Grading method of multi-head Chinese rose fresh cut flowers

Abstract

The invention relates to a grading method of multi-head Chinese rose fresh cut flowers, and relates to the technical field of grading of fresh cut flowers. According to the multi-head Chinese rose fresh cut flower grading method, illumination interference, shooting angle difference and background noise can be effectively eliminated through a multi-stage standardized preprocessing process and denoising-segmentation-enhancement, feature loss is remarkably reduced, flower areas are accurately extracted in combination with a petal mask segmentation technology, required features are enhanced, and the quality of the multi-head Chinese rose fresh cut flowers is improved. High-quality input is provided for subsequent feature analysis, and the problem of detection deviation caused by insufficient preprocessing in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grading fresh-cut flowers, and particularly relates to a grading method for fresh-cut multiflora roses. BACKGROUND

[0002] As a core category in the fresh-cut flower market, the grading quality of multiflora roses is directly related to production efficiency, market circulation efficiency and industrial competitiveness.

[0003] Traditional grading of multiflora roses relies on manual visual inspection and experience-based judgment, but such a grading method has been difficult to adapt to the needs of large-scale production and standardized trade. Manual grading is inefficient, has poor consistency, and is costly, resulting in that the flowering period of fresh-cut multiflora roses cannot meet the needs of the insurance period, logistics transportation and market sales of fresh-cut flowers.

[0004] At present, most existing fresh-cut flower grading devices only perform simple parameter detection and cannot accurately identify and comprehensively analyze complex features such as flower integrity and pest traces, so the identification effect cannot meet the grading requirements, resulting in economic losses. SUMMARY

[0005] To solve or partially solve the problems in the related art, the present application provides a grading method for fresh-cut multiflora roses, which can accurately identify and comprehensively analyze complex features and effectively ensure the accuracy of identification.

[0006] The present application provides a grading method for fresh-cut multiflora roses, which is applied to a fresh-cut flower grading system. The grading method comprises the following steps: An original image of a fresh-cut multiflora rose to be graded is collected by a multi-view cooperative detection unit, and the original image is preprocessed to obtain a standardized fresh-cut flower image. The multi-view cooperative detection unit is provided with three camera positions, including a front 45°, a side 90° and a top view, and multi-angle images of a single sample are synchronously collected. Target information is extracted from the standardized fresh-cut flower image, including flower stem length, bud number, flower integrity, flower color purity and pest traces, to obtain a multi-dimensional grading feature set of the fresh-cut flower. Based on grading rules, the multi-dimensional grading feature set is converted into a grade determination result, and a control instruction is generated to drive a grading execution mechanism to push the fresh-cut flower to a corresponding grade collection area.

[0007] Optionally, in some embodiments of the present application: The original image is preprocessed, specifically including: The original image is denoised to obtain a denoised image, wherein the denoising includes median filtering and bilateral filtering. The denoised image is input into a segmentation network model to generate an accurate petal mask area, and a segmented image is obtained. The segmented image is enhanced to highlight key features, and an enhanced image is obtained.

[0008] Optionally, in some embodiments of the present application: The segmented image is enhanced to highlight key features, and an enhanced image is obtained. Bud edge enhancement: through Laplacian operator enhancement, the contour edge of the bud in the front 45° image is strengthened to assist flower shape integrity detection; Stalk contrast enhancement: through adaptive histogram equalization, the contrast between the stalk and the background in the side 90° image is improved to assist length measurement; Flower color uniformity correction: based on the brightness correction of the illumination model, the brightness difference caused by uneven illumination on the flower head surface is eliminated to assist flower color purity detection.

[0009] Optionally, in some embodiments of the present application: The standardized cut flower image is subjected to target information extraction processing, which specifically includes: Based on the side 90° view image, the length of the cut flower is extracted through "pixel-actual size calibration" for length measurement, which specifically includes: The straight line edge of the bottom of the stalk is identified through the edge detection algorithm of the Canny operator, and the midpoint is taken as the bottom point A. The highest bud is located through the top overhead view, and the top point B is extracted in the side image. In the side image coordinate system, the Euclidean distance between A(x1, y1) and B(x2, y2) is calculated: The "pixel-millimeter" conversion coefficient k is calibrated by a standard ruler, and the unit is mm / pixel, so the actual length is: .

