temporal consistency of the data. Spatial alignment integrates multiple 2D observations into an accurate 3D model, which can overcome the problem of missing information from a single perspective. By fusing data from multiple perspectives to identify flower bud contours, misjudgments caused by occlusion or lighting are reduced, and the accuracy of flower bud analysis is improved.

[0093] Specifically, based on the bud distribution data, the curvature of the bud outline at different locations is calculated to obtain outline features. Individual bud data are then separated, and their corresponding volumes are calculated. Clustering separation through contour feature analysis identifies the natural boundaries between buds, improving the accuracy of bud segmentation and ensuring that subsequent volume calculations are performed on independent individuals, avoiding significant errors caused by incorrectly merging multiple bud volumes. Calculating the precise volume of each bud reflects its growth stage, providing accurate input features for subsequent maturity assessment.

[0094] like Figure 3 As shown, the images in the first image set are aligned temporally and spatially, and the morphological parameters of the flower buds are extracted to identify the flower bud contours and construct flower bud distribution data, including:

[0095] S501. Perform temporal and spatial alignment on the images in the first image set to obtain a corrected image set;

[0096] S502. Extract the key points and morphological parameters of each image from the set of corrected images, and construct a set of key point parameters;

[0097] S503. Calculate the matching degree of key point parameter sets between images, and select the two images with the highest matching degree as the image benchmark;

[0098] S504. Identify the flower bud outline in the image reference, and add the first image set to the corresponding distribution positions in sequence to construct flower bud distribution data.

[0099] In this embodiment, since the multi-angle images are taken sequentially, environmental factors such as natural wind can cause slight swaying of flower buds or leaves. The timestamp of each image is read from the camera's capture time, and the earliest captured image in the set is used as the time reference frame. For each other image, the pixel-level displacement field caused by the slight movement of the target between the image and the time reference frame is estimated using optical flow. Based on the relative time difference of the image and the estimated displacement field, the image is resampled using inverse bilinear interpolation to compensate for the pixel position offset caused by temporal asynchrony, resulting in a time-aligned image.

[0100] Specifically, images taken from different physical camera positions are spatially aligned to a single viewpoint plane to eliminate perspective distortion and scale differences caused by varying shooting angles. Using approximate camera extrinsic parameters obtained from the sensor, all images are mapped to a common reference imaging plane through perspective projection transformation. A plane approximately perpendicular to most of the shooting line of sight is chosen as the reference imaging plane. Through this reprojection transformation, the shape and scale differences of the same scene captured from different angles are largely eliminated, resulting in a set of corrected images where all image content is geometrically aligned.

[0101] Furthermore, for each image in the corrected image set, pixel locations with significant local structure are extracted using a scale-invariant feature transform (SMT) algorithm as keypoints. The SMT algorithm outputs a series of keypoint coordinates and corresponding feature vectors. These feature vectors describe the gradient direction distribution of the image patch surrounding the point and are invariant to illumination, rotation, and scale changes. For each bud region, morphological parameters are calculated, including but not limited to the region's pixel area, perimeter, aspect ratio of the minimum bounding rectangle, and density. These morphological parameters describe the two-dimensional shape and size characteristics of the bud region from different perspectives. The keypoint information and morphological parameters of each image are integrated to obtain a keypoint parameter set.

[0102] It should be noted that by extracting key points and feature vectors, positional information is provided for the image matching process. The invariant properties ensure that the corresponding relationship can be stably found even under different lighting conditions and slight viewpoint residuals. The extracted morphological parameters serve as supplementary features, providing accurate data support for the image matching process.

[0103] Specifically, the corrected image set is traversed, and the matching degree of each pair of images is calculated. Based on the keypoint feature vectors in the keypoint parameter sets of the two images, the nearest neighbor search method is used to find the closest and second closest keypoints in the other image for each keypoint in the first image. The ratio of the nearest neighbor distance to the second nearest neighbor distance is calculated, and matching results with a ratio less than a preset threshold are selected to obtain preliminary matching results. The geometric transformation relationship between the two images is analyzed from the preliminary matching results using a random sampling consensus algorithm. Outliers that do not conform to geometric constraints are removed, and the number of matching point pairs retained is used as the matching degree between the two images. After calculating the matching degree between all image pairs, the pair of images with the highest matching degree is selected as the image benchmark.

