Ice output equipment regulation and control method and system based on intelligent decision

Abstract

The invention relates to the technical field of commercial refrigeration equipment, in particular to an ice output equipment regulation and control method and system based on intelligent decision making. According to the pose internal reference of the industrial camera, emitting pixel mapping light rays towards each pixel point, located in the three-dimensional scene space of the ice outlet equipment, of the ideal transparent incomplete feature of the ice outlet ice block, and executing coding decision on a series of random sampling propagation points on the pixel mapping light rays by using a coding radiation domain of the three-dimensional scene space; if the intelligent ice making decision result shows that defective ice blocks exist, constraint planning is carried out on the incomplete local area surrounding model, voxelization is carried out on a voxel triangle of the incomplete local area surrounding model, and hash accumulation of negotiation judgment is carried out on the voxel triangle and M sub-cubic nodes of a voxel surrounding tree of a preset ice making control strategy division model; and regulating and controlling the ice outlet equipment according to the Hash accumulation result. Defect detection of the ice blocks can be optimized, reasonable regulation and control parameters of the ice can be intelligently decided according to the detection result, the ice making quality of the ice discharging equipment is improved, and the defect rate of the ice blocks is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of commercial refrigeration equipment, and in particular to an ice dispensing equipment control method and system based on intelligent decision-making. Background Art

[0002] In the modern commercial beverage industry, ice cubes are an indispensable ingredient, and the accuracy and quality of their supply directly impact the quality of the beverages and the consumer experience. Currently, existing ice-making and dispensing equipment on the market faces numerous challenges. During storage and transportation, ice cubes may melt, become stuck, or become irregularly shaped. However, the vision modules equipped with existing ice-dispensing equipment often struggle to effectively and accurately detect the missing portions of transparent ice cubes for quality inspection and comparison. Consequently, they are unable to promptly detect defective ice cubes and make appropriate screening and removal decisions. Providing poor-quality ice cubes to customers not only affects the appearance of the beverages but can also alter their taste and texture, negatively impacting businesses. Furthermore, some ice-dispensing equipment suffers from poor control performance and is unable to accurately regulate the appropriate condensation parameters during the production process. This results in the frequent production of low-quality ice cubes and the resulting erratic ice-dispensing behavior. Therefore, it is necessary to develop an ice-dispensing equipment control method that can effectively detect ice cube quality and appropriately adjust ice-dispensing parameters to address these issues, thereby significantly improving ice-dispensing accuracy and reliability. Summary of the Invention

[0003] The present invention overcomes the deficiencies of the prior art and provides an ice dispensing equipment control method and system based on intelligent decision-making.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A first aspect of the present invention provides an ice dispensing equipment control method based on intelligent decision-making, comprising the following steps:

[0006] S102: obtaining a real-time ice-making demand input by a user, and performing time sequence detection on ice cubes produced in response to the real-time ice-making demand through an industrial camera to obtain a real-time detection image of ice cubes produced in a preset time sequence;

[0007] S104: Constructing a pixel matching area, extracting a background reference image of a fixed-point reference background without an ice cube placed thereon, aligning the real-time detection image in the pixel matching area using the background reference image as a visual reference, and solving the optical flow displacement of the pixel points based on the characteristics of transparent refraction light to obtain the actual transparent defect features of the ice cube;

[0008] S106: emitting pixel mapping rays toward each pixel point in the three-dimensional scene space of the ice-discharging device where the ideal transparent defective feature of the ice cube is located based on the pose intrinsic parameters of the industrial camera, performing coding decisions on a series of randomly sampled propagation points on the pixel mapping rays using a coded radiation field in the three-dimensional scene space, and obtaining an intelligent ice-discharging decision result;

[0009] S108: If the intelligent ice-discharging decision result shows that there are defective ice cubes and the frequency of detection flashes exceeds the preset frequency, the incomplete local enclosure model is constrained and planned, the voxel triangles of the incomplete local enclosure model are voxelized, and the hash accumulation of the negotiation and judgment with the M sub-cube nodes of the voxel enclosure tree of the preset ice-making control pair planning sub-model is performed, and the condensing ice-making parameters of the ice-discharging equipment are adjusted according to the hash accumulation result.

[0010] More specifically, the step S104 includes the following steps:

[0011] Obtain the fixed-point reference background and detection log of ice cubes detected by the industrial camera, retrieve the detection log to obtain the detection image of the fixed-point reference background without ice cubes, and mark it as the background reference image;

[0012] The Sobel operator is introduced to perform a first-order Taylor expansion calculation on the pixel changes between the real-time detection image and the background reference image to obtain spatial gradients and temporal gradients. Based on the spatial gradients and temporal gradients, an optical flow constraint equation for the pixel changes between the real-time detection image and the background reference image is constructed.

[0013] Constructing a pixel pairing field by aligning each pixel of the real-time detection image one by one with each pixel of the background reference image in the pixel pairing field, outputting a plurality of groups of pixel pairs of interest, obtaining an optical flow window for each group of pixel pairs of interest, and continuously superimposing and solving the optical flow constraint equation corresponding to each group of pixel pairs of interest in the optical flow window during the pairing process to generate a two-dimensional pixel optical flow displacement field;

[0014] Through optical knowledge graph recognition and real-time detection of images, the refractive properties of light under the condition that the ice cube is the medium are obtained. Based on the refractive properties, a least squares energy function of the refracted light of the ice cube is established. The central difference method is introduced to transform the least squares energy function into the Poisson equation. The discretized refractive scalar is recovered and solved in the Poisson equation for the two-dimensional pixel optical flow displacement field to obtain the refractive phase gradient.

[0015] A feature imaging architecture is constructed, and the two-dimensional pixel optical flow displacement field is back-projected to the feature imaging architecture following the refraction phase gradient, and finally the actual transparent incomplete features of the ice cube are generated.