[0010] Optionally, in some embodiments of the present application: The standardized cut flower image is subjected to target information extraction processing, which specifically includes: Based on three-view image fusion, the number of flower buds is extracted through "target detection + de-duplication matching" for accurate counting, which specifically includes: For each view image, a flower bud detection model is used to output a set of flower bud candidate boxes for each view: Front view: ; Side view: ; Top view: ; Based on 3D spatial coordinate mapping, the IoU overlap of candidate boxes from different viewpoints is calculated. When IoU ≥ 0.7, they are determined to be the same flower bud, and the Hungarian algorithm is used to achieve matching and deduplication. IoU calculation: two candidate boxes b a (x) a1 ,y a1 ,x a2 ,y a2 The crossover ratio of bᵦ(xᵦ1,yᵦ1,xᵦ2,yᵦ2): Final count: .

[0011] Optionally, in some embodiments of this application: This involves extracting target information from standardized cut flower images, specifically including: Based on a 45° frontal camera image, feature extraction for the integrity of the flower shape is performed through "standard template comparison + defect detection," specifically including: The GrabCut algorithm is used to segment the area of ​​a single flower head, and its edge contour C is extracted. The standard flower shape template C0 for this variety is then used to calculate the Iou overlap between C and C0. Image thresholding segmentation technology was used to detect defects such as damage and wormholes in the outer petals, and the proportion of defective area to the total area of ​​the flower head was calculated. Flower shape integrity score: .

[0012] Optionally, in some embodiments of this application: The standard cut flower image is processed to extract target information, specifically including: Based on a 45° frontal camera image, the deviation is calculated in the HSV color space, and features of the color purity of cut flowers are extracted, specifically including: Flower head color region extraction: The flower head region is segmented using a mask, and the H and S values ​​of all pixels within the region are extracted and the mean is calculated. Standard color acquisition: Retrieve standard color chart data; Color deviation rate calculation: Color purity rating: For lighter-colored varieties, hue is more important, so take... For dark-colored varieties, saturation is more important. .

[0013] Optionally, in some embodiments of this application: The standard cut flower image is processed to extract target information, specifically including: Based on top-down and frontal 45° camera images, deep learning is used to segment lesion areas and extract pest and disease trace features, specifically including: The leaf segmentation model is trained to segment all leaf regions, resulting in a leaf pixel set L. A pest and disease segmentation model is then trained to segment pest and disease regions, resulting in a pixel set P. The percentage of pest and disease area is calculated as follows: This represents the number of pixels in the affected area. This represents the total number of pixels on the leaf blade.

[0014] Optionally, in some embodiments of this application: The extracted multidimensional hierarchical feature set is transformed into a level determination result, specifically including: The multi-dimensional hierarchical feature set is input into the PLC-based automated control system for analysis to obtain hierarchical control instructions for multi-headed cut roses. Based on the multi-headed rose cut flower grading control command, the action tasks are assigned to the corresponding single-control two-position five-way solenoid valve. Once the cut flowers are identified and located by the corresponding photoelectric sensor, the single-control two-position five-way solenoid valve drives the corresponding single-acting cylinder to perform a sorting operation, pushing the cut flowers to the corresponding grade collection area.

[0015] Optionally, in some embodiments of this application: This grading method for multi-headed cut roses is applicable to grading systems for multi-headed cut roses. The grading system for multi-headed cut roses includes: a visual grading system and a conveying grading system; This visual grading system includes: A housing used to hold hardware devices; A multi-view collaborative detection unit for acquiring original images of multi-headed cut roses to be graded includes: a top-view camera, a front 45° camera, and a side 90° camera. A circular light source used to eliminate shadows and unify lighting; Industrial touchscreens used to provide visual human-machine interfaces; Memory used for data storage, processing, and cross-module linkage; Industrial computers used to run machine vision algorithms and hierarchical decision models; PLC controller used for instruction execution and equipment linkage control; The conveying and grading system includes: The grading conveyor belt is used to move fresh cut flowers to be graded to the corresponding grade column, and the smooth guide rail is used to remove the flowers without damage. A photoelectric sensor is used to detect the position of the flower, and a cylinder is used to push the flower to the corresponding grade guide rail. Multiple collection bins are used to collect cut flowers from the target grading zone.