[0104] It should be noted that by performing matching degree analysis based on the number of feature matches and geometric consistency, the two images with the richest information and the most reliable correlation in the scene can be automatically identified. The image pair with the highest matching degree is selected as the benchmark, reflecting that there are the most and most accurate corresponding points between the images. This provides an accurate data foundation for the recovery of 3D information, improves the accuracy of the analysis process, and enhances the system's adaptability to complex scenes.

[0105] Furthermore, on the selected image benchmark, accurate flower bud contour recognition is performed. A semantic segmentation model based on the U-Net architecture is used, which needs to be pre-trained on a large number of tree cauliflower images labeled with flower bud contours. The benchmark image is directly processed using the cross-entropy loss function and the Adam optimizer to output a flower bud binary mask. The closed contour polygons of all flower buds are extracted from the flower bud binary mask by the contour tracking algorithm.

[0106] Preferably, based on the reference image and its precise contour information, a multi-view stereo vision algorithm is used to recover the initial sparse 3D point cloud of flower buds within the common viewing area of ​​the reference image. For each image in the calibration image set other than the reference image, the correspondence between its feature points and the existing 3D point cloud is analyzed by matching the projections of its feature points with those of the existing 3D points, and the camera pose of the image is solved. Using the viewpoint information of the newly added image and the existing 3D model, new 3D points are recovered through spatial forward intersection, resulting in a denser point cloud. The texture information of the new image is used to optimize the color attributes of the existing 3D points. During the fusion process, the contour information identified in the reference image serves as a spatial constraint to guide the semantic segmentation of the 3D point cloud, ensuring that the 3D points can be correctly classified into the corresponding flower bud instances. By iteratively adding and fusing the information of all images, a dense 3D point cloud model with texture and contour projection information is constructed as flower bud distribution data.

[0107] It should be noted that by starting with high-precision baseline image contours, the semantic accuracy and initial structural reliability of the 3D reconstruction are ensured; an incremental fusion strategy is adopted to efficiently and robustly integrate information from all multi-angle images, restore the 3D morphology of the flower bud, and obtain flower bud distribution data with geometric accuracy in 3D space, which can be used for volume measurement; it also associates multi-view 2D contours and textures, which can be used for morphological feature analysis; it can better handle occlusion and complex structures, and improve the accuracy of the flower bud analysis process.

[0108] Furthermore, based on the bud distribution data, the curvature at different positions of the bud outline is calculated to obtain the outline features. Individual bud data are then separated, and the corresponding bud volume is calculated, including:

[0109] S601. Calculate the curvature of the flower bud outline at different positions based on the flower bud distribution data to obtain the outline features, and then perform clustering according to the outline features to separate individual flower bud data.

[0110] S602. Project the data of each individual flower bud onto the vertical plane and identify the corresponding flower bud edge;

[0111] S603. Calculate the edge length and edge curvature according to the edge of the flower bud, and calculate the corresponding flower bud volume by integration.

[0112] In this embodiment, the flower bud distribution data includes a 3D point cloud of all flower bud points. For each point, a local surface is fitted by selecting a set of neighboring points. For example, the surface fitted by the 10 most recent points is selected. By analyzing the fitted surface, the principal curvature at that point is calculated, and the average curvature is used as the curvature value of that point. By traversing all points, a set of curvature values ​​corresponding one-to-one with the point cloud is obtained, reflecting the degree of drastic change in the shape of the flower bud surface. Clustering is performed by combining the 3D Euclidean geometric distance between points and the similarity of curvature values. The weighted sum of the geometric Euclidean distance and the curvature difference is calculated to obtain a comprehensive distance metric. Using the DBSCAN algorithm, the neighborhood search radius is set to 3 times the average point distance of the point cloud, and the minimum number of points to form a core point is set to 10. Points that are spatially close and have continuously and gently changing curvature are clustered into one class. Multiple independent subsets of the point cloud are obtained, and each subset represents a single flower bud data.

[0113] It should be noted that combining curvature with bud segmentation ensures that the segmentation boundary accurately falls at the true geometric intersection between buds, improving the accuracy of separation. This ensures that subsequent quantitative analyses, such as volume calculations, are performed on individual buds, avoiding volume data errors caused by erroneously merging multiple buds in the calculation, and thus improving the accuracy of the analysis process.

