Konjac heteromorphic slice detection method and system

Abstract

This invention relates to the field of machine vision technology and proposes a method and system for detecting irregularly shaped konjac slices. The method includes: acquiring initial image data of a target konjac slice under excitation by a variable-angle light source, and performing morphological reconstruction on the foreground region of the slice in the initial image data to obtain a slice contour image; performing edge curvature analysis on the slice contour image to identify contour abrupt change points; performing contour frequency domain quantization based on the contour abrupt change points to obtain contour feature parameters; performing inertial principal axis orientation and symmetry constraint correction based on the contour feature parameters to obtain a spatial attitude reference axis; performing radial offset analysis on edge points along the normal of the spatial attitude reference axis to construct a deformation distribution map; and performing deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result. This invention can improve the efficiency of detecting irregularly shaped konjac slices.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to a method and system for detecting irregularly shaped slices of konjac. Background Technology

[0002] Currently, the detection of irregularly shaped konjac slices still relies primarily on manual visual inspection. The detection process is highly dependent on the operator's visual judgment and experience. There is a lack of unified quantitative standards for detection, and the results are easily affected by subjective state, fatigue, and other factors, leading to frequent missed detections and incorrect detections. It is also difficult to guarantee the consistency and repeatability of the detection. At the same time, the manual detection speed is low and cannot match the high-speed turnover of industrial continuous production of konjac slices, becoming a bottleneck in the production process and restricting the improvement of overall production efficiency.

[0003] Traditional machine vision inspection solutions employ fixed-angle light sources and a single imaging mode, which can only acquire local image information of konjac slices. They cannot fully capture the slice's contour, texture, and deformation details. The contour feature extraction method is coarse, failing to perform fine analysis on key features such as edge curvature and contour topology. This makes it difficult to identify minute deformations and local irregular defects, resulting in low accuracy in irregularity classification, high false positive and false negative rates, and poor adaptability to the detection of konjac slices of different sizes and shapes, leading to insufficient detection reliability and generalization ability. Therefore, improving the accuracy, stability, adaptability, and efficiency of konjac irregularity slice detection has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for detecting irregularly shaped konjac slices to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for detecting irregularly shaped konjac slices, comprising:

[0006] P1. Obtain the initial image data of the target konjac slice under the excitation of a variable angle light source, and perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice.

[0007] P2. Perform edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image;

[0008] P3. Based on the contour abrupt change points, perform contour frequency domain quantization on the slice contour image to obtain the contour feature parameters of the slice contour image.

[0009] P4. Based on the contour feature parameters, perform inertial principal axis orientation on the slice contour image, and perform symmetry constraint correction on the initial principal axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice.

[0010] P5. Perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0011] P6. Perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

[0012] In a preferred embodiment, acquiring initial image data of the target konjac slice under variable-angle light source excitation, and performing morphological reconstruction of the slice foreground region in the initial image data to obtain the slice contour image of the target konjac slice, includes:

[0013] Acquire initial image data of the target konjac slices under illumination from light sources at different angles;

[0014] The pixels in the initial image data are encoded with texture features to construct a local texture feature map of the initial image data;

[0015] Morphological dilation reconstruction is performed on the local texture feature map to obtain a slice foreground mask of the local texture feature map;

[0016] Based on the slice foreground mask, background suppression mapping is performed on the initial image data to obtain a slice grayscale image of the initial image data;

[0017] The grayscale image of the slice is reconstructed using contour features to obtain the slice contour image of the target konjac slice.

[0018] In a preferred embodiment, performing edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image includes:

[0019] The contour direction of the slice contour image is interpreted to obtain the edge direction topology map of the slice contour image;

[0020] The local curvature response value of the topological node is obtained by sampling the curvature of the neighborhood of the topological node in the edge path topology graph.

[0021] Gradient transition detection is performed on the local curvature response value to obtain the curvature abrupt change label of the topological node;

[0022] Based on the curvature change label, the topological nodes are merged and identified to obtain the contour change points of the slice contour image.

