Automatic control method and device for flow line operation of sweet potato seedling raising

Abstract

The application discloses a kind of automatic control method and equipment for sweet potato seedling seedling pipeline operation, it utilizes differential pulse modulation to collect the excited state and background state image of vine, through space self-adaptive weighted difference modeling and non-local mean regularization enhancement technology, effectively suppress environmental light noise and retain weak biological fluorescence signal, to generate the high confidence vine node distribution map.On this basis, abandon traditional fixed-length cutting mode, construct dynamic programming model based on agronomic constraints, solve the optimal segmentation strategy of maximum output rate in the whole vine range.Finally, through the space-time fusion of visual coordinates and conveyor encoder, drive multi-axis servo system to track and accurately work the real-time virtual cutting point of dynamic movement, realize the efficient, low-loss automatic production of sweet potato seedling.

Description

Technical Field

[0001] This application relates to the field of agricultural automation technology, and more specifically, to an automated control method and equipment for a production line operation of sweet potato seedling cultivation. Background Technology

[0002] As an important food and industrial raw material crop, the standardization and automation level of the virus-free seedling propagation process of sweet potatoes directly determines the final planting yield and economic benefits. In modern sweet potato seedling factories, the core process is to precisely cut the continuously growing long vines into qualified segments (seedlings) that meet specific agronomic requirements (e.g., three sections per seedling). With rising agricultural labor costs, developing automated control systems for assembly line operations that can replace manual labor has become an inevitable trend in the industry.

[0003] However, most existing automatic cutting technologies follow the logic of industrial pipe processing, employing mechanical fixed-length cutting or cutting methods based on simple photoelectric timing. This rigid control logic ignores the significant non-standard characteristics of sweet potato vines as living organisms. Specifically, the internode spacing of the vines is greatly affected by light, water, fertilizer, and varietal genes, exhibiting a random, variable length distribution. Fixed-length cutting easily leads to insufficient effective nodes in the cut seedling segments (e.g., containing only one node) or accidental damage to buds at the cutting point, resulting in a large number of blind nodes or waste, severely reducing raw material utilization and finished product qualification rates. Although some advanced solutions attempt to introduce machine vision to identify node positions to assist decision-making, in actual high-speed assembly line operations, it is difficult to accurately distinguish stem nodes from petioles when they are obscured or against a background of the same color using conventional RGB images. While using the fluorescence characteristics of plant chlorophyll for imaging detection is a potential technical approach, existing dual-frame differential fluorescence extraction methods perform poorly under complex industrial ambient lighting conditions. Simple image subtraction operations often neglect the statistical independence of photon shot noise, leading to a deterioration in the signal-to-noise ratio after differentiation and making it impossible to obtain clear, high-fidelity nodal fluorescence distribution maps in dynamic contexts. This makes it difficult for the control system to obtain accurate nodal physical coordinate sequences, thus hindering the implementation of globally optimal variable-length segmentation planning and making it difficult to truly achieve high-output, low-damage intelligent operations.

[0004] Therefore, an optimized automated control scheme for the assembly line operation of sweet potato seedling cultivation is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an automated control method and equipment for a production line operation of sweet potato seedling cultivation.

[0006] According to one aspect of this application, an automated control method for a production line operation in sweet potato seedling cultivation is provided, comprising:

[0007] Acquire a first image and a second image of the physical entity of sweet potato vines. The first image is the original image containing ambient light and excitation fluorescence, and the second image is the background image containing only ambient light.

[0008] Differential denoising, region of interest extraction, and morphological optimization were performed on the first and second images to obtain the vine fluorescence distribution map;

[0009] Based on the transmission belt speed, the node coordinates of the vine fluorescence distribution map are located based on the physiological signal gradient to obtain the vine node position sequence;

[0010] Based on the set of agronomic parameters, the optimal segmentation planning under variable internode length constraints is performed on the vine node position sequence to obtain the set of physical coordinates of the target cutting point;

[0011] Dynamic tracking control based on vision-encoder fusion is performed on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system.

[0012] According to another aspect of this application, an automated control device for a production line operation of sweet potato seedling cultivation is provided, comprising:

[0013] The image acquisition module is used to acquire a first image and a second image of the sweet potato vine physical entity. The first image is the original image containing ambient light and excitation fluorescence, and the second image is the background image containing only ambient light.

[0014] The vine fluorescence distribution map acquisition module is used to perform differential denoising, region of interest extraction, and morphological optimization on the first and second images to obtain the vine fluorescence distribution map.

[0015] The node coordinate localization module is used to locate the node coordinates of the vine fluorescence distribution map based on the physiological signal gradient, based on the transmission belt speed, so as to obtain the vine node position sequence.

[0016] The optimal segmentation planning module is used to perform optimal segmentation planning on the vine node position sequence under variable internode length constraints based on the set of agronomic parameters to obtain the set of physical coordinates of the target cutting point.

[0017] The dynamic tracking control module is used to perform vision-encoder fusion-based dynamic tracking control on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system.

[0018] Compared with existing technologies, this solution effectively solves the problems of difficult node identification in sweet potato vines and wasted cutting due to differences in internode spacing by constructing a closed-loop control system of physiological perception, global planning, and dynamic execution. Multispectral imaging technology is used to acquire excited-state and background-state images separately. Differential denoising and morphological optimization are used to eliminate ambient light interference, accurately extracting fluorescent physiological signals representing stem node positions from the complex visual background. Based on this high-fidelity signal, combined with agronomic standards, a dynamic programming algorithm is used to calculate the optimal segmentation of variable-length internodes on the entire vine, generating a cutting strategy that maximizes seedling yield. Finally, vision-encoder fusion technology is used to map the virtual planning coordinates into servo motion commands in real time, controlling the mechanical device to precisely and synchronously track and cut the moving vines on the production line, achieving high-precision, high-yield automated operation under non-standard biological characteristics. 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 the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 This is a flowchart of an automated control method for a sweet potato seedling production line according to an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of data flow in an automated control method for a sweet potato seedling production line according to an embodiment of this application.

[0022] Figure 3 This is a flowchart illustrating the automatic control method for a production line operation of sweet potato seedling cultivation according to an embodiment of this application, which involves differential denoising, region of interest extraction, and morphological optimization of a first image and a second image to obtain a vine fluorescence distribution map.

[0023] Figure 4 This is a flowchart illustrating the process of differential denoising of a first image and a second image to obtain a fluorescence probability distribution map in an automated control method for a sweet potato seedling production line according to an embodiment of this application.

[0024] Figure 5 This is a flowchart illustrating the automatic control method for a production line operation of sweet potato seedling cultivation based on the transmission belt speed, which involves locating the node coordinates of the vine fluorescence distribution map based on the physiological signal gradient to obtain the vine node position sequence.

[0025] Figure 6This is a flowchart illustrating the optimal segmentation planning of the vine node position sequence under variable internode constraints to obtain the physical coordinate set of the target cutting point, based on the set of agronomic parameters in the automatic control method for assembly line operation of sweet potato seedling cultivation according to the embodiments of this application.