[0114] Specifically, for each individual flower bud data point, principal component analysis is used to calculate the three principal directions of the point cloud. The first principal direction is the direction in which the point cloud distribution is most dispersed, serving as the principal axis direction of the flower bud. A plane perpendicular to the first principal direction vector is defined as the perpendicular plane. All points in the 3D point cloud of the flower bud are projected onto the 2D plane along the direction parallel to the principal axis, i.e., perpendicular to the plane. After projection, a scattered point set is obtained on the plane. The flower bud edge is identified from this scattered point set using a concave hull algorithm. The concave hull algorithm defines a probe radius and sets a circle with a radius equal to the probe radius to roll outside the point set. The boundary of the blank area that the circle cannot enter constitutes the concave hull edge. The probe radius can be set to 2 to 5 times the average point distance of the point cloud. The algorithm outputs a polygon composed of a series of ordered vertices, which serves as the flower bud edge.

[0115] It should be noted that by adaptively determining the optimal projection plane for each flower bud through principal component analysis, the obtained two-dimensional contour is the most representative projection of the flower bud, thus improving the accuracy of the calculated volume. The concave hull algorithm is used to identify edges, which can better capture the non-convex shape features such as depressions and grooves in the flower bud, resulting in a more accurate contour representation and avoiding systematic errors in volume calculation caused by excessive contour simplification.

[0116] Preferably, based on the edge polygon of a single flower bud and its corresponding three-dimensional principal axis direction, the depth range of the flower bud in the projection direction in three-dimensional space is uniformly divided into N extremely thin slice layers along the principal axis direction, for example, the slice thickness is 0.1 mm. For the p-th slice layer, there is a cross-section on the projection plane that is similar to the original contour but has a different size.

[0117] Preferably, the edge curvature at the vertices of the original bud edge polygon and the edge length of each segment are analyzed, and the scaling ratio of the slice layer relative to the bud base and top is calculated. The bud's main axis is considered a growth axis, and the distance variation from each point on the bud edge polygon to the main axis determines the cross-sectional radius at different heights. Based on the distance between the vertices of the bud edge polygon and the main axis, at each slice height, the new positions of the vertices of the cross-sectional profile at that height are calculated using linear interpolation, constructing the cross-sectional polygon of that layer. The area of ​​the cross-sectional polygon of that layer is calculated, and the cross-sectional area of ​​each slice layer is multiplied by the slice thickness to obtain the approximate volume of the slice. The volumes of all slices are summed to obtain the estimated total volume of the bud, and the corresponding bud volume is calculated.

[0118] It should be noted that by using the slice integration method, combined with the three-dimensional geometric morphology information of the flower bud, the calculation accuracy can be improved, and the minute volume increment changes during the growth of the flower bud can be accurately captured. By combining edge length and curvature for cross-sectional interpolation, the actual shape changes of the flower bud along the main axis can be simulated, and the volume calculation process is more in line with the actual biological morphology, thus improving the accuracy of the calculated flower bud volume.

[0119] Furthermore, morphological features within the flower bud outline are extracted and combined with the flower bud volume. Feature fusion analysis is then used to calculate the probability value of the flower bud belonging to a preset maturity level, thus obtaining the flower bud's maturity, including:

[0120] S701. Extract the morphological features of the region within the flower bud outline, combine them with the flower bud volume, and construct the flower bud feature vector through feature fusion analysis.

[0121] S702. Based on the flower bud feature vector, calculate the probability value of the flower bud belonging to the preset maturity level, and obtain the maturity of the flower bud.

[0122] In this embodiment, morphological features within the flower bud outline are extracted and combined with flower bud volume. Through feature fusion analysis, a flower bud feature vector is constructed. By combining three-dimensional geometric shape, surface texture, simulated spectral index, and absolute volume, the resulting flower bud feature vector can simultaneously capture multiple key changes during flower bud maturation, such as the expansion and relaxation of external shape, changes in surface microstructure, biochemical transformation of internal pigments, and overall material accumulation. This effectively overcomes the problems of single-feature discrimination having a limited dimension and being susceptible to environmental interference, thereby improving the accuracy and computational efficiency of maturity analysis results.