[0016] More specifically, the step S106 includes the following steps:

[0017] Obtaining the desired ice cube output determined by the ice dispensing device based on the real-time ice making demand, obtaining the transparent defect feature of the desired ice cube placed on a fixed reference background based on a big data network search, defining the feature as an ideal transparent defect feature, and simultaneously obtaining the pose internal parameters of the industrial camera when capturing the ideal transparent defect feature;

[0018] Construct a 3D scene space for ice-producing equipment control and industrial camera detection. A multi-layer perceptron model is introduced to set the radiation network in the 3D space. Based on the radiation network, a coded radiation domain for the 3D scene space is constructed.

[0019] According to the preset illumination direction of the pose intrinsic parameter, a mapping ray is emitted from the detection optical center of the industrial camera along the illumination direction toward each pixel point where the ideal transparent defect feature is located in the three-dimensional scene space, thereby obtaining a plurality of pixel mapping rays;

[0020] A series of three-dimensional space points are randomly sampled uniformly along each pixel mapping ray to obtain N random sampling propagation points, and the position coordinates and observation direction of each random sampling propagation point in the three-dimensional scene space are obtained;

[0021] The volume density code and color code of each randomly sampled propagation point are obtained by injecting the coded radiation domain code according to the position coordinates at that time and the observation direction at that time. The ideal transparent incomplete features are rendered based on the volume density code and color code, and the quality of the ice cubes is judged by the hash misalignment decision between the rendered ideal transparent incomplete features and the actual transparent incomplete features to obtain the intelligent ice-making decision result.

[0022] More specifically, the method includes the following steps: obtaining a volume density code and a color code of each randomly sampled propagation point based on the position coordinates and the observation direction at that time, rendering an ideal transparent defect feature based on the volume density code and the color code, and determining the quality of the ice cubes through a hash misalignment decision between the rendered ideal transparent defect feature and the actual transparent defect feature to obtain an intelligent ice release decision result.

[0023] Inputting the N randomly sampled propagation points into a coded radiation field based on the position coordinates and the observation direction at that time, assigning a volume density and a color to each randomly sampled propagation point and performing sinusoidal position coding, and outputting a volume density code and a color code for each randomly sampled propagation point;

[0024] Rendering the contribution of each randomly sampled propagation point to each pixel mapping ray according to the volume density coding and color coding integrals to obtain an ideal rendering of the ideal transparent defect feature, and synchronously coding and rendering to obtain an actual rendering of the actual transparent defect feature;

[0025] The hash misalignment value between the actual rendering image and the ideal rendering image is calculated. If the hash misalignment value is less than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a complete ice cube; if the hash misalignment value is greater than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a defective ice cube, and the intelligent ice-making decision result is obtained.

[0026] More specifically, the step S108 includes the following steps:

[0027] If the intelligent ice-discharging decision result of the ice-discharging device shows that there are defective ice blocks, the detection flash frequency of the defective ice blocks is obtained by counting. If the detection flash frequency exceeds the preset detection flash frequency, a model voxel bounding tree is created;

[0028] The current ice-making rate of the ice-dispensing device and the preset ice-making control strategy are obtained, the voxel stacking resolution is set according to the current ice-making rate, the model voxel bounding tree is divided into M sub-cube nodes according to the tree branching level until the voxel stacking resolution is reached, and a hash algorithm is simultaneously introduced to calculate the local sensitive hash value of each sub-cube node;

[0029] A complete three-dimensional model of an ideal ice cube is established, a defect demarcation threshold is determined based on a hash dislocation function, and a defect boundary is planned in the complete three-dimensional model using the defect demarcation threshold as a termination constraint to obtain a defect local enclosure model of the actual ice cube relative to the ideal ice cube.

[0030] A voxelization algorithm is introduced, and based on a preset ice-making control strategy, the incomplete local enclosure model is triangulated in the voxelization algorithm to generate a plurality of voxel triangles belonging to the incomplete local enclosure model. Each of the voxel triangles is mapped to a model voxel enclosure tree, and each voxel triangle is checked to see whether it intersects with each sub-cube node.

[0031] If a voxel triangle interferes with a sub-cube node, the local sensitive hash values of adjacent filling cube nodes are concatenated and accumulated to output a filling cube aggregation node according to the filling voxel enclosure layout lattice. The hashes of the accumulated filling cube aggregation nodes are then concatenated to obtain a cumulative hash tree of the filling voxel enclosure layout lattice. The condensation and ice-making parameters of the ice-dispensing equipment are regulated based on the leaf hash growth of the cumulative hash tree.

[0032] More specifically, if a certain voxel triangle and a certain sub-cube node have an interaction phenomenon, then the local sensitive hash values of the adjacent filling cube nodes are concatenated and accumulated to output a filling cube aggregation node according to the filling voxel enclosing layout lattice, and the hashes of the accumulated filling cube aggregation nodes are further concatenated to obtain a cumulative hash tree of the filling voxel enclosing layout lattice. The condensation and ice-making parameters of the ice-dispensing device are regulated based on the leaf hash growth of the cumulative hash tree, specifically including the following steps:

[0033] If a certain voxel triangle interferes with a certain sub-cube node, the sub-cube node is marked as a filling cube node, and one or more filling cube nodes are obtained;

[0034] Obtaining a filling voxel enclosing layout lattice formed by one or more filling cube nodes, and concatenating and accumulating the local sensitive hash values of two adjacent filling cube nodes based on the filling voxel enclosing layout lattice to generate a filling cube aggregation node;

[0035] After concatenation and accumulation, perform independent secondary hashing on the filled cube aggregation nodes to generate the local sensitive hash values of the filled cube aggregation nodes;

[0036] The local sensitive hash values of the two adjacent filled cube aggregation nodes are concatenated and accumulated, and hash assignment is continued. This is repeated until only one filled cube aggregation node remains, and then hash accumulation is stopped. Finally, the accumulated hash tree of the filled voxel enclosing layout lattice is obtained.

[0037] The leaf hash growth pattern of the cumulative hash tree and the growth hash value of each leaf growth node in the leaf hash growth pattern are extracted. The ice-making control range of the ice-dispensing device is determined based on the leaf hash growth pattern. The condensing ice-making parameters of the ice-dispensing device are adjusted within the ice-making control range according to the growth hash value of each leaf growth node.