[0016] The technical solution provided in this application may include the following beneficial effects: This application employs a multi-stage standardized preprocessing workflow—denoising, segmentation, and enhancement—to effectively eliminate lighting interference, differences in shooting angles, and background noise, significantly reducing feature loss. Combined with petal mask segmentation technology, it accurately extracts flower regions, enhances the required features, and provides high-quality input for subsequent feature analysis, thus solving the detection bias problem caused by insufficient preprocessing in existing technologies.

[0017] This application reduces feature loss caused by flower bud occlusion through multi-view synchronous acquisition and standardized preprocessing, avoids single-view detection bias, and enables accurate identification and comprehensive analysis of complex features, effectively ensuring the accuracy of identification. At the same time, it adopts a variety-adaptive decision-making strategy to overcome the limitation of fixed algorithms that are difficult to adapt to the characteristics of different varieties. In addition, it combines a non-destructive linkage execution mechanism to reduce the sample damage rate during the grading process.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The above and other objects, features and advantages of this application will become more apparent from the more detailed description of exemplary embodiments thereof in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments thereof.

[0020] Figure 1 This is a schematic diagram of the grading method for multi-headed cut roses in the embodiments of this application; Figure 2 This is a schematic diagram of the grading system for multi-headed cut roses in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the visual grading system in the embodiments of this application; Figure 4 This is a schematic diagram of the conveying and grading system in an embodiment of this application.

[0021] Reference numerals: 1-Vision grading system, 101-House, 102-Industrial computer, 103-PLC controller, 104-Industrial touch screen, 105-Circular light source, 106-Top-view camera, 107-Front 45° camera, 108-Side 90° camera, 2-Conveying and grading system, 201-Grading conveyor belt, 202-Smooth guide rail, 203-Photoelectric sensor, 204-Cylinder, 3-Collection bin. Detailed Implementation

[0022] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to make this application more thorough and complete, and to fully convey the scope of this application to those skilled in the art.

[0023] It should be understood that although the terms "first," "second," "third," etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0024] In the description of this application, it should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0025] Unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0026] Traditional grading of multi-headed roses relies on manual visual inspection and experience, but this grading method is no longer suitable for the needs of large-scale production and standardized trade. Manual grading is inefficient, inconsistent, and costly, resulting in the flowering period of multi-headed rose cut flowers no longer meeting the needs of cut flower insurance, logistics, and market sales.

[0027] Currently, most existing cut flower grading devices only perform simple parameter detection and cannot accurately identify and comprehensively analyze complex characteristics such as flower shape integrity and traces of pests and diseases. The identification effect cannot meet the grading requirements, resulting in economic losses.

[0028] To address the aforementioned issues, this application provides a grading method for multi-headed cut roses, which can accurately identify and comprehensively analyze complex features, effectively ensuring the accuracy of identification.

[0029] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0030] See Figures 1-4 A grading method for multi-headed rose cut flowers, applied to a cut flower grading system, comprising: S100: The multi-view collaborative detection unit acquires the original images of the multi-headed rose cut flowers to be graded and preprocesses the original images to obtain standardized cut flower images. The multi-view collaborative detection unit is equipped with three camera positions, including: front 45°, side 90°, and top view, to simultaneously acquire multi-angle images of a single sample.

[0031] Specifically: the original image undergoes preprocessing, which includes: The original image is then denoised to obtain a denoised image. The denoising process includes median filtering and bilateral filtering. Specifically: Median filtering: Perform a 3×3 windowed mid-range filter on the image to remove salt-and-pepper noise and scatter artifacts: median is the median operation. The image after denoising; Bilateral filtering: To smooth noise while preserving edge details, the parameters are set as follows: in and Gaussian functions for the spatial domain and pixel domain, respectively, are set. =10, =25.