[0123] Specifically, based on the bud feature vector, the probability value of the bud belonging to a preset maturity level is calculated, thus obtaining the bud's maturity. By analyzing the correspondence between the bud feature vector and maturity, the complex nonlinear relationship between features and maturity can be handled, resulting in high accuracy. When it is necessary to add new maturity levels or introduce new features, the model can be incrementally trained or retrained by supplementing labeled data to adapt, thereby improving the system's robustness and environmental adaptability.

[0124] Furthermore, morphological features within the flower bud outline are extracted and combined with flower bud volume. Through feature fusion analysis, a flower bud feature vector is constructed, including:

[0125] S801. Extract the morphological features of the region within the flower bud outline. The morphological features include the curvature of each vertex in the flower bud outline, the area ratio and distribution dispersion of the positive Gaussian curvature region and the negative Gaussian curvature region on the flower bud surface, and the axial asymmetry of the outlines on both sides of the flower bud.

[0126] S802. Analyze the pixel value distribution skewness and kurtosis of the flower bud outline, perform spectral reflectance simulation for the accumulation process of target pigments during maturation, and calculate the corresponding spectral index.

[0127] S803, Extract the texture features of the region within the flower bud outline;

[0128] S804. Merge morphological features, spectral indices, texture features, and bud volume to construct a bud feature vector.

[0129] In this embodiment, based on the flower bud outline, for each vertex, the normal and area information at the vertex are analyzed, and the average curvature of each vertex is calculated through surface fitting. A threshold close to zero, such as 0.001, is set, and vertices with curvature greater than this threshold are classified as positive Gaussian curvature regions, corresponding to convex regions, while vertices with curvature less than the negative threshold are classified as negative Gaussian curvature regions, corresponding to concave regions. The sum of the areas belonging to these two types of regions is calculated, and their respective proportions to the total surface area are also calculated. The distance from all vertices in the positive and negative Gaussian curvature regions to the three-dimensional mass of the flower bud is calculated. The distance to the center is calculated, and the standard deviation of the distance is used as the distribution dispersion, reflecting the degree of concentration or dispersion of curvature features on the flower bud surface. The main axis direction of the flower bud is determined, and a series of cross-sectional planes perpendicular to the main axis are selected uniformly along the main axis direction. On each cross-section, the outline formed by the intersection of the flower bud and the plane is obtained, and all intersection points are projected onto the cross-sectional plane. The two points closest to the main axis are selected, and the difference in distance from these two points to the main axis is calculated. This difference is calculated for all cross-sections, and the mean of the absolute values ​​of the differences is used as a measure of axial asymmetry, reflecting the uniformity of flower bud growth along the main axis.

[0130] It should be noted that vertex curvature provides local shape information, and the area ratio of positive and negative Gaussian curvature regions can effectively distinguish between different stages where the surface is predominantly convex and where depressions begin to increase. Distribution dispersion reflects whether these feature points are concentrated in specific parts of the bud or widely distributed throughout the entire surface, corresponding to different swelling patterns. Axial asymmetry provides a basis for judgment from the perspective of overall symmetry; an increase in irregularity is related to non-uniform growth or slight deformation during the maturation process. By calculating the combination of morphological features, subtle geometric changes can be identified, improving the ability to distinguish between adjacent maturity levels.

[0131] Specifically, based on the analysis of the skewness and kurtosis of pixel value distribution according to the flower bud outline, the pixel intensity values ​​of all pixels within the flower bud area are extracted from the R, G, and B channels respectively. For the intensity value set of each channel, its third and fourth statistical moments are calculated as skewness and kurtosis respectively. Skewness reflects the asymmetry of pixel value distribution. For example, when the green channel value distribution shifts to the left, i.e., more low values ​​appear, it reflects a reduction in chlorophyll. Kurtosis reflects the sharpness or flatness of the distribution, and the change in peak value is related to the change in color uniformity.

[0132] Preferably, based on plant physiology, an empirical relationship model is established between the target pigment content and RGB channel values. For example, chlorophyll has strong absorption in the red band and high reflectance in the near-infrared band. However, with only RGB information available, this relationship can be simulated by combining red and green channels. The green channel value is subtracted from the red channel value, and then divided by the sum of the green channel value, the red channel value, and a small constant (to prevent division by zero) to obtain the greenness index as a spectral index. This index value is positively correlated with chlorophyll content; as chlorophyll degrades with maturity, the index value decreases. Skewness, kurtosis, and spectral index values ​​are obtained.