[0026] Figure 7 This is a block diagram of an automated control device for sweet potato seedling production line operation according to an embodiment of this application. Detailed Implementation

[0027] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0028] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0029] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0030] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0031] Existing sweet potato seedling production lines mainly rely on mechanical fixed-length cutting or simple photoelectric triggering logic. This rigid control method cannot adapt to the random variation in the internode length of sweet potato vines due to differences in the growth environment. This can easily lead to the seedlings being scrapped because they do not contain effective buds. Furthermore, conventional visual methods are unable to accurately identify the weak stem node positions under complex backgrounds and petiole obstruction. Therefore, the technical solution of this application proposes an automatic control method for assembly line operations in sweet potato seedling cultivation. This method first constructs a physiological feature perception mechanism based on multispectral imaging. By acquiring dual-frame images of excited and background states, differential denoising and morphological optimization are used to eliminate ambient light interference. The fluorescence distribution map characterizing the activity of stem nodes is extracted with high fidelity from the high-noise background, and the node coordinates are accurately located based on the physiological signal gradient. On this basis, the traditional fixed-length cutting logic is abandoned, and an optimal segmentation planning strategy under the constraint of variable-length internodes is introduced. A qualified seedling evaluation model is constructed based on agronomic parameters, and a dynamic programming algorithm is used to solve the cutting path that maximizes output on the whole vine. Finally, through dynamic tracking control with vision-encoder fusion, the optimal coordinates of the virtual planning are transformed into real-time instructions of the multi-axis servo system, realizing precise follow-up cutting of non-standard biological entities on the conveyor belt, thereby significantly improving the utilization rate of raw materials while ensuring the agronomic compliance of seedlings.

[0032] Figure 1 This is a flowchart of an automated control method for a sweet potato seedling production line according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in an automated control method for a sweet potato seedling production line according to an embodiment of this application. Figure 1 and Figure 2 As shown, the automatic control method for sweet potato seedling production line operation according to an embodiment of this application includes the following steps: S100, acquiring a first image and a second image of the physical entity of sweet potato vines, wherein the first image is an original image containing ambient light and excitation fluorescence, and the second image is a background image containing only ambient light; S200, performing differential denoising, region of interest extraction, and morphological optimization on the first and second images to obtain a vine fluorescence distribution map; S300, based on the conveyor belt speed, performing node coordinate positioning based on physiological signal gradient on the vine fluorescence distribution map to obtain a vine node position sequence; S400, based on a set of agronomic parameters, performing optimal segmentation planning under variable internode constraints on the vine node position sequence to obtain a set of physical coordinates of target cutting points; S500, performing dynamic tracking control based on vision-encoder fusion on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of target cutting points to obtain a position and speed command stream sent to the multi-axis servo system.

[0033] Specifically, in step S100, a first image and a second image of the sweet potato vine physical entity are acquired. The first image is the original image containing ambient light and excitation fluorescence, and the second image is a background image containing only ambient light. It is understood that, due to uncontrollable diffuse reflection of natural light and interference from artificial lighting at the sweet potato seedling production line, and the relatively weak intensity of the biofluorescence signal generated by the stimulated sweet potato stem nodes compared to the ambient background light, if a single-frame imaging method is directly used, background noise often masks key physiological node features, making it difficult to segment the cutting target in subsequent processing stages. Therefore, in the technical solution of this application, a time-modulated dual-frame differential acquisition strategy is adopted to acquire a first image containing excitation fluorescence and ambient light, and a second image containing only ambient light, within a very short time window, thereby constructing real-time reference data with ambient light noise and mixed data containing the target signal. This allows the spatial correlation of the two images in the background light component to provide a data basis for eliminating environmental interference and preserving bud fluorescence signals through algebraic operations, ensuring the accuracy of physiological feature perception in non-dark room environments.

[0034] More specifically, in a particular example of this application, a high-frequency pulse timing-coordinated control of the imaging sensor and a specific wavelength excitation light source is used to perform the data acquisition task. When the sweet potato vine moves to the visual detection area along the conveyor belt, the ultraviolet excitation light source is first turned on and maintained with a stable output. UVA light with a center wavelength of 365nm is used to irradiate the vine surface to induce fluorescence in the chlorophyll and meristematic tissue at the stem nodes. Simultaneously, the imaging sensor performs the first exposure operation, converting the mixed light signal received within the field of view, including ambient background reflected light and biological stimulated fluorescence, into a digital signal to generate the first image. Then, within millisecond-level time intervals, the ultraviolet excitation light source is turned off to block the fluorescence excitation path, while maintaining the external ambient lighting conditions. Simultaneously, the imaging sensor performs the second exposure operation, acquiring the light signal within the field of view consisting only of ambient background reflected light to generate the second image, thus completing a full two-frame data acquisition process.

[0035] Specifically, in step S200, differential denoising, region of interest extraction, and morphological optimization are performed on the first and second images to obtain a vine fluorescence distribution map. It is understood that photoelectric imaging sensors inevitably introduce photon shot noise and readout noise during the acquisition process, and simple algebraic subtraction operations lead to the variance superposition of random noise, causing the signal-to-noise ratio of the original differential image to deteriorate significantly while removing background light. Furthermore, environmental vibrations or leaf micro-movements may introduce edge artifacts, resulting in a large number of discrete noise points and discontinuous signal holes in the original signal. Therefore, in the technical solution of this application, differential denoising, region of interest extraction, and morphological optimization are further performed on the first and second images to suppress the noise amplification effect in high background light areas, while using morphological operations to filter out isolated salt-and-pepper noise, fill in holes within the effective signal, and smooth edge spikes, thereby reconstructing a continuous and complete biological feature region. In this way, a vine fluorescence distribution map with high signal-to-noise ratio and good geometric morphology fidelity can be generated, providing a clean data foundation for subsequent accurate positioning of node coordinates based on physiological signal gradients, and avoiding false detection or missed detection of buds due to noise interference.

[0036] Figure 3 This is a flowchart illustrating the process of performing differential denoising, region of interest extraction, and morphological optimization on a first and second image to obtain a vine fluorescence distribution map, according to an embodiment of the automatic control method for a sweet potato seedling production line according to this application. Figure 3 As shown, step S200 includes: S210, performing differential denoising and region of interest extraction on the first image and the second image to obtain a fluorescence probability distribution map; S220, performing morphological optimization on the fluorescence probability distribution map to obtain the vine fluorescence distribution map.

[0037] In step S210, differential denoising and region of interest extraction are performed on the first and second images to obtain a fluorescence probability distribution map. It is understood that photoelectric imaging systems (such as CMOS or CCD sensors) inevitably introduce photon shot noise and readout noise during the acquisition process. Furthermore, conventional direct pixel subtraction operations are based on the idealized assumption of a noise-free background image, ignoring the statistical independence of noise in the spatiotemporal dimensions. According to the principle of variance summation, simple subtraction can lead to a near doubling of the noise variance in the final image. Especially in areas with strong ambient light, photon shot noise increases significantly, causing the weak biological fluorescence signal to be overwhelmed by a drastically deteriorated signal-to-noise ratio. Therefore, in the technical solution of this application, differential denoising and region of interest extraction are further performed on the first and second images to obtain a fluorescence probability distribution map. This establishes an optimal computational mechanism that dynamically adjusts the differential weights based on local background intensity. Nonlinear suppression relationships are used to suppress the noise amplification effect in bright background areas. Combined with a nonlocal mean algorithm, the spatial structural self-similarity of biological signals is used for texture reconstruction, transforming deterministic binary segmentation into a confidence-based soft decision mapping. In this way, the traditional differential algorithm can be fundamentally overcome to misjudge the characteristics of noise. While effectively filtering out random noise, it can restore the edge contour and internal texture of the vine stem nodes with high fidelity, generating a probabilistic digital map that can accurately reflect the distribution of physiological activity, providing robust data support for subsequent precise positioning.