[0032] In this embodiment, by performing noise reduction processing, random noise and background artifacts caused by changes in lighting, camera noise, and reflections on the surface of petals during the shooting process can be removed to obtain a denoised image.

[0033] The denoised image is input into a segmentation network model to generate an accurate petal mask region, resulting in a segmented image. Specifically: Obtain a flower image sample dataset, which includes individual sample flower images used to train the segmentation network model and their corresponding petal edge information; A segmentation network model was trained based on a semantic segmentation model in deep learning and a sample set of flower images.

[0034] In this embodiment, the generation of an accurate petal mask region by a segmentation network model ensures that subsequent feature extraction is performed only on the flower region, avoiding the bias introduced by background noise.

[0035] The segmented image is enhanced to highlight key features, resulting in an enhanced image.

[0036] Specifically: the segmented image is enhanced to highlight key features, including: Flower bud edge enhancement: The Laplacian operator is used to enhance the contour edges of flower buds in a 45° frontal image, aiding in flower shape integrity detection. Specifically: c is the enhancement factor, set to 0.5 to avoid excessive noise amplification. This is the image of the segmented flower head region.

[0037] In this embodiment, the Laplacian operator can efficiently enhance the edges and details in the image, reduce blur or noise, and the enhancement coefficient c is set to 0.5, which suppresses noise amplification while highlighting the edges, ensuring that the processed image is not distorted, and facilitating more accurate segmentation of the flower head area in the future.

[0038] Stem contrast enhancement: Adaptive histogram equalization improves the contrast between the stem and background in 90° side views, aiding in length measurement. Specifically: tileSize is the size of the local area, 8 8. Adapts to the characteristics of slender stems; clipLimit controls the upper limit of contrast to avoid overexposure.

[0039] In this embodiment, adaptive histogram equalization makes the stem edges more distinct, enhances stem contrast, improves the accuracy and efficiency of length measurement, and reduces the false detection rate of length.

[0040] Flower color uniformity correction: Based on the illumination model, brightness correction eliminates brightness differences on the flower head surface caused by uneven illumination, assisting in the detection of flower color purity. Specifically: Gaussian blur is applied to the V channel of the flower head region. ), to obtain the illumination component ; Corrected brightness: , This is the original V channel mean; The corrected V channel is recombined with the original H and S channels to form an HSV image, which is then converted back to RGB for subsequent pattern extraction.

[0041] In this embodiment, the illumination component is extracted by Gaussian blur, which simulates real lighting conditions and corrects the brightness difference on the flower head surface caused by shadows or reflections. This makes the color extraction more uniform, avoids color misjudgment caused by changes in lighting, and ensures that the grading results meet actual needs.

[0042] In this embodiment, a multi-stage standardized preprocessing process—denoising, segmentation, and enhancement—effectively eliminates lighting interference, shooting angle differences, and background noise, significantly reducing feature loss. Combined with petal mask segmentation technology, the flower region is accurately extracted, and the required features are enhanced, providing high-quality input for subsequent feature analysis. This solves the detection bias problem caused by insufficient preprocessing in existing technologies.

[0043] S200: Extract target information from standardized cut flower images. The target information includes flower stem length, number of flower buds, flower shape integrity, flower color purity, and traces of pests and diseases, resulting in a multi-dimensional hierarchical feature set of cut flowers.

[0044] Specifically: the target information extraction process for the standardized cut flower images includes: Based on a 90° side view image, the length of the cut flower is extracted using a "pixel-to-actual-size calibration" method. Specifically, this includes: The Canny edge detection algorithm is used to identify the straight edge at the bottom of the stem, and the midpoint is taken as the bottom point A. The highest flower bud is located with the assistance of a top-down camera position, and its vertex coordinates are extracted from the side view image as the top point B. In the side view image coordinate system, the Euclidean distance between A(x1,y1) and B(x2,y2) is calculated. By calibrating the pixel-to-millimeter conversion factor k using a standard ruler, with the unit being mm / pixel, the actual length is: .