[0038] A second aspect of the present invention provides an ice-dispensing device control system based on intelligent decision-making. The ice-dispensing device control system includes a memory and a processor. The memory stores an ice-dispensing device control method program based on intelligent decision-making. When the ice-dispensing device control method program is executed by the processor, any one of the steps of the ice-dispensing device control method is implemented.

[0039] The present invention solves the technical defects existing in the background technology, and the beneficial technical effects of the present invention are:

[0040] The real-time ice-making demand input by the user is obtained, and the ice cubes discharged in accordance with the real-time ice-making demand are detected in time sequence by an industrial camera to obtain a real-time detection image of the ice cubes discharged at a preset time sequence; a pixel matching field is constructed, and a background reference image without ice cubes placed at a fixed reference background is extracted; the real-time detection image is aligned in the pixel matching field with the background reference image as a visual reference, and the optical flow displacement of the pixel point is solved based on the characteristics of transparent refraction light to obtain the actual transparent defect feature of the ice cubes discharged; the ideal transparent defect feature of the ice cubes discharged is located at the ice discharge device according to the internal reference of the industrial camera's posture Each pixel in a three-dimensional scene space emits a pixel mapping ray, and a coded radiation domain in the three-dimensional scene space is used to perform coding decisions on a series of randomly sampled propagation points on the pixel mapping ray, resulting in an intelligent ice-making decision result. If the intelligent ice-making decision result indicates the presence of defective ice cubes and exceeds a preset detection flash frequency, a constraint planning incomplete local enclosing model is used. The voxel triangles of the incomplete local enclosing model are voxelized and hashed and accumulated with the M sub-cubic nodes of the voxel enclosing tree of the pre-set ice-making control pair planning sub-model. The condensation ice-making parameters of the ice-making device are adjusted based on the hash accumulation results. The present invention can extract incomplete features under transparent characteristics from ice-making images captured by an industrial camera and use these incomplete features to determine whether the ice cubes are qualified. If the ice cubes are unqualified, the missing local features are hashed and accumulated to rationally control the ice-making device, thereby improving the ice-making integrity of the ice-making device, reducing the ice defect rate, optimizing ice-making quality, and enhancing user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0042] Figure 1 A first method flow chart of a method for controlling ice discharging equipment based on intelligent decision-making is shown;

[0043] Figure 2 A second method flow chart of an ice dispensing equipment control method based on intelligent decision-making is shown;

[0044] Figure 3 The system framework diagram of an ice dispensing equipment control system based on intelligent decision-making is shown. DETAILED DESCRIPTION

[0045] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0047] The first aspect of the present invention provides a method for controlling ice dispensing equipment based on intelligent decision making, such as Figure 1 As shown, the following steps are included:

[0048] S102: obtaining a real-time ice-making demand input by a user, and performing time sequence detection on ice cubes produced in response to the real-time ice-making demand through an industrial camera to obtain a real-time detection image of ice cubes produced in a preset time sequence;

[0049] S104: Constructing a pixel matching area, extracting a background reference image of a fixed-point reference background without an ice cube placed thereon, aligning the real-time detection image in the pixel matching area using the background reference image as a visual reference, and solving the optical flow displacement of the pixel points based on the characteristics of transparent refraction light to obtain the actual transparent defect features of the ice cube;

[0050] S106: emitting pixel mapping rays toward each pixel point in the three-dimensional scene space of the ice-discharging device where the ideal transparent defective feature of the ice cube is located based on the pose intrinsic parameters of the industrial camera, performing coding decisions on a series of randomly sampled propagation points on the pixel mapping rays using a coded radiation field in the three-dimensional scene space, and obtaining an intelligent ice-discharging decision result;

[0051] S108: If the intelligent ice-discharging decision result shows that there are defective ice cubes and the frequency of detection flashes exceeds the preset frequency, the incomplete local enclosure model is constrained and planned, the voxel triangles of the incomplete local enclosure model are voxelized, and the hash accumulation of the negotiation and judgment with the M sub-cube nodes of the voxel enclosure tree of the preset ice-making control pair planning sub-model is performed, and the condensing ice-making parameters of the ice-discharging equipment are adjusted according to the hash accumulation result.

[0052] More specifically, the step S104 includes the following steps:

[0053] Obtain the fixed-point reference background and detection log of ice cubes detected by the industrial camera, retrieve the detection log to obtain the detection image of the fixed-point reference background without ice cubes, and mark it as the background reference image;

[0054] The Sobel operator is introduced to perform a first-order Taylor expansion calculation on the pixel changes between the real-time detection image and the background reference image to obtain spatial gradients and temporal gradients. Based on the spatial gradients and temporal gradients, an optical flow constraint equation for the pixel changes between the real-time detection image and the background reference image is constructed.

[0055] Constructing a pixel pairing field by aligning each pixel of the real-time detection image one by one with each pixel of the background reference image in the pixel pairing field, outputting a plurality of groups of pixel pairs of interest, obtaining an optical flow window for each group of pixel pairs of interest, and continuously superimposing and solving the optical flow constraint equation corresponding to each group of pixel pairs of interest in the optical flow window during the pairing process to generate a two-dimensional pixel optical flow displacement field;

[0056] Through optical knowledge graph recognition and real-time detection of images, the refractive properties of light under the condition that the ice cube is the medium are obtained. Based on the refractive properties, a least squares energy function of the refracted light of the ice cube is established. The central difference method is introduced to transform the least squares energy function into the Poisson equation. The discretized refractive scalar is recovered and solved in the Poisson equation for the two-dimensional pixel optical flow displacement field to obtain the refractive phase gradient.

[0057] A feature imaging architecture is constructed, and the two-dimensional pixel optical flow displacement field is back-projected to the feature imaging architecture following the refraction phase gradient, and finally the actual transparent incomplete features of the ice cube are generated.