[0038] Figure 4 This is a flowchart illustrating the process of differential denoising of a first image and a second image to obtain a fluorescence probability distribution map in an automated control method for sweet potato seedling production line operations according to an embodiment of this application. Figure 4 As shown, step S210 includes: S211, performing spatial adaptive weighted difference on the first image and the second image to obtain the original fluorescence signal map; S212, performing signal regularization enhancement based on nonlocal mean on the original fluorescence signal map to obtain the enhanced fluorescence signal map; S213, performing region of interest probabilization on the enhanced fluorescence signal map to obtain the fluorescence probability distribution map.

[0039] In step S211, spatial adaptive weighted difference is performed on the first image and the second image to obtain the original fluorescence signal image. It is understood that any photoelectric imaging system, especially CMOS or CCD sensors, introduces random noise during the acquisition process, mainly including photon shot noise and readout noise. According to noise theory, the variance of the difference between two independent random variables is equal to the sum of the variances of the two variables. Therefore, simple image subtraction not only fails to effectively eliminate noise but also causes the noise variance of the final difference image to nearly double, significantly deteriorating the signal-to-noise ratio, especially in areas with strong ambient light. Therefore, in the technical solution of this application, spatial adaptive weighted difference is further performed on the first image and the second image to obtain the original fluorescence signal image, thereby introducing a signal reliability weight. This weight is dynamically determined by the local pixel intensity of the background image, thus using an exponentially decaying nonlinear weight to replace simple algebraic subtraction to estimate the truly latent pure fluorescence signal in the excited state image. In this way, by using an exponential decay function to construct this nonlinear suppression relationship, the noise amplification effect in the high background light region is effectively suppressed, while the real fluorescence signal with high signal-to-noise ratio in the dark background region is completely preserved, thus obtaining a preliminary purified original fluorescence signal map.

[0040] More specifically, in a particular example of this application, the processing unit performs a pixel-by-pixel nonlinear weighted calculation on the acquired image data based on the physical characteristic that photon shot noise increases with increasing light intensity. This calculation process can be represented as follows:

[0041]

[0042] in, Pixel coordinates in the original fluorescence signal image representing the output The intensity value at that location, The intensity value at that point is the excited-state image, i.e., the first image. This represents the intensity value of the background image, i.e., the second image, at that point. It is a gain correction factor used to correct for small systematic gain differences that may exist between two exposures. It is a natural exponential function. As a calibrable attenuation coefficient, it is used to control the attenuation rate of the weights as the background intensity changes. In a specific scenario: if a local area of ​​the production line is directly exposed to strong external light (such as a light spot projected from a skylight), resulting in bright background illumination at that location, the greater the corresponding photon shot noise, the lower the reliability of the difference result. In this case, the algorithm uses an exponential decay function to make the weighting coefficient of the bright area approach zero, thereby effectively suppressing the noise amplification effect in the high background light area. Conversely, in the shadow area formed by the mutual occlusion of the blades, the background image intensity value is low, and the calculated weight remains high, while the high signal-to-noise ratio real fluorescence signal in the dark background area is completely preserved. Thus, a dynamic adjustment calculation strategy for areas with different lighting conditions is realized in the same image.

[0043] In step S212, the original fluorescence signal image is enhanced by signal regularization based on nonlocal means to obtain an enhanced fluorescence signal image. It is understood that although some noise is suppressed in the original fluorescence signal image output in the first step, residual random noise remains. Traditional morphological filtering and other methods, while denoising, often blur or distort the edges of the signal, destroying its true geometric shape. Therefore, in the technical solution of this application, the original fluorescence signal image is further enhanced by signal regularization based on nonlocal means. Local operators are abandoned in favor of a nonlocal means algorithm to fully utilize the prior knowledge of the inherent structural self-similarity of biological signals in space. This allows for the filtering of random noise to a great extent while maintaining the sharp edge contours and fine internal texture of the fluorescence spots with high fidelity, providing high-quality input data for subsequent precise localization.

[0044] More specifically, in a concrete example of this application, the processing unit executes a full-image similarity search and weighted reconstruction strategy. This strategy is based on the physiological structural characteristics of sweet potato vines, namely, the morphology and texture of a real fluorescent spot on a sweet potato stem node are likely to be repeated in other locations in the image (e.g., at another adjacent stem node). Based on this, the algorithm reconstructs the true signal value of the current pixel by searching all image patches similar to the neighborhood of the current pixel across the entire image and taking a weighted average of the center pixel values ​​of these similar patches. This process can be represented as follows:

[0045]

[0046]

[0047] in, The pixel represents the enhanced pixel. The output intensity value is q, which iterates through all pixels in the image. It is a pixel and Similarity weights between them and They are in pixels and The central image patch (neighborhood). This represents the square of the Gaussian-weighted Euclidean distance between two neighborhoods, used to measure their similarity. As a filter strength parameter, it controls the Gaussian decay rate of the weighting function. This is the normalization factor that ensures the total weights are equal to 1. In a real-world sweet potato seedling production line scenario, if a stem node on the conveyor belt exhibits mottled or broken fluorescence signals due to minor surface stains or localized sensor thermal noise, the algorithm can automatically use information from other clear stem nodes (i.e., areas with high texture similarity) on the same vine to repair the damaged node. This eliminates unstructured random noise without blurring the bud edges, ensuring that subsequent steps can accurately identify the complete node morphology.

[0048] In step S213, the enhanced fluorescence signal image is probabilistically transformed into a region of interest (ROI) to obtain a fluorescence probability distribution map. It is understood that traditional hard-threshold segmentation methods are too rigid and can lead to information loss or artifacts due to improper threshold selection; this is the reason for performing this step. If the enhanced signal image is simply binarized, weak bud signals at critical intensities are easily misclassified as background, or strong background noise is misclassified as targets. Therefore, in the technical solution of this application, the enhanced fluorescence signal image is further probabilistically transformed into a region of interest. Instead of performing binary processing of either 0 or 1, an adaptive sigmoid function is introduced to smoothly map the enhanced fluorescence signal intensity to a probability value between 0 and 1. That is, a deterministic segmentation problem is transformed into a probability estimation problem based on uncertainty. This generates a fluorescence probability distribution map, where the value of each pixel directly represents the confidence level that the point belongs to the true fluorescence region. This soft decision graph provides downstream node localization algorithms with far richer decision-making criteria than binary images, allowing subsequent processing units to adopt different processing strategies based on probability values, thereby improving the robustness and intelligence of the entire system and realizing an intelligent upgrade from fragile hard decision-making to robust soft decision-making.