[0023] In a preferred embodiment, the step of performing contour frequency domain quantization on the slice contour image based on the contour abrupt change points to obtain the contour feature parameters of the slice contour image includes:

[0024] Using the contour abrupt change points as segmentation anchor points, the slice contour image is segmented by anchor point segmentation to obtain the contour segment sequence of the slice contour image;

[0025] Radial distance encoding is performed on the contour segment sequence to obtain a one-dimensional distance waveform of the contour segment sequence;

[0026] The one-dimensional distance waveform is decomposed into a waveform spectrum to obtain the frequency domain amplitude distribution sequence of the contour segment sequence;

[0027] The main lobe features are extracted from the frequency domain amplitude distribution sequence to obtain the contour feature parameters of the slice contour image.

[0028] In a preferred embodiment, the step of performing waveform spectral decomposition on the one-dimensional distance waveform to obtain the frequency domain amplitude distribution sequence of the contour segment sequence includes:

[0029] Amplitude analysis is performed on the time-series sampling points of the one-dimensional distance waveform to obtain the amplitude sequence and amplitude sequence length of the one-dimensional distance waveform;

[0030] The amplitude sequence is subjected to discrete cosine spectral mapping to obtain the frequency domain mapped spectral coefficients of the amplitude sequence. The calculation formula for the frequency domain mapped spectral coefficients is as follows:

[0031] ;

[0032] in, Indicates the magnitude sequence of the first The amplitude of each sampling point Indicates the length of the amplitude sequence. This represents the frequency domain index of the frequency domain mapping spectral coefficients. Indicates the first The frequency domain mapping spectral coefficients. Represents the cosine function;

[0033] The frequency domain mapping spectral coefficients are rearranged and recombined in frequency order to obtain the frequency domain amplitude distribution sequence of the contour segment sequence.

[0034] In a preferred embodiment, the step of orienting the slice contour image by inertial principal axis based on the contour feature parameters, and correcting the initial principal axis direction after orienting by symmetry constraints to obtain the spatial attitude reference axis of the target konjac slice, includes:

[0035] Using the contour feature parameters as weighting factors, a weighted field tensor is constructed on the slice contour image to obtain the weighted gray-level distribution field of the slice contour image;

[0036] The initial principal axis pointing orientation of the slice contour image is obtained by performing inertial principal axis positioning on the weighted gray-level distribution field.

[0037] Based on the initial principal axis pointing direction, the slice contour image is divided into mirror half-regions to obtain the left and right contour half-regions of the initial principal axis pointing direction.

[0038] A topological deviation measurement is performed on the left and right contour halves to obtain the bilateral contour difference value of the initial principal axis pointing direction.

[0039] Based on the difference value of the two-sided contour, the initial principal axis pointing direction is finely adjusted by symmetry constraint, and the finely adjusted principal axis pointing direction is calibrated as the spatial attitude reference axis of the target konjac slice.

[0040] In a preferred embodiment, the step of performing inertial principal axis localization on the weighted grayscale distribution field to obtain the initial principal axis pointing orientation of the slice contour image includes:

[0041] The gray-level centroid of the weighted gray-level distribution field is calibrated to obtain the gray-level centroid coordinates of the weighted gray-level distribution field;

[0042] Using the gray-level centroid coordinates as the origin, the moment field tensor of the weighted gray-level distribution field is extracted to obtain the inertial moment tensor matrix of the weighted gray-level distribution field;

[0043] The eigenor orientation of the moment of inertia tensor matrix is ​​identified to obtain the eigenvector direction corresponding to the largest eigenvalue of the moment of inertia tensor matrix.

[0044] The direction of the feature vector is mapped to the image coordinate system of the slice contour image to obtain the initial principal axis pointing direction of the slice contour image.

[0045] In a preferred embodiment, the step of performing radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct a deformation distribution map of the edge points relative to the spatial attitude reference axis includes:

[0046] By performing normal field radiation on the spatial attitude reference axis, a cluster of normal rays of the spatial attitude reference axis is obtained;

[0047] The edge points of the slice contour image are projected onto the normal ray cluster to obtain the radial landing position of the edge points along the normal ray cluster;

[0048] The radial offset amplitude of the radial landing point position is measured to obtain the radial offset distance of the edge point relative to the spatial attitude reference axis;

[0049] Based on the traversal order of the edge points, the radial offset distance is continuously expanded and constructed to obtain the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0050] In a preferred embodiment, the step of performing deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice includes:

[0051] The deformation distribution map is encoded at the amplitude level to obtain the offset level distribution map of the deformation distribution map;

[0052] High-density domain mining is performed on the offset level distribution map to obtain high offset clusters of the deformation distribution map spectrum;

[0053] The high-offset clusters are subjected to morphological feature capture to obtain the maximum radial span within the clusters and the inter-cluster distribution spacing of the high-offset clusters.