[0045] In this embodiment, the Canny operator accurately identifies the stem edge, avoiding mispositioning caused by flower tilt or background interference. Multi-view image collaboration ensures the reliability of endpoint coordinates, reduces deviations caused by occlusion, and provides a reliable data foundation for subsequent grading decisions.

[0046] Specifically: the target information extraction process for the standardized cut flower images includes: Based on three-view image fusion, the method achieves accurate counting and feature extraction of flower bud count through "object detection + deduplication matching," specifically including: For each viewpoint image, a flower bud detection model is used to output a set of candidate flower bud boxes for each viewpoint: Frontal view: ; Side view: ; Top view: ; Based on 3D spatial coordinate mapping, the IoU overlap of candidate boxes from different viewpoints is calculated. When IoU ≥ 0.7, they are determined to be the same flower bud, and the Hungarian algorithm is used to achieve matching and deduplication. IoU calculation: two candidate boxes b a (x) a1 ,y a1 ,x a2 ,y a2 The crossover ratio of bᵦ(xᵦ1,yᵦ1,xᵦ2,yᵦ2): Final count: .

[0047] The construction of this flower bud detection model includes: Obtain a flower image sample dataset, which includes sample flower images used to train the YOLOv8 object detection model and their corresponding flower bud annotation information; A flower bud detection model was obtained by training a YOLOv8 object detection network model in deep learning and a flower image sample set.

[0048] In this embodiment, the YOLOv8 model can quickly detect candidate boxes for flower buds, adapt to different lighting and pose changes, and ensure the consistency of multi-view data through 3D coordinate mapping and the Hungarian algorithm. This effectively avoids duplicate counting, improves processing efficiency, and provides a reliable data foundation for subsequent hierarchical decision-making.

[0049] Specifically: the target information extraction process for the standardized cut flower images includes: Based on a 45° frontal camera image, feature extraction for the integrity of the flower shape is performed through "standard template comparison + defect detection," specifically including: The GrabCut algorithm is used to segment the area of ​​a single flower head, and its edge contour C is extracted. The standard flower shape template C0 for this variety is then used to calculate the Iou overlap between C and C0. Image thresholding segmentation technology was used to detect defects such as damage and wormholes in the outer petals, and the proportion of defective area to the total area of ​​the flower head was calculated. Flower shape integrity score: .

[0050] In this embodiment, the rabbitCut algorithm can accurately segment the flower head region, ensuring the integrity of the contour extraction. Furthermore, it provides a quantifiable similarity index through standard template comparison, which can provide a reliable data foundation for subsequent hierarchical decision-making.

[0051] Specifically: the target information extraction process for the standardized cut flower images includes: Based on a 45° frontal camera image, the deviation is calculated in the HSV color space, and features of the color purity of cut flowers are extracted, specifically including: Flower head color region extraction: The flower head region is segmented using a mask, and the H and S values ​​of all pixels within the region are extracted and the mean is calculated. Standard color acquisition: Retrieve standard color chart data; Color deviation rate calculation: Color purity rating: For lighter-colored varieties, hue is more important, so take... For dark-colored varieties, saturation is more important. .

[0052] In this embodiment, mask segmentation enables color extraction to be performed only on the flower head area, avoiding background interference. Furthermore, adaptive weighting improves adaptability to different varieties, reduces misjudgment due to color, and provides a reliable data foundation for subsequent grading decisions.

[0053] Specifically: the target information extraction process for the standardized cut flower images includes: Based on top-down and frontal 45° camera images, deep learning is used to segment lesion areas and extract pest and disease trace features, specifically including: The leaf segmentation model is trained to segment all leaf regions, resulting in a leaf pixel set L. A pest and disease segmentation model is then trained to segment pest and disease regions, resulting in a pixel set P. The percentage of pest and disease area is calculated as follows: This represents the number of pixels in the affected area. This represents the total number of pixels on the leaf blade.

[0054] The construction of the blade segmentation model includes: Obtain a leaf image sample dataset, which includes sample leaf images used to train the U-Net model and their corresponding leaf annotation information; A leaf segmentation model is obtained by training the U-Net segmentation network model in deep learning and the leaf image sample set.