[0049] More specifically, in a particular example of this application, the processing unit uses a nonlinear activation function to perform soft mapping on the signal intensity space, constructing a confidence map reflecting the probability of the existence of sweet potato stem nodes. This mapping process can be represented as:

[0050]

[0051] in, For the pixels in the final output fluorescence probability map The probability value at that location. This represents the signal strength after the previous step. It is a segmentation bias benchmark dynamically calculated using adaptive algorithms such as Otsu, which enhances the method's adaptability to different batches and vines in different growth stages. This controls the sharpness of the probability transition curve. In actual assembly line operations, different batches of sweet potato vines may exhibit different fluorescence baseline intensities due to differences in water content or chlorophyll concentration (e.g., older vines show strong fluorescence, while younger shoots show weak fluorescence). The dynamic adjustment can automatically adapt to these biological differences, ensuring that the probability value of the core region of both strong and weak fluorescent targets approaches 1, while the probability decays smoothly in the edge transition region, thus preserving complete geometric topology information for subsequent path planning.

[0052] In step S220, the fluorescence probability distribution map is morphologically optimized to obtain the vine fluorescence distribution map. It is understood that due to inherent discrete noise interference from the imaging sensor and the physical characteristics of the sweet potato vine surface, such as petiole occlusion or uneven chlorophyll distribution, the generated fluorescence probability distribution map often still contains isolated artifact noise points, or breaks and voids within the actual stem node signal region. These unstructured defects directly affect the accuracy of subsequent node topology analysis. Therefore, in the technical solution of this application, the fluorescence probability distribution map is further morphologically optimized to obtain the vine fluorescence distribution map. This filters out discrete noise that does not possess biological structural characteristics based on the principle of geometric connectivity, and spatially bridges and fills the broken effective signals. This outputs a vine fluorescence distribution map with a clean background and continuous, complete target features, providing a high signal-to-noise ratio morphological basis for subsequent node localization based on one-dimensional projection.

[0053] More specifically, in a specific example of this application, the processing unit invokes a preset morphological operation operator to perform cascaded spatial filtering on the probabilistic image. First, a morphological opening operation is performed on the image using a structuring element that matches the minimum physical size of the sweet potato axillary bud. This process peels away and removes isolated bright spots smaller than the structuring element through erosion, thereby eliminating false positive signals caused by diffuse reflection from air dust or sensor dead pixels. Next, a morphological closing operation is performed on the image using the structuring element. This process expands the boundary of the high-probability region outward through dilation, fuses and connects adjacent signal blocks that are broken due to minor occlusion, and fills low-probability voids inside the region caused by uneven excitation. Then, the region boundary is shrunk back to its original scale through erosion, thus completing the integrity restoration of biological features while maintaining the geometric contour of the bud.

[0054] Specifically, in step S300, based on the conveyor belt speed, the vine fluorescence distribution map is used to locate the node coordinates based on the physiological signal gradient to obtain the vine node position sequence. It is understood that because the petioles and stems of sweet potato vines highly overlap in space, traditional visual recognition methods based on geometric contours struggle to accurately locate hidden axillary bud nodes through occlusion. Furthermore, the visual system acquires discrete instantaneous pixel coordinates, which cannot directly guide downstream mechanical actuators to perform precise operations in a continuously moving physical space. Therefore, in the technical solution of this application, the vine fluorescence distribution map is further located based on the conveyor belt speed using the physiological signal gradient to obtain the vine node position sequence. This utilizes the local maxima characteristic of fluorescence intensity exhibited by vigorous biological metabolism at the stem nodes, and uses mathematical extreme value search of the signal gradient to lock the physiological node center. Combined with the kinematic parameters of the conveyor belt, the pixel index in the image domain is mapped to distance coordinates in the physical domain. This constructs an ordered dataset containing the precise physical locations of all effective buds on the vine, providing an accurate geometric reference for subsequent globally optimal cutting planning under variable internode length constraints.

[0055] Figure 5 This is a flowchart illustrating the automatic control method for a sweet potato seedling production line based on conveyor belt speed, which uses physiological signal gradients to locate the node coordinates of a vine fluorescence distribution map to obtain a sequence of vine node positions. (See the flowchart for an example.) Figure 5 As shown, step S300 includes: S310, performing longitudinal integral projection on the vine fluorescence distribution map to obtain a one-dimensional fluorescence intensity signal waveform; S320, performing feature peak search based on second-order derivative on the one-dimensional fluorescence intensity signal waveform to obtain a set of node positions in the image coordinate system; S330, performing multi-dimensional parameter fusion physical coordinate mapping on the set of node positions in the image coordinate system based on the transmission belt speed and the calibration coefficient of the vision system to obtain a sequence of vine node positions.

[0056] In step S310, the fluorescence distribution map of the vines is subjected to longitudinal integral projection to obtain a one-dimensional fluorescence intensity signal waveform. It is understandable that although the posture of the sweet potato vines on the conveyor belt is generally distributed along the direction of movement, random displacement and bending inevitably occur in the lateral width. Furthermore, the fluorescence intensity of a single pixel is easily affected by discrete noise from the sensor, causing fluctuations. Directly performing feature search in the two-dimensional image space would not only result in a huge computational load but also make it difficult to establish a unified geometric criterion to stably lock the centroid of the nodes. Therefore, in the technical solution of this application, the fluorescence distribution map of the vines is further subjected to longitudinal integral projection to obtain a one-dimensional fluorescence intensity signal waveform. This mathematically reduces the dimensionality of the two-dimensional image pixel matrix along the direction perpendicular to the conveyor belt movement, utilizing the integral accumulation effect to aggregate the fluorescence energy of the stem node region and smooth the randomly distributed background noise. In this way, the complex two-dimensional morphological recognition problem can be transformed into a one-dimensional signal peak detection problem with higher computational efficiency and stronger anti-interference capability, improving the real-time response speed and robustness of node positioning in high-speed assembly line operations.

[0057] More specifically, in a concrete example of this application, the processing unit first determines the mapping relationship between the image coordinate system and the physical conveyor belt coordinate system based on the calibration parameters of the vision system. The horizontal pixel coordinate axis of the image is defined as the principal axis along the direction of conveyor belt movement, and the vertical pixel coordinate axis is defined as the projection axis perpendicular to the direction of conveyor belt movement. Subsequently, the algorithm traverses the entire vine fluorescence distribution map pixel by pixel along the principal axis direction. At each principal axis coordinate point, it performs an accumulation operation on all grayscale values ​​of the corresponding column of pixels along the projection axis direction to obtain the total fluorescence intensity value of the cross-section at that location. Next, a Gaussian smoothing filter is applied to the original discrete sequence composed of the accumulated values ​​of all principal axis coordinate points. High-frequency spikes caused by imaging quantization errors or minor impurities are eliminated through convolution operations, ultimately generating a smooth and continuous one-dimensional fluorescence intensity signal waveform. The local maxima in this waveform represent the physical centers of the stem nodes with the strongest biological activity on the vine.