[0054] Based on the maximum radial span within the cluster and the distribution spacing between the clusters, the topological configuration of the high-offset clusters is determined to obtain the irregularity category of the target konjac slice.

[0055] To address the above problems, the present invention also provides a konjac irregular slice detection system, the system comprising:

[0056] The foreground reconstruction module is used to acquire the initial image data of the target konjac slice under the excitation of a variable angle light source, and to perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice.

[0057] The mutation point identification module is used to perform edge curvature analysis on the slice contour image to identify the contour mutation points of the slice contour image.

[0058] The frequency domain quantization module is used to perform contour frequency domain quantization on the slice contour image based on the contour abrupt change point to obtain the contour feature parameters of the slice contour image.

[0059] The principal axis orientation module is used to perform inertial principal axis orientation on the slice contour image based on the contour feature parameters, and to perform symmetry constraint correction on the initial principal axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice.

[0060] The radial offset analysis module is used to perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis, so as to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0061] The deviation pattern recognition module is used to perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. This invention employs variable-angle light source excitation imaging combined with morphological reconstruction technology to completely extract the foreground region and contour information of konjac slices. It optimizes image quality through texture feature encoding and background suppression mapping, accurately stripping the slice contour and eliminating background interference. Based on edge curvature analysis, it accurately identifies contour abrupt change points, using these points as anchors to complete frequency domain quantization and discrete cosine spectrum mapping, generating standardized contour feature parameters. This significantly improves the accuracy of contour feature analysis and data processing efficiency, ensuring the integrity and stability of feature extraction, and providing standardized, highly reliable feature data support for subsequent detection processes.

[0064] 2. This invention achieves inertial principal axis orientation and symmetry constraint correction based on contour feature parameters, accurately establishing the spatial attitude reference axis of konjac slices. A comprehensive deformation distribution map is constructed through normal radial offset analysis, fully capturing the deformation information of the slice edges. High-density offset domain mining and topological configuration discrimination are completed through deviation pattern recognition, achieving automated and accurate classification of irregularly shaped konjac slices. This effectively improves the automation level and overall detection efficiency of the detection process, ensuring the consistency, stability, and accuracy of the detection results, meeting the detection application needs of continuous industrial production. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating a method for detecting irregularly shaped slices of konjac provided in an embodiment of the present invention.

[0066] Figure 2 This is a functional block diagram of a konjac irregular slice detection system provided in an embodiment of the present invention;

[0067] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0068] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0069] This application provides a method for detecting irregularly shaped konjac slices. The execution subject of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for detecting irregularly shaped konjac slices can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0070] Reference Figure 1 The diagram shown is a flowchart illustrating a method for detecting irregularly shaped konjac slices according to an embodiment of the present invention. In this embodiment, the method for detecting irregularly shaped konjac slices includes:

[0071] P1. Obtain the initial image data of the target konjac slice under the excitation of a variable angle light source, and perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice.

[0072] In this embodiment of the invention, the step of acquiring initial image data of the target konjac slice under variable angle light source excitation, and performing morphological reconstruction on the slice foreground region in the initial image data to obtain the slice contour image of the target konjac slice, includes:

[0073] Acquire initial image data of the target konjac slices under illumination from light sources at different angles;

[0074] The pixels in the initial image data are encoded with texture features to construct a local texture feature map of the initial image data;

[0075] Morphological dilation reconstruction is performed on the local texture feature map to obtain a slice foreground mask of the local texture feature map;

[0076] Based on the slice foreground mask, background suppression mapping is performed on the initial image data to obtain a slice grayscale image of the initial image data;

[0077] The grayscale image of the slice is reconstructed using contour features to obtain the slice contour image of the target konjac slice.

[0078] Multi-angle light sources are directionally deployed around the target konjac slice. Each light source is activated sequentially at a preset angle to illuminate the target konjac slice. An image acquisition device is used to fully image the target konjac slice under the illumination of each angle light source. All imaging results are integrated and collected to form the initial image data of the target konjac slice.

[0079] The process involves iterating through all pixels in the initial image data one by one, analyzing the texture distribution and brightness variation characteristics of each pixel within its surrounding neighborhood, encoding the features of each pixel according to the texture distribution rules, and then converting the initial image data into a local texture feature map based on the encoding results.