[0133] It should be noted that by combining high-order color statistics and simulated spectral indices, the biochemical changes in the early stages of flower bud maturation can be sensitively detected. The skewness and kurtosis of pixel value distribution can capture subtle changes in the overall color distribution pattern. For example, when some areas of the flower bud begin to lose their green color, the pixel value distribution of the green channel will change from a single peak to a double peak, resulting in significant changes in skewness and kurtosis, which occurs earlier than changes in the average color value. Simulated spectral indices, on the other hand, correlate RGB information with the intrinsic pigment content through empirical models, enabling the analysis of maturation trends before obvious color changes are observed.

[0134] Specifically, the RGB image of the flower bud region is converted to a grayscale image, and a gray-level co-occurrence matrix (GLCM) is constructed to reflect the probability that a pixel with gray level i and a pixel with gray level j will appear simultaneously at a given spatial distance and direction. Based on the GLCM, contrast, correlation, energy, and homogeneity are calculated as texture features. Contrast measures the degree of local variation in the image, reflecting the sharpness and groove depth of the texture; correlation measures the similarity between rows or columns of elements in the GLCM, reflecting the linear directional regularity of the texture; energy measures the uniformity or smoothness of the image texture; and homogeneity measures the local homogeneity of the image texture.

[0135] It should be noted that by extracting texture features, subtle structural information in the visual pattern of flower buds that is difficult to describe using shape or color can be obtained. For example, the surface of young flower buds is smoother and shinier, while the surface of mature flower buds becomes slightly rougher or has fine wrinkles due to cell enlargement. By extracting texture features, they can be fused with morphological and spectral features, providing complementary information dimensions and improving the robustness and accuracy of maturity classification.

[0136] Specifically, morphological features, spectral features, texture features, and bud volume are integrated and spliced ​​in a predefined order to obtain a bud feature vector. Through feature fusion, multi-source information from geometry, optics, texture, and size is integrated into a unified feature vector, enabling joint decision-making. By utilizing the complementarity between features, accurate data is provided for the bud maturity analysis process, improving the accuracy of maturity analysis results.

[0137] Furthermore, based on the bud feature vector, the probability value of the bud belonging to a preset maturity level is calculated to obtain the maturity level of the bud, including:

[0138] S901. Based on the flower bud feature vector, analyze the maturity of the flower bud through a preset neural network model and calculate the probability value of the flower bud belonging to the preset maturity level.

[0139] S902. Identify the maturity of the flower buds according to the probability value.

[0140] In this embodiment, the preset neural network model includes, but is not limited to, a multilayer perceptron model. The number of neurons in the input layer of this model is equal to the dimension of the constructed flower bud feature vector. The model includes at least one hidden layer, and two hidden layers can be set. The number of neurons in the first hidden layer is 1.5 times the input dimension, and the number of neurons in the second hidden layer is the same as that in the first hidden layer. Each hidden layer is followed by a non-linear activation function, such as a modified linear unit function. The number of neurons in the output layer of the model is equal to the preset number of maturity levels, for example, three neurons, corresponding to the three levels of immature, moderately mature, and overripe. The output layer uses the Softmax function as the activation function, which activates each... The original output value of a neuron is transformed into a probability value between 0 and 1, and the sum of the probability values ​​of all output neurons is 1. The dataset required to train this model is obtained through historical accumulation. A large number of flower bud samples were collected over multiple growth cycles, and corresponding flower bud feature vectors were extracted. Experienced agricultural experts labeled each sample with its true maturity level. The classification cross-entropy loss function is used as the optimization objective of the model. The Adam optimizer is used during training, with an initial learning rate of 0.001. During training, the dataset is randomly divided into training and validation sets, and the model parameters are iteratively updated on the training set. After training, the model is deployed to an online monitoring system. The feature vector of the flower bud to be tested is input into the pre-trained model, and the model calculates and outputs a probability vector, for example [0.05, 0.88, 0.07], representing the probability estimates of whether the flower bud belongs to the three levels of immature, moderately mature, and overripe, respectively. Through the pre-trained neural network model, the complex relationship between features and maturity can be automatically learned and modeled, improving the accuracy of maturity analysis results.