[0058] It should be noted that most ice cubes are transparent, and when transparent ice cubes are placed on the inspection table for inspection, the blurred outline of the inspection background may be revealed through the ice cubes themselves. This makes the blurred effect of the inspection background when the industrial camera detects transparent ice cubes interfere with the extraction of the transparent features of the ice cubes. In addition, the defective surfaces of ice cubes with local defects are mostly reflective smooth mirrors. The refraction of light on these surfaces will also produce varying degrees of loss error in the defect feature recognition and extraction of the industrial camera, making it difficult to identify the defects of the transparent ice cubes, greatly increasing the error rate of the industrial camera in identifying defects in transparent ice cubes. To address this, the present method obtains image data of the inspection background when an industrial camera detects ice cubes, such as an inspection table without ice cubes or an ice cube conveyor belt. This background image is used as a visual noise reference benchmark for pixel comparison extraction between ice cubes with and without ice cubes, which can provide a basis for eliminating the contour interference of the subsequent detection background contour mapped on the transparent ice cubes. Since the detection process from no ice cubes to ice cubes on the background is a dynamic reference detection process from nothing to something, the pixel changes have a motion reference optical flow gradient with a certain velocity vector in the detection space and detection time of the ice-discharging device. Therefore, this method uses Taylor expansion to detect the pixel changes between the real-time detection image and the background reference image, so that the linear gradient trend from no ice cubes to ice cubes in space and time can be accurately captured, thereby making the optical flow constraints between adjacent pixels jump more smoothly, making the overlap of the detection background and the transparent ice cube area flatter, effectively filtering and eliminating background noise, and improving the robustness of transparent feature extraction.

[0059] It should be noted that this method uses a pixel pairing field to align the real-time detection image and the background reference image. During the pairing process, the optical flow windows of the pixel pairs of interest are continuously superimposed to solve the corresponding optical flow constraint equations for each group of pixel pairs of interest. For example, an optical flow window is taken with a certain pixel pair of interest as the center, such as 5×5 (25 pixels in total), and an optical flow constraint equation is written for each pixel in the window, so that a linear equation matrix can be obtained. In this way, the uncertainty of each pixel can be aggregated into an optical flow as a whole, eliminating the phenomenon that the gradient of a single pixel is easily affected by noise, thereby stably solving the local motion, and further estimating the incomplete relative pixel displacement of the real-time detection image with the background reference image as a reference, reducing the degradation of transparent features when the detection reference background is eliminated, and maximally retaining and reflecting the texture information of transparent objects after the detection background noise is eliminated. Next, a least-squares energy function for the refraction of light by the ice cube is established based on the refractive properties of light in the transparent ice medium. This least-squares energy function measures the difference between the refraction gradient of a standard transparent ice cube and the two-dimensional pixel optical flow displacement field, representing the tolerable light refraction noise and inconsistency. The two-dimensional pixel optical flow displacement field is then restored to its original scalar field from the gradient field using this least-squares energy function. Transforming the least-squares energy function into a Poisson equation reduces the characteristic scalar error of pixel displacement under light refraction noise and improves the clarity of feature interpretation. The discretized refraction scalar is then recovered and solved within the Poisson equation to obtain a linear refraction scalar array represented by a sparse gradient at the ice cube pixel. The gradient function value at each pixel can be solved on a discrete image grid, thereby converting the pixel displacement into a more physically meaningful refractive index gradient field, and inferring the refractive index phase change of the ice cube's missing surface. Finally, the two-dimensional pixel optical flow displacement field is back-projected onto a feature imaging architecture based on the refraction phase gradient, thus restoring the missing features of the transparent ice cube detected by industrial cameras.

[0060] It should be noted that this method can calculate the pixel displacement of the actual ice image relative to the reference image with the detection background as a reference, and back-project the pixel displacement based on the refraction characteristics of light on the defective ice surface, so as to infer the missing features on the transparent ice. Compared with traditional visual inspection methods, it can cope with the feature extraction errors of transparent ice surface images, eliminate the noise interference of transparent factors, and make the feature detection of transparent objects more accurate, thereby improving the accuracy of ice-discharging equipment in judging ice defects.

[0061] More specifically, the step S106 includes the following steps:

[0062] Obtaining the desired ice cube output determined by the ice dispensing device based on the real-time ice making demand, obtaining the transparent defect feature of the desired ice cube placed on a fixed reference background based on a big data network search, defining the feature as an ideal transparent defect feature, and simultaneously obtaining the pose internal parameters of the industrial camera when capturing the ideal transparent defect feature;

[0063] Construct a 3D scene space for ice-producing equipment control and industrial camera detection. A multi-layer perceptron model is introduced to set the radiation network in the 3D space. Based on the radiation network, a coded radiation domain for the 3D scene space is constructed.

[0064] According to the preset illumination direction of the pose intrinsic parameter, a mapping ray is emitted from the detection optical center of the industrial camera along the illumination direction toward each pixel point where the ideal transparent defect feature is located in the three-dimensional scene space, thereby obtaining a plurality of pixel mapping rays;

[0065] A series of three-dimensional space points are randomly sampled uniformly along each pixel mapping ray to obtain N random sampling propagation points, and the position coordinates and observation direction of each random sampling propagation point in the three-dimensional scene space are obtained;

[0066] The volume density code and color code of each randomly sampled propagation point are obtained by injecting the coded radiation domain code according to the position coordinates at that time and the observation direction at that time. The ideal transparent incomplete features are rendered based on the volume density code and color code, and the quality of the ice cubes is judged by the hash misalignment decision between the rendered ideal transparent incomplete features and the actual transparent incomplete features to obtain the intelligent ice-making decision result.