[0055] The construction of the pest and disease segmentation model includes: Obtain a sample dataset of leaf disease and pest images, which includes sample leaf disease and pest images used to train the U-Net model and their corresponding leaf disease and pest annotation information. A disease and pest segmentation model is obtained by training the U-Net segmentation network model in deep learning and the leaf disease and pest image sample set.

[0056] In this embodiment, the pest and disease segmentation model can quickly and accurately identify minute traces such as insect holes and spots, providing a reliable data foundation for subsequent classification decisions.

[0057] In this embodiment, multi-view synchronous acquisition and standardized preprocessing are used to reduce feature loss caused by flower bud occlusion and avoid single-view detection bias. A variety-adaptive decision-making strategy is adopted to overcome the limitation that fixed algorithms are difficult to adapt to the characteristics of different varieties. At the same time, the sample damage rate during the grading process is reduced by combining a non-destructive linkage execution mechanism.

[0058] S300: Based on the grading rules, the extracted multi-dimensional grading feature set is transformed into a grading determination result, and control instructions are generated to drive the grading execution mechanism to push the cut flowers to the corresponding grading collection area.

[0059] Specifically: the extracted multidimensional hierarchical feature set is transformed into a level determination result, which includes: The multi-dimensional hierarchical feature set is input into the PLC-based automated control system for analysis, resulting in hierarchical control instructions for multi-headed cut roses. Specifically: Based on the grading rules of GB / T 41201-2021, the five core features extracted—flower branch length, number of flower buds, flower shape integrity, flower color purity, and traces of pests and diseases—are converted into grading results, and control commands are generated to drive the grading execution mechanism. When the characteristic indicators span multiple levels, the principle of "choosing the lower level" shall be followed.

[0060] Based on the multi-head rose cut flower grading control command, action tasks are assigned to the corresponding single-control two-position five-way solenoid valves. Specifically, after the decision module outputs the grade, a control command is sent to the PLC to drive the pneumatic actuator and other actuators to complete the grading: The control command (Code) is sent to the actuator via MODBUS to assign action tasks to the corresponding single-control two-position five-way solenoid valve, which is used to drive the cylinder to perform staged actions.

[0061] Once the cut flowers are identified and located by the corresponding photoelectric sensor, the single-control two-position five-way solenoid valve drives the corresponding single-acting cylinder to perform a sorting operation, pushing the cut flowers to the corresponding grade collection area.

[0062] Specifically: the grading method for multi-headed cut roses is applicable to the grading system for multi-headed cut roses; the grading system for multi-headed cut roses includes: a visual grading system 1 and a transport grading system 2; The visual grading system 1 includes: Housing 101 for housing hardware devices; The multi-view collaborative detection unit for acquiring original images of multi-headed cut roses to be graded includes: a top-view camera 106, a front 45° camera 107, and a side 90° camera 108. Circular light source 105 used to eliminate shadows and unify lighting; Industrial touchscreen 104 for providing a visual human-machine interface; An industrial computer 102 for running machine vision algorithms and hierarchical decision models, and a memory for data storage, processing and cross-module linkage. PLC controller 103 is used for instruction execution and equipment linkage control.

[0063] The conveying and grading system 2 includes: The grading conveyor belt 201 is used to move fresh cut flowers to be graded to the corresponding grade column; Smooth guide rail 202 for non-destructive flower cutting; The photoelectric sensor 203 is used to detect the position of the flower, and the cylinder 204 is used to push the flower to the corresponding grade guide rail. Multiple collection bins 3 are used to collect cut flowers from the target grading area.

[0064] The technical solution provided in this application may include the following beneficial effects: This application employs a multi-stage standardized preprocessing workflow—denoising, segmentation, and enhancement—to effectively eliminate lighting interference, differences in shooting angles, and background noise, significantly reducing feature loss. Combined with petal mask segmentation technology, it accurately extracts flower regions, enhances the required features, and provides high-quality input for subsequent feature analysis, thus solving the detection bias problem caused by insufficient preprocessing in existing technologies.