[0058] In step S320, a feature peak search based on second-order derivative is performed on the one-dimensional fluorescence intensity signal waveform to obtain a set of node positions in the image coordinate system. It is understood that due to differences in the physiological activity of individual sweet potato vines, the absolute fluorescence intensity values ​​produced by different stem nodes fluctuate significantly with changes in plant water content and growth cycle. If only a single intensity threshold is used for judgment, it is easy to miss weak fluorescence nodes or falsely detect strong fluorescence internode regions due to improper threshold setting. Therefore, in the technical solution of this application, a feature peak search based on second-order derivative is further performed on the one-dimensional fluorescence intensity signal waveform to obtain a set of node positions in the image coordinate system. This utilizes the geometric property that the first derivative is zero and the second derivative is negative at local maxima points to accurately locate the peak position representing the bud center while ignoring signal baseline drift. This ensures that the node localization algorithm has a high degree of adaptability to changes in signal amplitude, thereby outputting a set of accurate node coordinates in the image coordinate system that is unaffected by uneven illumination.

[0059] More specifically, in a concrete example of this application, the processing unit treats the discrete one-dimensional fluorescence intensity signal waveform as a continuously differentiable function, approximates its first and second derivatives at each pixel displacement point using difference operations, and filters points across the entire domain that simultaneously satisfy three necessary conditions: gradient crossing zero, downward convexity of the curve, and intensity higher than the floor noise, as candidate nodes. This feature peak search logic based on second-order derivatives is executed according to the following formula:

[0060]

[0061] in, The one-dimensional fluorescence intensity signal waveform generated in the preceding steps characterizes the distribution of bioactivity along the vine length. This is the set of node positions in the final output image coordinate system, i.e., a list of all identified bud pixel coordinates. This represents the displacement variable along the direction of the conveyor belt movement, corresponding to the column coordinate index of the image. This is the first derivative of the one-dimensional fluorescence intensity signal waveform, used to detect the rate of change of the signal and the location of extreme points. The second derivative of the one-dimensional fluorescence intensity signal waveform is used to determine the concavity / convexity of extreme points to distinguish between peaks and troughs. The preset fluorescence intensity noise threshold is used to filter out weak clutter from non-biological features in the background. The AND operator indicates that all of the above conditions must be met simultaneously. In a specific scenario, suppose a sweet potato vine on a conveyor belt contains both a section of brightly fluorescent new shoot and a section of weakly fluorescent aging stem. Traditional thresholding methods are prone to misclassifying the bright internodes of the new shoot as nodes, or missing aging, dark nodes due to excessively high thresholds. However, the second-order differential search mechanism described in this embodiment focuses on the topological shape of the signal waveform rather than absolute brightness. Therefore, even weak fluorescent bulges at aging stem nodes, as long as they form a local peak structure, can be accurately captured and their coordinates included. The set, while the internodes of the new shoots are bright but have flat waveforms (the second derivative does not meet the condition), will be correctly excluded, thus achieving a highly robust localization of the nodes of the entire vine.

[0062] In step S330, based on the conveyor belt speed and the calibration coefficient of the vision system, a multi-dimensional parameter fusion physical coordinate mapping is performed on the set of node positions in the image coordinate system to obtain the vine node position sequence. It is understandable that since the node positions output by the visual recognition algorithm are initially only represented as discrete pixel indices in a virtual image coordinate system, this data lacks a metrical meaning corresponding to real physical space. Furthermore, the sweet potato vines are in a continuous dynamic transmission process on the assembly line, and static pixel coordinates cannot be directly used as input commands for the servo control system to guide the cutting tool to accurately position itself in three-dimensional space. Therefore, in the technical solution of this application, a multi-dimensional parameter fusion physical coordinate mapping is further performed on the set of node positions in the image coordinate system based on the conveyor belt speed and the calibration coefficient of the vision system to obtain the vine node position sequence. This transforms abstract image feature points into metric length information with clear physical dimensions and eliminates spatial deviations caused by temporal differences between image acquisition and calculation times or by the stitching process. In this way, a precise and ordered set of physical coordinates of all effective buds on the vine relative to the root reference point can be constructed, providing rigorous geometric data support for subsequent intelligent segmentation planning that conforms to agronomic standards over the entire length.

[0063] More specifically, in a concrete example of this application, the processing unit first establishes a linear or nonlinear conversion model between image pixel distance and actual physical length based on the calibration coefficients determined by the installation height of the visual imaging system and the lens parameters. This model defines the millimeter value represented by a unit pixel. Next, the algorithm identifies and locks the morphological root endpoint of the sweet potato vine in the image coordinate system as the zero-point reference for physical measurement, and calculates the pixel difference between each candidate node in the image coordinate system relative to this zero-point reference. Simultaneously, the processing unit introduces the conveyor belt speed parameter as a dynamic correction factor. For scenarios using linear scan or multi-frame stitching imaging, it calculates the physical displacement between different acquisition rows or frames using speed integration, and compensates for this displacement in the aforementioned pixel difference conversion calculation, thereby performing multi-dimensional parameter fusion operations. Finally, the calculated physical distances of each node are sorted and indexed according to the spatial order from root to tip, outputting a node position sequence that accurately reflects the current vine biological structure.

[0064] Specifically, in step S400, based on the set of agronomic parameters, optimal segmentation planning under variable-length internode constraints is performed on the vine node position sequence to obtain the set of physical coordinates of the target cutting points. It is understandable that, due to the significant random variable length characteristics of sweet potato vine internodes influenced by the growth environment and varietal genes, and the strict dual constraints of agronomic standards on the length range and effective number of nodes per seedling, traditional mechanical fixed-length cutting or local greedy strategies are prone to result in substandard seedling segments (such as blind nodes) or waste of the entire vine's remaining material due to internode fluctuations. Therefore, in the technical solution of this application, optimal segmentation planning under variable-length internode constraints is further performed on the vine node position sequence based on the set of agronomic parameters to obtain the set of physical coordinates of the target cutting points. This transforms the simple cutting action into a constrained global combinatorial optimization problem. Based on a preset qualified seedling evaluation model, a cutting strategy is calculated over the entire length range that strictly meets agronomic requirements such as three nodes per seedling while maximizing the number of qualified seedlings produced by the entire vine. This enables a fundamental shift from rigid, fixed-length processing to flexible, intelligent decision-making, effectively overcoming the processing challenges posed by the non-standard nature of biological materials. It ensures that each finished seedling has the potential for a high survival rate while improving raw material utilization and production economic efficiency.

[0065] Figure 6 This is a flowchart illustrating the process of obtaining the set of physical coordinates of target cutting points by performing optimal segmentation planning under variable-length internode constraints on the vine node position sequence based on a set of agronomic parameters in an automated control method for sweet potato seedling production line operations according to embodiments of this application. Figure 6As shown, step S400 includes: S410, based on the set of agronomic parameters, performing Boolean evaluation modeling of qualified seedling segments on the vine node position sequence to obtain the qualified seedling segment evaluation function; S420, based on the qualified seedling segment evaluation function, solving the maximum output path based on dynamic programming for the total vine length to obtain the dynamic programming state table and the optimal path backtracking table; S430, performing optimal solution backtracking and cutting point sequence generation on the dynamic programming state table, the optimal path backtracking table, and the total vine length to obtain the set of physical coordinates of the target cutting point.