[0067] It should be noted that, in order to satisfy the user's dining experience, most existing ice-dispensing equipment is equipped with an ice quality detection and rejection module, which is used to screen out and reject ice cubes that are melted, irregular in shape, or other quality that does not meet the requirements, so that the ice cubes produced by the ice-dispensing equipment have a higher degree of integrity. However, traditional image defect detection methods have a mapping penetration phenomenon when depicting the detailed and incomplete features of transparent objects. This causes a large deviation in the comparison of local transparent features, which easily leads to false detection and omission of defective ice cubes, resulting in a high error rate in ice rejection decisions, reducing the overall quality of the ice cubes. To this end, this method first obtains an expected ice cube output that meets the real-time ice-making needs. The expected ice cube output is a quantitative benchmark template that reflects the degree of defects in the actual ice cube output. Since the industrial camera detects the continuously conveyed ice cubes in the internal space of the ice-discharging equipment while adjusting different postures, angles, and focus parameters, the reconstruction of the transparent and incomplete features of the ice cubes needs to maintain the mapping criteria of the three-dimensional scene space of the industrial camera adjusting the posture and internal parameters for detecting ice cubes. Therefore, this method uses a multi-layer perceptron to map the position and observation direction in the above three-dimensional scene space to specific volume density and color, thereby realizing nonlinear fitting of the scene radiation field, making the interpretation of information such as transparent image texture contour clearer and richer in details. A mapping ray is then emitted from the industrial camera's detection optical center, following the illumination direction of the pose intrinsic parameter, toward each pixel in the 3D scene space where the ideal transparent defect feature is located. This mapping ray serves as a bridge connecting the scene to the transparent defect features detected by the industrial camera in multiple dimensions. The ray carries the camera's viewpoint information. By sampling and calculating points along the mapping ray, the rendering result of the scene detection of the ideal transparent defect feature at that viewpoint can be obtained. A series of random sampling propagation points in 3D space are then uniformly and randomly sampled along each pixel mapping ray. This series of random sampling propagation points is designed to obtain sufficient information along the ray to calculate the volume rendering integral, thereby maximally reproducing the local detailed texture and contour of the actual transparent defect feature of the ice block, significantly improving the structural fidelity and reproducibility of the ideal transparent defect feature detected by the industrial camera. Furthermore, appropriate ray sampling can reduce the computational effort while ensuring rendering quality. Too few sampling points can lead to issues such as aliasing and holes in the rendering result; too many sampling points increase the computational burden and reduce rendering efficiency.

[0068] It should be noted that the random sampling propagation points and the corresponding positions and observation directions obtained by sampling are input into the constructed coded radiation domain to predict the volume density and color of each point, providing the necessary parameters for subsequent volume rendering. Finally, a rendering image of the ideal transparent defect feature can be output. By comparing this ideal rendering image with the rendering image of the actual transparent defect feature, it is possible to further determine whether the defects of the actual ice cubes are qualified, thereby realizing intelligent decision-making for ice cube quality detection and rejection. This method can accurately render the feature information of the ideal template of transparent ice cubes in detail, retain and restore the feature scale of local missing parts of transparent characteristic objects to the greatest extent, thereby providing a comparison basis for the transparent defect features of the actual ice cubes to be rejected, effectively improving the accuracy of the ice-dispensing equipment in rejecting defective ice cubes, reducing false detection, missed detection and misjudgment, and optimizing the control accuracy of the ice-dispensing equipment.

[0069] More specifically, the method includes the following steps: obtaining a volume density code and a color code of each randomly sampled propagation point based on the position coordinates and the observation direction at that time, rendering an ideal transparent defect feature based on the volume density code and the color code, and determining the quality of the ice cubes through a hash misalignment decision between the rendered ideal transparent defect feature and the actual transparent defect feature to obtain an intelligent ice release decision result.

[0070] Inputting the N randomly sampled propagation points into a coded radiation field based on the position coordinates and the observation direction at that time, assigning a volume density and a color to each randomly sampled propagation point and performing sinusoidal position coding, and outputting a volume density code and a color code for each randomly sampled propagation point;

[0071] Rendering the contribution of each randomly sampled propagation point to each pixel mapping ray according to the volume density coding and color coding integrals to obtain an ideal rendering of the ideal transparent defect feature, and synchronously coding and rendering to obtain an actual rendering of the actual transparent defect feature;

[0072] The hash misalignment value between the actual rendering image and the ideal rendering image is calculated. If the hash misalignment value is less than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a complete ice cube; if the hash misalignment value is greater than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a defective ice cube, and the intelligent ice-making decision result is obtained.

[0073] It should be noted that for the mapping of randomly sampled propagation points within the coded radiation domain, the position coordinates and observation direction of each randomly sampled propagation point are input into the coded radiation domain. Through forward propagation of the coded radiation domain, the input ideal transparent defect feature geometric information is converted into the physical properties of the detection scene. Since the original input position coordinates may not be able to highly restore the feature details of the complex ice surface in the low-dimensional scene space, in order to enhance the radiation domain's ability to express the defects of high-frequency transparent feature information, this method performs sinusoidal position encoding on the input position and direction of the industrial camera multi-dimensional detection scene mapping. Sinusoidal position encoding can encode the local surface detail information of the ideal ice corresponding to the actual transparent defect feature into high-dimensional features, making it possible to capture subtle changes in the defect feature and significantly improve the rendering quality of ice quality inspection. Compared with traditional feature mapping description methods without position encoding, underfitting of the transparent surface may occur, resulting in the rendering results being penetrating, blurred, and lacking important details, greatly improving the accuracy of ice defect detection. If the hash misalignment value is less than the preset hash misalignment value, it indicates that the degree of defect on the actual ice cube is small, equal to or similar to a complete ice cube, and the ice cube can be used by customers. Therefore, the ice cube with the actual transparent defect feature is marked as a complete ice cube. Otherwise, it means that the degree of defect on the actual ice cube is greater than that of the ideal ice cube, and does not meet the use scope of complete ice cubes. Therefore, it is marked as a defective ice cube. This method can provide correct position encoding for the defect feature mapping of the ideal ice cube, thereby improving the detail and restoration of the transparent defect texture rendering on the ideal ice cube, making defect detection and judgment more accurate and reliable. At the same time, it can determine whether the surface defect of the current ice cube is qualified, thereby ensuring that the ice cubes produced by the ice dispensing equipment have a higher degree of integrity and significantly improving the quality of the ice cubes.