[0065] This application reduces feature loss caused by flower bud occlusion through multi-view synchronous acquisition and standardized preprocessing, avoids single-view detection bias, and can accurately identify and comprehensively analyze complex features, effectively ensuring the accuracy of identification. At the same time, it adopts a variety-adaptive decision-making strategy to overcome the limitation of fixed algorithms that are difficult to adapt to the characteristics of different varieties. In addition, it combines a non-destructive linkage execution mechanism to reduce the sample damage rate during the grading process.

[0066] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0068] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0069] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0070] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method of grading fresh cut Rosa multiflora flowers, characterized by, The grading method is applied to a fresh-cut flower grading system, and the grading method comprises: The original image of the multi-flower Rosa chinensis hybrid cut flower to be graded is collected by a multi-view cooperative detection unit, and the original image is preprocessed to obtain a standardized fresh-cut flower image, wherein the multi-view cooperative detection unit is provided with three camera positions, including: front 45°, side 90°, and top view, and multi-angle images of a single sample are synchronously collected; Target information extraction processing is performed on the standardized fresh-cut flower image, and the target information includes flower stem length, bud number, flower shape integrity, flower color purity, and pest trace, and a multi-dimensional grading feature set of the fresh-cut flower is obtained; Based on the grading rules, the multi-dimensional grading feature set extracted is converted into a grade determination result, and a control instruction is generated to drive the grading execution mechanism to push the fresh-cut flower to the corresponding grade collection area.

2. The grading method of the multi-flower Rosa chinensis hybrid fresh-cut flower according to claim 1, wherein: The preprocessing of the original image specifically comprises: The original image is denoised to obtain a denoised image, wherein the denoising processing includes: median filter processing, bilateral filter processing; The denoised image is input into a segmentation network model to generate an accurate petal mask region, and a segmented image is obtained; The segmented image is enhanced to highlight the key features, and an enhanced image is obtained.

3. The grading method of the multi-flower Rosa chinensis hybrid fresh-cut flower according to claim 2, wherein: The enhancement of the segmented image to highlight the key features specifically comprises: Bud edge enhancement: the Laplacian operator is used for enhancement to strengthen the contour edge of the bud in the front 45° image, which assists in the detection of flower shape integrity; Stalk contrast enhancement: adaptive histogram equalization is used to improve the contrast between the stalk and the background in the side 90° image, which assists in length measurement; Flower color uniformity correction: based on the brightness correction of the illumination model, the brightness difference caused by uneven illumination on the flower head surface is eliminated, which assists in the detection of flower color purity.

4. The grading method of the multi-flower Rosa chinensis hybrid fresh-cut flower according to claim 1 or 3, wherein: The target information extraction processing of the standardized fresh-cut flower image specifically comprises: Based on the side 90° camera position image, the length of the cut flower is extracted by "pixel-actual size calibration" for length measurement, which specifically comprises: The straight edge of the bottom of the stalk is identified by the edge detection algorithm of the Canny operator, the midpoint is taken as the bottom point A, the highest bud is located with the assistance of the top view camera position, and the top point coordinates of the highest bud are extracted as the top point B in the side image, and the Euclidean distance between A(x1, y1) and B(x2, y2) is calculated in the side image coordinate system: The "pixel-millimeter" conversion factor k is calibrated by a standard ruler, in mm / pixel, so that the actual length is: .

5. The grading method of the multi-flower Rosa chinensis hybrid fresh-cut flower according to claim 4, wherein: The target information extraction processing of the standardized fresh-cut flower image specifically comprises: Based on three-view image fusion, the flower bud number is accurately counted by "target detection + duplicate matching", which specifically comprises: For each view image, a flower bud detection model is used to output a set of flower bud candidate boxes for each view: frontal view: ; Side view: ; Top view: ; The IoU coincidence degree of the candidate boxes of different view angles is calculated based on 3D space coordinate mapping, and when IoU≥0.7, it is determined as the same flower bud, and the matching and deduplication are realized by using the Hungarian algorithm: IoU computation: Intersection over Union of two candidate boxes b a (x a1 ,y a1 ,x a2 ,y a2 ) and bᵦ(xᵦ1,yᵦ1,xᵦ2,yᵦ2): Final count: .