[0066] In step S410, based on the set of agronomic parameters, a Boolean evaluation model of qualified seedling segments is performed on the vine node position sequence to obtain a qualified seedling segment evaluation function. It is understandable that sweet potato seedlings face strict dual constraints on physical morphology and biological characteristics during cutting propagation: the stem segment must reach a specific length to ensure the cutting depth, and it must contain a sufficient number of nodes (buds) to guarantee rooting and germination rates. Simple fixed-length cutting cannot guarantee that every segment cut simultaneously meets these two combined conditions. Therefore, in the technical solution of this application, a Boolean evaluation model of qualified seedling segments is further performed on the vine node position sequence based on the set of agronomic parameters to obtain a qualified seedling segment evaluation function. This encapsulates the complex agronomic selection criteria into a computable mathematical logic operator. This provides a rigorous logical basis for subsequent global optimization algorithms, ensuring that any candidate cutting scheme output by the algorithm strictly conforms to the survival standards for agricultural planting from a biological perspective.

[0067] More specifically, in a concrete example of this application, the construction of the evaluation logic includes two stages: parameter loading and geometric verification. First, the standard numerical range of agronomic requirements is loaded into the calculation unit. These parameters include the minimum lower limit of the length of a qualified seedling, the maximum upper limit of the length, and the minimum node threshold required for a single seedling. Then, a logical judgment function is defined to receive arbitrary starting and ending cut coordinates on the vine as input variables. In the execution logic of this function, the physical distance difference between the starting and ending coordinates is calculated to determine whether it falls within a closed interval formed by the minimum and maximum lengths. Simultaneously, based on the vine node position sequence obtained in the previous steps, the number of valid node coordinates falling within this spatial interval is traversed and counted. The function outputs a qualified value only if the physical length of the segment is within the allowable range and the number of nodes contained is greater than or equal to the preset node threshold; otherwise, it outputs a unqualified value, thereby achieving the digital translation of agricultural standards.

[0068] In step S420, based on the qualified seedling segment evaluation function, the maximum output path of the total vine length is solved using dynamic programming to obtain a dynamic programming state table and an optimal path backtracking table. It is understandable that due to the non-standard biological structure of sweet potato vines, simple sequential cutting decisions often fall into local optima traps. That is, in order to meet the agronomic requirements of the current segment, excessively long vine resources are consumed prematurely, resulting in the remaining material, although physically sufficient, failing to form qualified seedlings due to uneven node distribution, causing a hidden loss in seedling yield. Therefore, in the technical solution of this application, the maximum output path of the total vine length is further solved using dynamic programming based on the qualified seedling segment evaluation function to obtain a dynamic programming state table and an optimal path backtracking table. This utilizes the global optimization capability of the dynamic programming algorithm, treating the entire vine as a finite resource space, traversing all possible cutting combinations to find the specific segmentation path that can produce the most qualified seedlings. This ensures that the placement of each cut not only meets current agronomical compliance requirements but also reserves optimal resource allocation for subsequent cutting, thereby maximizing overall yield.

[0069] More specifically, in a concrete example of this application, the computing unit first allocates two linear array spaces in memory corresponding to the total physical length of the vine: a dynamic programming state table for recording the maximum seedling yield at the current position and an optimal path backtracking table for recording the optimal predecessor decision position. The initial values ​​of the state tables are all set to zero. Subsequently, the algorithm performs bottom-up recursive calculations, traversing every possible cut position from the root to the tip of the vine in millimeter-level steps. When calculating the state value at each current position, the algorithm not only inherits the state value of the previous position to represent the strategy of not cutting at this location, but more importantly, it backtracks to search for all possible previous cut positions and calls the qualified seedling segment evaluation function constructed in the preceding steps to verify the compliance of the virtual stem segment formed by these two points. Once the verification passes, it indicates that a potential qualified seedling has been found. The algorithm compares the total number of seedlings obtainable by cutting at this location with the currently recorded maximum number of seedlings. If the former is greater, the state value of the current position is updated, and the index of the previous cut position is written into the path backtracking table. This state transition process is executed according to the following formula:

[0070]

[0071] in, This represents the state value in dynamic programming, and its physical meaning is the distance from the root of the vine to a physical length of... The maximum number of qualified seedlings that can be planned is the core indicator for measuring the effectiveness of the plan. This is a function for maximizing output, used to select the optimal solution with the highest yield from multiple possible cutting schemes. and These are the current discrete position index and the preorder traversal index along the length of the vine, respectively. For a qualified seedling segment evaluation function, if and only if the position arrive The state transition returns true when the vine segments simultaneously satisfy both the length and node number constraints, constituting the necessary boundary conditions. If the representative is in position and If a qualified seedling is successfully cut from the current location, the current total output will equal the position. The optimal output is increased by 1. To illustrate this with a scenario: Suppose a sweet potato vine is 100 cm long, with a non-uniform node distribution of dense-sparse-dense. If a greedy strategy is used for sequential cutting, after cutting three standard 25 cm sections, the remaining 25 cm section, located in a sparse area with only two nodes, might be considered waste. However, the dynamic programming algorithm in this embodiment, through global calculation, might automatically adjust the cutting points of the first three sections (for example, slightly extending the first two sections to 28 cm to cross the sparse area), so that the last section, although physically shorter, contains exactly three dense nodes, ultimately successfully planning four qualified seedlings, transforming the original waste into effective production capacity.

[0072] In step S430, the dynamic programming state table, optimal path backtracking table, and total vine length are used to generate an optimal solution backtracking and cutting point sequence to obtain the set of physical coordinates of the target cutting points. It is understandable that, since the dynamic programming algorithm generates a state matrix representing the maximum output and a path backtracking matrix recording decision nodes during the forward recursive calculation process, these data structures are essentially abstract mathematical indexes stored in the memory of the computing unit and do not possess the explicit spatial geometric attributes required for physical execution. Furthermore, the calculation direction is along the vine from the root to the tip, and determining the specific segmentation boundary requires reverse derivation based on the global optimal result at the terminal. Therefore, in the technical solution of this application, the dynamic programming state table, optimal path backtracking table, and total vine length are further used to generate an optimal solution backtracking and cutting point sequence to obtain the set of physical coordinates of the target cutting points. This reconstructs the exact physical location of each cutting action from the discrete decision chain and reverses the reverse logical deduction sequence into a forward spatial sequence compatible with the timing of the mechanical actuator. In this way, the abstract mathematical optimal solution can be transformed into a set of executable target cutting point physical coordinates that can directly drive the servo system, ensuring that the theoretical maximum output solution is accurately materialized into physical cutting operations.