[0074] More specifically, the step S108 is as follows: Figure 2 As shown, the specific steps include:

[0075] S202: If the intelligent ice-discharging decision result of the ice-discharging device indicates that defective ice cubes exist, a detection flash frequency of the defective ice cubes is obtained by counting and if the detection flash frequency exceeds a preset detection flash frequency, a model voxel bounding tree is created;

[0076] S204: Obtain the current ice-making rate of the ice-dispensing device and the preset ice-making control strategy, set the voxel stacking resolution according to the current ice-making rate, divide the model voxel bounding tree into M sub-cube nodes according to the tree branching level until the voxel stacking resolution is reached, and simultaneously introduce a hash algorithm to calculate the local sensitive hash value of each sub-cube node;

[0077] S206: Establishing a complete 3D model of the ideal ice cube, determining a defect boundary threshold based on a hash dislocation function, planning a defect boundary in the complete 3D model using the defect boundary threshold as a termination constraint, and obtaining a defect local enclosing model of the actual ice cube relative to the ideal ice cube.

[0078] S208: Introducing a voxelization algorithm, performing triangular cutting decomposition on the incomplete local bounding model in the voxelization algorithm based on a preset ice making control strategy, generating a plurality of voxel triangles belonging to the incomplete local bounding model, mapping each of the voxel triangles to a model voxel bounding tree, and checking whether each voxel triangle intersects with each sub-cube node;

[0079] S210: If there is an interaction between a certain voxel triangle and a certain sub-cube node, the local sensitive hash values of the adjacent filling cube nodes are concatenated and accumulated according to the filling voxel enclosing layout lattice to output a filling cube aggregation node, and the hashes of the accumulated filling cube aggregation nodes are further concatenated to obtain a cumulative hash tree of the filling voxel enclosing layout lattice. The condensation and ice-making parameters of the ice-dispensing equipment are regulated based on the leaf hash growth of the cumulative hash tree.

[0080] It should be noted that ice production in ice-dispensing equipment primarily involves the physical transformation of circulating water from liquid to solid through refrigeration and compression. This transformation is a gradual process, and the efficiency of this process depends on the condensation parameters. One of the causes of localized ice defects is improper control of the condensation parameters in the ice-dispensing equipment. Therefore, the presence of defects in ice is positively correlated with the rationality of the condensation parameters. To address this, this method first determines whether the frequency of ice defects produced by the ice-dispensing equipment exceeds a specified limit. If the detection frequency exceeds a preset detection frequency, it indicates that the ice-dispensing equipment's condensation parameters are being controlled inappropriately, resulting in frequent ice defects. Further control of the condensation parameters is necessary. A model voxel bounding tree is then created. This voxel bounding tree is a spatial index structure for the voxelized model data of each preset ice-making control strategy, providing a hierarchical query framework for subsequent voxelization of defective models. The model voxel bounding tree is divided into a hierarchical structure of M sub-cubic nodes according to the tree branching level. This reduces unnecessary computation steps for incomplete voxels. For example, completely blank areas do not need to be further subdivided, improving the control and response efficiency of the ice-making equipment during condensation. The current ice-making rate reflects the rate required to condense the incomplete area and is a reflection of the resolution of the voxel stack. Compared to a uniform-resolution voxel grid, dividing the voxel stack according to its resolution allows for localized refinement only where needed, improving the control accuracy of the condensation ice-making parameters to fully produce ice cubes in the incomplete areas. Then, an incomplete local enclosure model is constructed according to the hash dislocation value, that is, the incomplete part of the actual ice cube relative to the ideal ice cube. The reasonable regulation of the condensation ice-making parameters is to repair the incomplete part. Therefore, based on the preset ice-making control strategy, the incomplete local enclosure model voxel triangles are further divided in the voxelization algorithm, and each voxel triangle is mapped to the model voxel enclosure tree, so that the continuous geometric shape can be converted into a discrete voxel expression of the index-joint preset ice-making control strategy, so that the ice-making control of the final accumulated incomplete voxels is more accurate and responsive, and the condensation ice-making parameter regulation required to repair the incomplete part of the ice cube is more accurate and reliable.

[0081] It should be noted that after the segmentation is completed, the incomplete local enclosing model is traversed to determine the intersection of the voxel triangles of each geometric unit with the sub-cube nodes. If a voxel triangle intersects with a sub-cube node, it means that the sub-cube node has been filled by the voxel triangle, indicating that the control policy parameters indexed within the spatial structure of the sub-cube node are the key components for regulating the voxel triangle's belonging to the incomplete part. Therefore, it is marked as a filled-cube node and further spliced. During the splicing, the local sensitive hash values of the filled-cube nodes are continuously accumulated to form a cumulative hash tree that reveals the refrigeration control policy parameter matrix required to repair the incomplete part, that is, the cumulative hash tree of the filled-cube layout lattice. This cumulative hash tree presents the refrigeration parameter structure required to voxelize and repair the incomplete model. Through this method, the condensation refrigeration control parameters based on the preset ice making control policy can be indexed and accumulated by voxelizing the incomplete part of the ice cube, thereby avoiding the incomplete phenomenon caused by the ice dispensing equipment in refrigerating ice cubes, eliminating the defective ice cube rate, and effectively improving the quality and reliability of ice cube production.

[0082] More specifically, if a certain voxel triangle and a certain sub-cube node have an interaction phenomenon, then the local sensitive hash values of the adjacent filling cube nodes are concatenated and accumulated to output a filling cube aggregation node according to the filling voxel enclosing layout lattice, and the hashes of the accumulated filling cube aggregation nodes are further concatenated to obtain a cumulative hash tree of the filling voxel enclosing layout lattice. The condensation and ice-making parameters of the ice-dispensing device are regulated based on the leaf hash growth of the cumulative hash tree, specifically including the following steps:

[0083] If a certain voxel triangle interferes with a certain sub-cube node, the sub-cube node is marked as a filling cube node, and one or more filling cube nodes are obtained;

[0084] Obtaining a filling voxel enclosing layout lattice formed by one or more filling cube nodes, and concatenating and accumulating the local sensitive hash values of two adjacent filling cube nodes based on the filling voxel enclosing layout lattice to generate a filling cube aggregation node;

[0085] After concatenation and accumulation, perform independent secondary hashing on the filled cube aggregation nodes to generate the local sensitive hash values of the filled cube aggregation nodes;

[0086] The local sensitive hash values of the two adjacent filled cube aggregation nodes are concatenated and accumulated, and hash assignment is continued. This is repeated until only one filled cube aggregation node remains, and then hash accumulation is stopped. Finally, the accumulated hash tree of the filled voxel enclosing layout lattice is obtained.