[0141] Specifically, the maturity level corresponding to the highest probability value in the probability vector is selected as the maturity level of the flower bud. For example, if the probability vector is [0.05, 0.88, 0.07], then the maturity level corresponding to the maximum value of 0.88 is the identification result. This simple maximum value determination rule ensures the accuracy and efficiency of the decision-making process, providing accurate data references for subsequent system operations.

[0142] like Figure 4 As shown, a smart monitoring system for the maturity of cauliflower buds is used to implement a method for smart monitoring of the maturity of cauliflower buds, including:

[0143] The image acquisition module analyzes the bud distribution density based on the pre-acquired tree cauliflower images, and performs multi-angle image acquisition on areas where the bud distribution density is greater than a preset density threshold to obtain a first image set.

[0144] The flower bud analysis module performs temporal and spatial alignment on the images in the first image set, extracts the morphological parameters of the flower buds, identifies the flower bud outline, and calculates the flower bud volume.

[0145] The fusion calculation module extracts the morphological features of the region within the flower bud outline, combines them with the flower bud volume, and calculates the probability value of the flower bud belonging to the preset maturity level through feature fusion analysis to obtain the maturity of the flower bud.

[0146] The maturity monitoring module combines the maturity of flower buds with historical growth data to simulate the growth process of flower buds and predict the maturity time of flower buds, so as to intelligently monitor the maturity of tree cauliflower flower buds.

[0147] Example 2:

[0148] This embodiment uses a specific application scenario to describe in detail the complete implementation process of this technical solution. The solution is applied to a tree cauliflower planting base covering approximately five acres. A fixed, track-mounted mobile monitoring platform is installed within the base, integrating a high-definition visible light camera, a controllable pan-tilt unit, and an edge computing unit.

[0149] After the cauliflower enters the critical stage of bud development, the system initiates periodic monitoring, controlling the mobile platform to move along a preset track and perform a rapid global scan of the field from a vertical overhead angle, acquiring a high-resolution orthophoto covering most of the field, thus obtaining the pre-acquired cauliflower image. This image is transmitted in real time to the edge computing unit, where a pre-trained bud semantic segmentation model is run. This model, trained on a large amount of labeled data, can classify each pixel in the image as either a bud or background. After processing, a binarized bud distribution map is generated, where white pixel areas represent identified bud groups.

[0150] Connectivity analysis was performed on the distribution map, and interconnected white pixel blocks were marked as independent flower cluster regions, identifying a total of 120 such initial regions. For each region, the system calculated its bud distribution density, specifically the total number of bud pixels within the region divided by the area of ​​its smallest bounding rectangle. The calculations showed that the density values ​​of most regions fluctuated between 0.3 and 0.8. A preset density threshold of 0.6 was set, and all regions with densities greater than this threshold were filtered out, resulting in 35 regions identified as high-density regions, forming the first region set. For each of these 35 first regions, the coordinates of its bounding box center point were calculated, and multi-angle shooting paths were planned based on the position of this center point in the preset world coordinate system. For example, for an area with center point coordinates (X: 10.5m, Y: 3.2m, Z: 1.0m), the control unit carrying the camera is moved sequentially to four positions that can be aimed at the point with azimuth angles of 0°, 90°, 180°, and 270° and a pitch angle of -20°. A high-definition close-up image is captured at each position, completing the multi-angle acquisition of all 35 high-density areas, resulting in a total of 140 multi-angle images, forming the first image set.

[0151] Based on the first image set, temporal alignment is performed. Due to millisecond-level delays in image capture, the system uses the first image of each group as a reference, employing optical flow to estimate and compensate for the minute displacements of the flower buds caused by a light breeze, based on the image timestamps. Spatial alignment is then performed, extracting SIFT feature points from all images. Through feature matching and structure-of-motion (SOMO) algorithms, a sparse 3D point cloud of the entire scene is reconstructed, and the precise camera pose parameters for each image are optimized and calculated. Building upon this, a dense 3D point cloud containing the flower bud region is generated using multi-view stereo vision technology. Many flower buds in this point cloud are tightly clustered together. The curvature features of each point in the point cloud are calculated, and a density clustering algorithm fusing geometric distance and curvature similarity is used to separate the clustered flower bud point clouds.