6. The grading method of Rosa hybrida fresh-cut flowers according to claim 5, characterized in that: The target information extraction processing of the standardized fresh-cut flower image specifically includes: Based on the 45° front view image, the feature extraction of flower shape integrity is realized through "standard template comparison + defect detection", specifically including: The GrabCut algorithm is used to segment the single flower head region, extract its edge contour C, call the standard flower shape template C0 of the variety, and calculate the Iou coincidence degree of C and C0: The image threshold segmentation technology is used to detect the damage, wormhole and other defects of the outer petals, and calculate the proportion of the defect area to the total area of the flower head: Flower integrity score: .

7. The grading method of Rosa hybrida fresh-cut flowers according to claim 6, characterized in that: The target information extraction processing of the standardized fresh-cut flower image specifically includes: Based on the 45° front view image, the deviation is calculated in the HSV color space, and the flower color purity of the fresh-cut flower is extracted, specifically including: Flower color area extraction: the flower head area is segmented by Mask, and the H and S values of all pixels in the area are extracted, and the average value is calculated: Standard color acquisition: call the standard color card data; Color deviation rate calculation: Flower color purity score: For weights, hue is more important for light color varieties, taken , saturation is more important for dark color varieties, taken .

8. The grading method of Rosa hybrida fresh-cut flowers according to claim 7, characterized in that: The target information extraction processing of the standardized fresh-cut flower image specifically includes: Based on the top view + 45° front view image, the disease spot area is segmented by deep learning, and the disease and pest trace feature extraction is realized, specifically including: Train the leaf segmentation model to segment all leaf areas to get the leaf pixel set L, and train the disease and pest segmentation model to segment the disease and pest area pixel set P, and the disease and pest area ratio: is the number of pixels in the pest area, is the total number of pixels in the leaf.

9. The grading method of Rosa hybrida fresh-cut flowers according to claim 1, characterized in that: The multi-dimensional grading feature set is converted into a grade judgment result, specifically including: The multi-dimensional grading feature set is input into the automatic control system based on PLC controller (103) for analysis, and the grading control instruction of Rosa hybrida fresh-cut flowers is obtained; Based on the grading control instruction of Rosa hybrida fresh-cut flowers, the action task is allocated to the corresponding single-control two-position five-way electromagnetic valve; When the fresh-cut flower is recognized and positioned by the corresponding photoelectric sensor (203), the single-control corresponding two-position five-way electromagnetic valve drives the corresponding single-acting cylinder (204) to execute the sorting operation, and pushes the fresh-cut flower to the corresponding grade collection area.

10. The grading method of Rosa hybrida fresh-cut flowers according to claim 1, characterized in that: The grading method of Rosa hybrida fresh-cut flowers is suitable for a grading system of Rosa hybrida fresh-cut flowers; The grading system of Rosa hybrida fresh-cut flowers includes a visual grading system (1) and a conveying grading system (2); The visual grading system (1) includes: a shell (101) for carrying hardware devices; A multi-view cooperative detection unit for acquiring the original image of the multi-headed rose fresh-cut flower to be graded, comprising: a top-down camera (106), a front 45° camera (107), and a side 90° camera (108); A circular light source (105) for eliminating shadows and unifying illumination; An industrial touch screen (104) for providing a visual human-computer interaction interface; A memory for data storage, processing, and cross-module linkage; An industrial computer (102) for running machine vision algorithms and grading decision models; A PLC controller (103) for instruction execution and device linkage control; The conveying and grading system (2) comprises: A grading conveyor belt (201) for moving the fresh-cut flower to be graded to the corresponding grade column, and a smooth guide rail (202) for lossless flower dropping; A photoelectric sensor (203) for detecting the position of the flower, and a pneumatic cylinder (204) for pushing the flower to the corresponding grading guide rail; A plurality of collection columns (3) for collecting fresh-cut flowers in the target grading area.

Images