[0073] More specifically, in a concrete example of this application, the execution logic initiates a reverse search process starting from the physical endpoint coordinates of the total vine length. The processing logic first accesses the value corresponding to the total length position in the dynamic programming state table to confirm the existence of a valid solution and sets the current search pointer to that total length value. Then, it enters a loop iteration state, using the current pointer value as an index to query the optimal path backtracking table, extracting the predecessor decision node coordinate index of that position record. This predecessor index represents the starting position of the last qualified seedling segment. After obtaining the predecessor coordinates, the current cutting termination coordinates are added to a temporary buffer, and the search pointer is updated to the predecessor coordinate value, thus jumping back to the previous decision state. This recursive backtracking process continues until the pointer value returns to zero, representing the zero point coordinates of the vine root, completing the extraction of all segment nodes. Finally, given that the physical cutting sequence on the pipeline must follow the vine's transmission direction, a reversal operation is performed on the coordinate sequence stored in the temporary buffer to correct the order, generating a strictly monotonically increasing set of physical coordinates of the target cutting points. This set defines the absolute position of all cutting events relative to the vine root reference.

[0074] Specifically, in step S500, dynamic tracking control based on vision-encoder fusion is performed on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system. It is understandable that due to the inevitable time lag in image acquisition, data transmission and complex algorithm processing, and the fact that the sweet potato vines are in a continuous high-speed motion with the conveyor belt, there is a dynamic deviation between the static physical coordinates calculated based on historical images and the actual physical position at the current moment. If cutting is performed directly based on static data, the cutter will not be able to accurately hit the already moved target node, and the flexible stem may even be pulled and broken or produce an uneven tear due to the relative speed difference between the cutter and the vine. Therefore, in the technical solution of this application, dynamic tracking control based on vision-encoder fusion is further applied to the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system. This establishes a spatiotemporal mapping relationship between the static visual space and the dynamic motion space. Using high-precision encoder pulses as a globally unified spatiotemporal reference, the virtual cutting point output by the planning layer is attached to the real-time conveyor belt motion stream, and an electronic cam motion curve is generated. In this way, the multi-axis servo system can be driven to perform highly dynamic response following motion, ensuring that the cutting end actuator achieves strict positional coincidence and zero relative velocity synchronization at the moment of contact with the vine, thereby completing non-destructive, smooth, and precise cutting in a non-stop assembly line operation mode.

[0075] More specifically, in a concrete example of this application, the motion control logic first performs a global pulse mapping in the spatiotemporal coordinate system. It reads the pulse value of the spindle encoder latched at the moment of image acquisition as a spatiotemporal anchor point, and combines this with the inherent physical offset between the center of the vision system and the mechanical cutting origin to transform each planned static physical distance coordinate into an absolute target pulse value in the conveyor belt coordinate system. This pulse mapping process is executed according to the following formula:

[0076]

[0077] in, For the first The target pulse value of the conveyor belt absolute encoder corresponding to each cutting point is the global spatiotemporal reference for triggering the tracking action. The encoder pulse value latched at the moment of image acquisition is used to eliminate temporal errors caused by image processing delays. The first step is planned based on the previous steps. The physical distance between each cutting point and the root of the vine This is a fixed physical distance between the visual imaging center and the center of the cutting actuator, used to compensate for spatial differences caused by the installation position. The encoder pulse equivalent coefficient (pulses / mm) is used to uniformly map the physical millimeter unit to the pulse unit in the control domain. Subsequently, the control logic establishes a master-slave following model for the electronic cam, setting the conveyor belt encoder as the master axis and the cutting servo axis as the slave axis. Based on the target pulse value, a fifth-order polynomial motion curve including acceleration, synchronization, and return segments is planned. When the real-time monitored conveyor belt pulse count enters the preset tracking window, the servo system activates synchronization segment control, dynamically adjusting the cutter's feed speed and position to keep it relatively stationary with the conveyor belt. To illustrate this in a scenario: Assume the conveyor belt is transporting sweet potato vines at a speed of 500 mm / s. The vision system takes a picture when the encoder reading is 10000. After 200 milliseconds of calculation, it plans a cut 300 mm from the vine's root, while the camera is 1000 mm from the cutter. At this point, the vine has already moved 100 mm forward; cutting according to static coordinates would result in an incomplete cut. In this embodiment, the absolute target pulse is calculated by formula. No matter how long the calculation takes, as long as the pulse value of the conveyor belt has not reached the target value, the servo system will start acceleration and cut at the same speed at the exact moment (that is, when the cutting point on the vine just reaches below the blade), ensuring accurate cutting after accurate calculation.

[0078] In summary, the automated control method for sweet potato seedling production line operations according to the embodiments of this application is explained. It effectively solves the problems of difficulty in identifying sweet potato vine nodes and cutting waste caused by differences in internode spacing by constructing a closed-loop control system of physiological perception-global planning-dynamic execution. Multispectral imaging technology is used to acquire excited-state and background-state images respectively. Differential denoising and morphological optimization are used to eliminate ambient light interference, accurately extracting the fluorescent physiological signals representing the stem node position from the complex visual background. Based on this high-fidelity signal, and further combined with agronomic standards, a dynamic programming algorithm is used to calculate the optimal segmentation of variable-length internodes on the entire vine, generating a cutting strategy that maximizes seedling yield. Finally, vision-encoder fusion technology is used to map the virtual planning coordinates into servo motion commands in real time, controlling the mechanical device to accurately and synchronously track and cut the moving vines on the production line, achieving high-precision, high-yield automated operations under non-standard biological characteristics.

[0079] Furthermore, an automated control device for a production line operation of sweet potato seedling cultivation is also provided.

[0080] Figure 7 This is a block diagram of an automated control device for a sweet potato seedling production line according to an embodiment of this application. Figure 7 As shown, the automated control device 100 for sweet potato seedling production line according to an embodiment of this application includes: an image acquisition module 110, used to acquire a first image and a second image of the physical entity of sweet potato vines, wherein the first image is an original image containing ambient light and excitation fluorescence, and the second image is a background image containing only ambient light; a vine fluorescence distribution map acquisition module 120, used to perform differential denoising, region of interest extraction, and morphological optimization on the first image and the second image to obtain a vine fluorescence distribution map; a node coordinate positioning module 130, used to perform node coordinate positioning based on physiological signal gradients on the vine fluorescence distribution map based on the conveyor belt speed to obtain a vine node position sequence; an optimal segmentation planning module 140, used to perform optimal segmentation planning under variable internode constraints on the vine node position sequence based on a set of agronomic parameters to obtain a set of physical coordinates of target cutting points; and a dynamic tracking control module 150, used to perform dynamic tracking control based on vision-encoder fusion on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of target cutting points to obtain a position and speed command stream sent to a multi-axis servo system.