[0087] The leaf hash growth pattern of the cumulative hash tree and the growth hash value of each leaf growth node in the leaf hash growth pattern are extracted. The ice-making control range of the ice-dispensing device is determined based on the leaf hash growth pattern. The condensing ice-making parameters of the ice-dispensing device are adjusted within the ice-making control range according to the growth hash value of each leaf growth node.

[0088] It should be noted that for the voxel accumulation of the filling cube nodes, this method accumulates the local sensitive hash values of the two adjacent filling cube nodes by splicing them together. At this time, the local sensitive hash value generated by the splicing efficiently represents the repair summary of the preset ice-making control strategy for eliminating the local structure of the incomplete part of the ice cube, and the generated filling cube aggregation node is the inheritance relationship between the step-by-step accumulation of voxelized repair hash data. After the splicing and accumulation, an independent secondary hash is performed on the filling cube aggregation node, and the local sensitive hash values of the two adjacent filling cube aggregation nodes are continued to be spliced and accumulated and hashed. And so on, the hash accumulation is stopped until only one filled cube aggregation node is left. Through this hierarchical hash accumulation, the condensation ice-making control range based on the preset ice-making control strategy can be made clearer, avoiding the cumulative avalanche effect of the traditional method, and improving the condensation ice-making parameter control of the ice-dispensing equipment to reduce the occurrence of ice cube defects. It is more accurate and reliable, and reasonable condensation ice-making parameters can be adjusted according to the voxel shape of the defective part to achieve the controllability and integrity of the ideal ice cube shape, effectively improving the intelligence of the ice-dispensing equipment's control decision-making, and further ensuring the quality of ice cube production by the ice-dispensing equipment.

[0089] The second aspect of the present invention provides an ice dispensing equipment control system based on intelligent decision making, such as Figure 3 As shown, the ice dispensing device control system includes a memory 31 and a processor 32. The memory 31 stores an ice dispensing device control method program based on intelligent decision-making. When the ice dispensing device control method program is executed by the processor 32, any one of the steps of the ice dispensing device control method is implemented.

[0090] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for controlling ice dispensing equipment based on intelligent decision-making, characterized in that: The following steps are involved: S102: obtaining a real-time ice-making demand input by a user, and performing time sequence detection on ice cubes produced in response to the real-time ice-making demand through an industrial camera to obtain a real-time detection image of ice cubes produced in a preset time sequence; S104: Constructing a pixel matching area, extracting a background reference image of a fixed-point reference background without an ice cube placed thereon, aligning the real-time detection image in the pixel matching area using the background reference image as a visual reference, and solving the optical flow displacement of the pixel points based on the characteristics of transparent refraction light to obtain the actual transparent defect features of the ice cube; S106: emitting pixel mapping rays toward each pixel point in the three-dimensional scene space of the ice-discharging device where the ideal transparent defective feature of the ice cube is located based on the pose intrinsic parameters of the industrial camera, performing coding decisions on a series of randomly sampled propagation points on the pixel mapping rays using a coded radiation field in the three-dimensional scene space, and obtaining an intelligent ice-discharging decision result; S108: If the intelligent ice-discharging decision result shows that there are defective ice cubes and the frequency of detection flashes exceeds the preset frequency, the incomplete local enclosure model is constrained and planned, the voxel triangles of the incomplete local enclosure model are voxelized, and the hash accumulation of the negotiation and judgment with the M sub-cube nodes of the voxel enclosure tree of the preset ice-making control pair planning sub-model is performed, and the condensing ice-making parameters of the ice-discharging equipment are adjusted according to the hash accumulation result.

2. The method for controlling ice dispensing equipment based on intelligent decision-making according to claim 1, characterized in that: The step S104 specifically includes the following steps: Obtain the fixed-point reference background and detection log of ice cubes detected by the industrial camera, retrieve the detection log to obtain the detection image of the fixed-point reference background without ice cubes, and mark it as the background reference image; The Sobel operator is introduced to perform a first-order Taylor expansion calculation on the pixel changes between the real-time detection image and the background reference image to obtain spatial gradients and temporal gradients. Based on the spatial gradients and temporal gradients, an optical flow constraint equation for the pixel changes between the real-time detection image and the background reference image is constructed. Constructing a pixel pairing field by aligning each pixel of the real-time detection image one by one with each pixel of the background reference image in the pixel pairing field, outputting a plurality of groups of pixel pairs of interest, obtaining an optical flow window for each group of pixel pairs of interest, and continuously superimposing and solving the optical flow constraint equation corresponding to each group of pixel pairs of interest in the optical flow window during the pairing process to generate a two-dimensional pixel optical flow displacement field; Through optical knowledge graph recognition and real-time detection of images, the refractive properties of light under the condition that the ice cube is the medium are obtained. Based on the refractive properties, a least squares energy function of the refracted light of the ice cube is established. The central difference method is introduced to transform the least squares energy function into the Poisson equation. The discretized refractive scalar is recovered and solved in the Poisson equation for the two-dimensional pixel optical flow displacement field to obtain the refractive phase gradient. A feature imaging architecture is constructed, and the two-dimensional pixel optical flow displacement field is back-projected to the feature imaging architecture following the refraction phase gradient, and finally the actual transparent incomplete features of the ice cube are generated.