[0152] For example, a cluster of approximately 5000 points was separated into three independent point cloud clusters, containing approximately 1800, 2200, and 1000 points respectively, each corresponding to an independent flower bud. For each separated flower bud point cloud, principal component analysis was used to determine its principal axis direction. The point cloud was then projected onto a plane perpendicular to the principal axis, and its two-dimensional projected contour was extracted using the Alpha Shapes algorithm. Virtual slices were made along the principal axis at intervals of 0.5 mm. The area of ​​the projected contour at each slice was calculated, and the volume of each flower bud was calculated by summing the product of the areas of all slices and the intervals. For example, the volumes of the three flower buds were calculated to be approximately 0.8 cubic centimeters, 2.2 cubic centimeters, and 0.9 cubic centimeters, respectively.

[0153] After obtaining the precise volume, multi-dimensional features are extracted from the 3D model and corresponding texture image of each flower bud. The statistical value of the vertex mean curvature, the area ratio and distribution dispersion of positive and negative Gaussian curvature regions, and axial asymmetry are calculated. Skewness and kurtosis of color channels are calculated from the RGB image, and a spectral index reflecting greenness is simulated. Four texture features—contrast, correlation, energy, and homogeneity—are extracted using the gray-level co-occurrence matrix. All features are concatenated with the volume values ​​to obtain the flower bud feature vector.

[0154] The flower bud feature vector is input into a pre-trained multilayer perceptron neural network model. This model has 10 input neurons, a hidden layer with 16 neurons, and three output neurons, corresponding to three maturity levels: immature, moderately mature, and overripe. The model is trained on 5000 historical samples and their expert-annotated labels, using the cross-entropy loss function and the Adam optimizer. After thorough training, it has been deployed in the system. The model performs forward computation on the input flower bud feature vector and outputs a probability distribution. For example, for a flower bud with a volume of 2.2 cubic centimeters, the model outputs a probability value of [0.02, 0.91, 0.07], indicating that it has a very high probability of belonging to the moderately mature level, reaching 91%. Based on the principle of maximizing probability, the system determines that the flower bud is moderately mature.

[0155] The system combines the current maturity assessment results of all flower buds with the historical volume and spectral index data of the buds over the past three days to fit their growth curves. By extrapolating the growth curves, it predicts when the key indicators of the buds will reach the center value of the optimal maturity threshold range. For example, it predicts that the aforementioned buds will reach the optimal harvest maturity point in approximately 48 hours. The system summarizes the maturity status and predicted maturity time of all flower buds, generates a monitoring report, and sends it to managers via a network interface, providing clear decision support for them to arrange precise harvesting plans.

[0156] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A method for intelligent monitoring of the maturity of cauliflower buds, characterized in that, include: Based on the pre-acquired images of tree cauliflower, the distribution density of flower buds is analyzed, and multi-angle image acquisition is performed on areas where the distribution density of flower buds is greater than a preset density threshold to obtain the first image set. The images in the first image set are aligned temporally and spatially, and the morphological parameters of the flower buds are extracted, the flower bud outline is identified, and the flower bud volume is calculated. Extract the morphological features of the region within the flower bud outline. The morphological features include the curvature of each vertex in the flower bud outline, the area ratio and distribution dispersion of the positive Gaussian curvature region and the negative Gaussian curvature region on the flower bud surface, and the axial asymmetry of the two sides of the flower bud outline. The pixel value distribution skewness and kurtosis of the flower bud outline are analyzed, and the spectral reflectance is simulated for the accumulation process of target pigments during the maturation process to calculate the corresponding spectral index. Extract texture features from the region within the flower bud outline; The morphological features, spectral indices, texture features, and bud volume are fused to construct a bud feature vector; Based on the flower bud feature vector, calculate the probability value of the flower bud belonging to the preset maturity level, and obtain the maturity of the flower bud. By combining the maturity of flower buds with historical growth data, the growth process of flower buds is simulated to predict the maturity time of flower buds, so as to intelligently monitor the maturity of tree cauliflower flower buds.

2. The intelligent monitoring method for the maturity of cauliflower buds according to claim 1, characterized in that, The process involves analyzing the bud distribution density based on pre-acquired tree flower images, and acquiring multi-angle images of areas where the bud distribution density is greater than a preset density threshold to obtain a first image set, including: Based on the pre-acquired images of tree cauliflower, the location of tree cauliflower is identified to obtain a bud distribution map; Analyze the connected components in the flower bud distribution map, treat each connected component as a flower bud region, and construct a set of flower bud regions. Based on the flower bud region set, multi-angle image acquisition is performed on regions where the flower bud distribution density is greater than a preset density threshold to obtain the first image set.