[0081] As described above, the automated control device 100 for sweet potato seedling production line operation according to the embodiments of this application can be implemented in various types of computing devices or control units. For example, it can be a ruggedized industrial computer deployed in the main control cabinet of the seedling production line, an embedded vision controller installed at the vision acquisition station, or a high-performance PLC processing unit integrated into a motion control system. In one possible implementation, the automated control device 100 for sweet potato seedling production line operation according to the embodiments of this application can be integrated into the computing device as a software module and / or a hardware module. For example, the automated control device 100 for sweet potato seedling production line operation can be a resident data processing service in the operating system of the computing device. This software module is configured to perform differential denoising and morphological optimization of excited-state and background-state images, node coordinate localization based on physiological signal gradients, optimal segmentation planning of variable-length internodes combined with agronomic parameters, and dynamic tracking instruction generation based on vision-encoder fusion. Alternatively, it can be a dedicated intelligent cutting control algorithm program for sweet potato seedlings developed specifically for the computing device. Of course, the automated control equipment 100 for the assembly line operation of sweet potato seedling cultivation can also be one of the many hardware modules of the computing device or control unit, or it can be embedded in a field-programmable gate array circuit to accelerate image difference operations and one-dimensional signal projection analysis in parallel, or it can be a physiological fluorescence signal processing integrated circuit for a specific application.

[0082] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, and are not 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. An automated control method for a production line operation of sweet potato seedling cultivation, characterized in that, include: Acquire a first image and a second image of the physical entity of sweet potato vines. The first image is the original image containing ambient light and excitation fluorescence, and the second image is the background image containing only ambient light. Differential denoising, region of interest extraction, and morphological optimization were performed on the first and second images to obtain the vine fluorescence distribution map; Based on the transmission belt speed, the node coordinates of the vine fluorescence distribution map are located based on the physiological signal gradient to obtain the vine node position sequence; Based on the set of agronomic parameters, the optimal segmentation planning under variable internode length constraints is performed on the vine node position sequence to obtain the set of physical coordinates of the target cutting point; Dynamic tracking control based on vision-encoder fusion is performed on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system; Differential denoising, region of interest extraction, and morphological optimization are performed on the first and second images to obtain the vine fluorescence distribution map, including: Differential denoising and region of interest extraction are performed on the first and second images to obtain fluorescence probability distribution maps; Morphological optimization was performed on the fluorescence probability distribution map to obtain the fluorescence distribution map of the vine. Differential denoising is performed on the first and second images to obtain a fluorescence probability distribution map, including: Spatial adaptive weighted difference is performed on the first image and the second image to obtain the original fluorescence signal map; The original fluorescence signal map is enhanced by signal regularization based on nonlocal mean to obtain an enhanced fluorescence signal map; The region of interest is probabilized in the enhanced fluorescence signal map to obtain the fluorescence probability distribution map; To obtain the original fluorescence signal map, spatial adaptive weighted difference is performed on the first image and the second image, including: performing spatial adaptive weighted difference on the first image and the second image using the following formula: in, Pixel coordinates in the original fluorescence signal image The intensity value at that location, The intensity value of the first image at that point. The intensity value of the second image at that point. It is a gain correction factor used to correct for small systematic gain differences that may exist between two exposures. It is a natural exponential function. As a calibrable attenuation coefficient.

2. The automated control method for a production line operation of sweet potato seedling cultivation according to claim 1, characterized in that, Based on the transmission belt speed, node coordinate localization based on physiological signal gradients was performed on the fluorescence distribution map of the vines to obtain the vine node position sequence, including: A one-dimensional fluorescence intensity signal waveform is obtained by longitudinal integral projection of the fluorescence distribution map of the vines. A feature peak search based on second-order derivative is performed on the one-dimensional fluorescence intensity signal waveform to obtain the set of node positions in the image coordinate system. Based on the transmission belt speed and the calibration coefficient of the vision system, a physical coordinate mapping with multi-dimensional parameter fusion is performed on the set of node positions in the image coordinate system to obtain the vine node position sequence.

3. The automated control method for a production line operation of sweet potato seedling cultivation according to claim 2, characterized in that, To obtain the set of node positions in the image coordinate system by performing a feature peak search based on the second derivative of a one-dimensional fluorescence intensity signal waveform, the method includes: performing a feature peak search based on the second derivative of the one-dimensional fluorescence intensity signal waveform using the following formula: in, This is a one-dimensional fluorescence intensity signal waveform. It is the set of node positions in the image coordinate system. Let be the displacement variable along the direction of movement of the conveyor belt. The first derivative of the one-dimensional fluorescence intensity signal waveform. The second derivative of the one-dimensional fluorescence intensity signal waveform. The preset fluorescence intensity noise threshold, For logical AND operator.

4. The automated control method for assembly line operation of sweet potato seedling cultivation according to claim 1, characterized in that, Based on a set of agronomic parameters, optimal segmentation planning under variable-length internode constraints is performed on the vine node position sequence to obtain the set of physical coordinates of the target cutting points, including: Based on the set of agronomic parameters, a Boolean evaluation model for qualified seedling segments is performed on the sequence of vine node positions to obtain the evaluation function for qualified seedling segments; Based on the evaluation function of qualified seedling segments, the maximum output path of the total vine length is solved by dynamic programming to obtain the dynamic programming state table and the optimal path backtracking table. The optimal solution backtracking and cutting point sequence generation are performed on the dynamic programming state table, optimal path backtracking table, and total vine length to obtain the set of physical coordinates of the target cutting points.

5. An automated control device for a production line of sweet potato seedling cultivation, used to perform the method as described in any one of claims 1 to 4, characterized in that, include: The image acquisition module is used to acquire a first image and a second image of the sweet potato vine physical entity. The first image is the original image containing ambient light and excitation fluorescence, and the second image is the background image containing only ambient light. The vine fluorescence distribution map acquisition module is used to perform differential denoising, region of interest extraction, and morphological optimization on the first and second images to obtain the vine fluorescence distribution map. The node coordinate localization module is used to locate the node coordinates of the vine fluorescence distribution map based on the physiological signal gradient, based on the transmission belt speed, so as to obtain the vine node position sequence. The optimal segmentation planning module is used to perform optimal segmentation planning on the vine node position sequence under variable internode length constraints based on the set of agronomic parameters to obtain the set of physical coordinates of the target cutting point. The dynamic tracking control module is used to perform vision-encoder fusion-based dynamic tracking control on the real-time pulse count of the conveyor belt encoder and the set of physical coordinates of the target cutting point to obtain the position and speed command stream sent to the multi-axis servo system.

6. The automated control equipment for sweet potato seedling production line as described in claim 5, characterized in that, The vine fluorescence distribution map acquisition module includes: The fluorescence probability distribution map acquisition unit is used to perform differential denoising and region of interest extraction on the first image and the second image to obtain the fluorescence probability distribution map; A morphological optimization unit is used to perform morphological optimization on the fluorescence probability distribution map to obtain the fluorescence distribution map of the vine.

7. The automated control equipment for sweet potato seedling production line as described in claim 6, characterized in that, The node coordinate positioning module includes: The longitudinal integral projection unit is used to perform longitudinal integral projection on the vine fluorescence distribution map to obtain a one-dimensional fluorescence intensity signal waveform. The feature peak search unit is used to perform a feature peak search based on the second-order derivative on the one-dimensional fluorescence intensity signal waveform to obtain the set of node positions in the image coordinate system. The physical coordinate mapping unit is used to perform multi-dimensional parameter fusion physical coordinate mapping on the set of node positions in the image coordinate system based on the transmission belt speed and the calibration coefficient of the vision system to obtain the vine node position sequence.

Images