3. The method for controlling ice dispensing equipment based on intelligent decision-making according to claim 1, characterized in that: The step S106 specifically includes the following steps: Obtaining the desired ice cube output determined by the ice dispensing device based on the real-time ice making demand, obtaining the transparent defect feature of the desired ice cube placed on a fixed reference background based on a big data network search, defining the feature as an ideal transparent defect feature, and simultaneously obtaining the pose internal parameters of the industrial camera when capturing the ideal transparent defect feature; Construct a 3D scene space for ice-producing equipment control and industrial camera detection. A multi-layer perceptron model is introduced to set the radiation network in the 3D space. Based on the radiation network, a coded radiation domain for the 3D scene space is constructed. According to the preset illumination direction of the pose intrinsic parameter, a mapping ray is emitted from the detection optical center of the industrial camera along the illumination direction toward each pixel point where the ideal transparent defect feature is located in the three-dimensional scene space, thereby obtaining a plurality of pixel mapping rays; A series of three-dimensional space points are randomly sampled uniformly along each pixel mapping ray to obtain N random sampling propagation points, and the position coordinates and observation direction of each random sampling propagation point in the three-dimensional scene space are obtained; The volume density code and color code of each randomly sampled propagation point are obtained by injecting the coded radiation domain code according to the position coordinates at that time and the observation direction at that time. The ideal transparent incomplete features are rendered based on the volume density code and color code, and the quality of the ice cubes is judged by the hash misalignment decision between the rendered ideal transparent incomplete features and the actual transparent incomplete features to obtain the intelligent ice-making decision result.

4. The method for controlling ice dispensing equipment based on intelligent decision-making according to claim 3, characterized in that: The method calculates the volume density code and color code of each randomly sampled propagation point based on the position coordinates at that time and the coded radiation domain code injected into the observation direction at that time, renders the ideal transparent incomplete feature based on the volume density code and color code, and determines the quality of the ice cube by hashing the rendered ideal transparent incomplete feature and the actual transparent incomplete feature to obtain the intelligent ice release decision result, specifically including the following steps: Inputting the N randomly sampled propagation points into a coded radiation field based on the position coordinates and the observation direction at that time, assigning a volume density and a color to each randomly sampled propagation point and performing sinusoidal position coding, and outputting a volume density code and a color code for each randomly sampled propagation point; Rendering the contribution of each randomly sampled propagation point to each pixel mapping ray according to the volume density coding and color coding integrals to obtain an ideal rendering of the ideal transparent defect feature, and synchronously coding and rendering to obtain an actual rendering of the actual transparent defect feature; The hash misalignment value between the actual rendering image and the ideal rendering image is calculated. If the hash misalignment value is less than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a complete ice cube; if the hash misalignment value is greater than the preset hash misalignment value, the ice cube with the actual transparent incomplete feature is marked as a defective ice cube, and the intelligent ice-making decision result is obtained.

5. The method for controlling ice dispensing equipment based on intelligent decision-making according to claim 1, characterized in that: The step S108 specifically includes the following steps: If the intelligent ice-discharging decision result of the ice-discharging device shows that there are defective ice blocks, the detection flash frequency of the defective ice blocks is obtained by counting. If the detection flash frequency exceeds the preset detection flash frequency, a model voxel bounding tree is created; The current ice-making rate of the ice-dispensing device and the preset ice-making control strategy are obtained, the voxel stacking resolution is set according to the current ice-making rate, the model voxel bounding tree is divided into M sub-cube nodes according to the tree branching level until the voxel stacking resolution is reached, and a hash algorithm is simultaneously introduced to calculate the local sensitive hash value of each sub-cube node; A complete three-dimensional model of an ideal ice cube is established, a defect demarcation threshold is determined based on a hash dislocation function, and a defect boundary is planned in the complete three-dimensional model using the defect demarcation threshold as a termination constraint to obtain a defect local enclosure model of the actual ice cube relative to the ideal ice cube. A voxelization algorithm is introduced, and based on a preset ice-making control strategy, the incomplete local enclosure model is triangulated in the voxelization algorithm to generate a plurality of voxel triangles belonging to the incomplete local enclosure model. Each of the voxel triangles is mapped to a model voxel enclosure tree, and each voxel triangle is checked to see whether it intersects with each sub-cube node. If a voxel triangle interferes with a sub-cube node, the local sensitive hash values of adjacent filling cube nodes are concatenated and accumulated to output a filling cube aggregation node according to the filling voxel enclosure layout lattice. The hashes of the accumulated filling cube aggregation nodes are then concatenated to obtain a cumulative hash tree of the filling voxel enclosure layout lattice. The condensation and ice-making parameters of the ice-dispensing equipment are regulated based on the leaf hash growth of the cumulative hash tree.

6. The method for controlling ice dispensing equipment based on intelligent decision-making according to claim 5, characterized in that: If a certain voxel triangle and a certain sub-cube node have an interaction phenomenon, then the local sensitive hash values of the adjacent filling cube nodes are spliced and accumulated according to the filling voxel enclosing layout lattice to output a filling cube aggregation node, and then the hashes of the accumulated filling cube aggregation nodes are further spliced to obtain a cumulative hash tree of the filling voxel enclosing layout lattice. The condensation and ice making parameters of the ice-dispensing device are regulated based on the leaf hash growth of the cumulative hash tree, specifically including the following steps: If a certain voxel triangle interferes with a certain sub-cube node, the sub-cube node is marked as a filling cube node, and one or more filling cube nodes are obtained; Obtaining a filling voxel enclosing layout lattice formed by one or more filling cube nodes, and concatenating and accumulating the local sensitive hash values of two adjacent filling cube nodes based on the filling voxel enclosing layout lattice to generate a filling cube aggregation node; After concatenation and accumulation, perform independent secondary hashing on the filled cube aggregation nodes to generate the local sensitive hash values of the filled cube aggregation nodes; The local sensitive hash values of the two adjacent filled cube aggregation nodes are concatenated and accumulated, and hash assignment is continued. This is repeated until only one filled cube aggregation node remains, and then hash accumulation is stopped. Finally, the accumulated hash tree of the filled voxel enclosing layout lattice is obtained. The leaf hash growth pattern of the cumulative hash tree and the growth hash value of each leaf growth node in the leaf hash growth pattern are extracted. The ice-making control range of the ice-dispensing device is determined based on the leaf hash growth pattern. The condensing ice-making parameters of the ice-dispensing device are adjusted within the ice-making control range according to the growth hash value of each leaf growth node.

7. An ice dispensing equipment control system based on intelligent decision-making, characterized in that: The ice dispensing device control system includes a memory and a processor. The memory stores an ice dispensing device control method program based on intelligent decision-making. When the ice dispensing device control method program is executed by the processor, the ice dispensing device control method steps according to any one of claims 1 to 6 are implemented.

Images