3. The intelligent monitoring method for the maturity of cauliflower buds according to claim 2, characterized in that, The first image set is obtained by acquiring multi-angle images of regions with a flower bud distribution density greater than a preset density threshold, according to the flower bud region set, including: Based on the flower bud region set, calculate the flower bud distribution density of each region, and filter out regions whose flower bud distribution density is greater than a preset density threshold to obtain the first region set; Analyze the boundary range of each region in the first region set and select the corresponding boundary midpoints; Based on the positional relationship between the midpoint of the boundary and the preset physical coordinate system, the corresponding acquisition angle is calculated, and multi-angle image acquisition is performed according to the acquisition angle to obtain the first image set.

4. The intelligent monitoring method for the maturity of cauliflower buds according to claim 1, characterized in that, The steps of performing temporal and spatial alignment on the images in the first image set, extracting the morphological parameters of the flower buds, identifying the flower bud outline, and calculating the flower bud volume include: The images in the first image set are aligned temporally and spatially, and the morphological parameters of the flower buds are extracted to identify the flower bud outlines and construct flower bud distribution data. Based on the bud distribution data, the curvature of the bud outline at different positions is calculated to obtain the outline features, and the individual bud data is separated and the corresponding bud volume is calculated.

5. The intelligent monitoring method for the maturity of cauliflower buds according to claim 4, characterized in that, The process of performing temporal and spatial alignment on the images in the first image set, extracting the morphological parameters of the flower buds, identifying the flower bud outlines, and constructing flower bud distribution data includes: The images in the first image set are time-aligned and spatially aligned to obtain the corrected image set; From the set of corrected images, extract the key points and morphological parameters of each image to construct a set of key point parameters; Calculate the matching degree of key point parameter sets between images, and select the two images with the highest matching degree as the image benchmark; Identify the flower bud outlines in the image baseline, and sequentially add the first image set to the corresponding distribution locations to construct flower bud distribution data.

6. The intelligent monitoring method for the maturity of cauliflower buds according to claim 5, characterized in that, The process of calculating the curvature of the flower bud contour at different positions based on the flower bud distribution data to obtain contour features, separating individual flower bud data, and calculating the corresponding flower bud volume includes: Based on the flower bud distribution data, the curvature of the flower bud outline at different positions is calculated to obtain the outline features. Clustering is then performed according to the outline features to separate individual flower bud data. The data of each individual flower bud is projected onto a vertical plane to identify the corresponding flower bud edge; Based on the flower bud edge, calculate the edge length and edge curvature, and calculate the corresponding flower bud volume by integration.

7. The intelligent monitoring method for the maturity of cauliflower buds according to claim 1, characterized in that, The step of calculating the probability value of a flower bud belonging to a preset maturity level based on the flower bud feature vector to obtain the maturity of the flower bud includes: Based on the flower bud feature vector, the maturity of the flower bud is analyzed through a preset neural network model, and the probability value of the flower bud belonging to the preset maturity level is calculated. The maturity of the flower buds is identified based on the probability values.

8. A smart monitoring system for the maturity of florets of a flowering vegetable, characterized in that, A method for intelligent monitoring of the maturity of tree cauliflower buds as described in any one of claims 1 to 7, comprising: The image acquisition module analyzes the bud distribution density based on the pre-acquired tree cauliflower images, and performs multi-angle image acquisition on areas where the bud distribution density is greater than a preset density threshold to obtain a first image set. The flower bud analysis module performs temporal and spatial alignment on the images in the first image set, extracts the morphological parameters of the flower buds, identifies the flower bud outline, and calculates the flower bud volume. The fusion calculation module extracts the morphological features of the region within the flower bud outline, combines them with the flower bud volume, and calculates the probability value of the flower bud belonging to the preset maturity level through feature fusion analysis to obtain the maturity of the flower bud. The maturity monitoring module combines the maturity of flower buds with historical growth data to simulate the growth process of flower buds and predict the maturity time of flower buds, so as to intelligently monitor the maturity of tree cauliflower flower buds.

Images