[0080] Using local texture feature maps as the processing object, a morphological dilation operation is performed to expand and reconstruct the slice foreground texture region in the image outward, fill the small gaps inside the foreground region, weaken the texture interference of the background region, accurately distinguish the slice foreground and background regions, and generate a slice foreground mask of local texture feature maps.

[0081] The slice foreground mask is matched pixel-level with the initial image data. The background pixel information in the initial image data is masked by the slice foreground mask, while the pixel information of the slice foreground area is preserved and converted into grayscale representation to generate a slice grayscale image of the initial image data.

[0082] The edge pixels of the konjac slice in the grayscale image are extracted in a specific direction. The extracted edge pixels are then subjected to continuity regularization and line enhancement processing to reconstruct the contour features of the slice edge and generate the slice contour image of the target konjac slice.

[0083] The beneficial effects are that the image information of konjac slices can be obtained comprehensively through multi-angle light source imaging. Combined with the step-by-step processing of texture feature encoding, morphological dilation reconstruction, background suppression mapping and contour feature reconstruction, the foreground and background of the slices can be accurately separated, and a clear and regular slice contour image can be completely extracted. The interference of background and irrelevant textures can be eliminated, ensuring the accuracy and integrity of the contour image, and providing high-quality basic data for subsequent contour analysis and irregularity detection.

[0084] P2. Perform edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image;

[0085] In this embodiment of the invention, the step of performing edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image includes:

[0086] The contour direction of the slice contour image is interpreted to obtain the edge direction topology map of the slice contour image;

[0087] The local curvature response value of the topological node is obtained by sampling the curvature of the neighborhood of the topological node in the edge path topology graph.

[0088] Gradient transition detection is performed on the local curvature response value to obtain the curvature abrupt change label of the topological node;

[0089] Based on the curvature change label, the topological nodes are merged and identified to obtain the contour change points of the slice contour image.

[0090] Starting from the starting point, the entire closed edge of the slice contour image is traced point by point. The extension direction of each contour point and the turning relationship between adjacent points are recorded. All contour points are transformed into an ordered combination of topological nodes according to the connection relationship. The connection path between each topological node and the overall contour direction structure are clarified, and a complete edge direction topology map that can reflect the contour extension and turning characteristics is constructed.

[0091] Taking each topological node in the edge-oriented topology graph as the core, several contour points adjacent to the node are selected to form a local contour segment. The curvature and curvature trend of the local contour segment are analyzed, and the curvature of the local contour segment is converted into the corresponding response result. A local curvature response value that can reflect the local curvature state is generated for each topological node.

[0092] According to the node order of the edge-oriented topology graph, the local curvature response values ​​of adjacent topology nodes are compared in turn to determine the magnitude of change between response values. When a sudden change occurs in the response value, a unique abrupt change identifier is marked for the topology node, and a curvature abrupt change label is assigned to each topology node with a curvature abrupt change.

[0093] Traverse all topological nodes with curvature change labels in the edge orientation topology graph, group the scattered label nodes together, remove invalid label nodes caused by small contour fluctuations, retain valid nodes that can truly reflect the contour morphology change, and finally screen and determine the contour change points of the slice contour image.

[0094] The beneficial effects are as follows: by interpreting the contour direction to construct an edge direction topology map, the overall structure of the contour is clarified; by sampling the curvature of the neighborhood, the accurate local curvature response values ​​of each topology node are obtained; by relying on gradient transition detection to identify curvature change nodes and complete the curvature change labeling; and by merging and filtering to remove invalid nodes, the contour change points of the slice contour image can be accurately located, and the morphological change positions of the contour edge can be completely captured, providing stable and accurate segmentation anchor points for subsequent contour frequency domain quantization, and comprehensively improving the accuracy and reliability of contour feature analysis.

[0095] P3. Based on the contour abrupt change points, perform contour frequency domain quantization on the slice contour image to obtain the contour feature parameters of the slice contour image.

[0096] In this embodiment of the invention, the step of performing contour frequency domain quantization on the slice contour image based on the contour abrupt change point to obtain the contour feature parameters of the slice contour image includes:

[0097] Using the contour abrupt change points as segmentation anchor points, the slice contour image is segmented by anchor point segmentation to obtain the contour segment sequence of the slice contour image;

[0098] Radial distance encoding is performed on the contour segment sequence to obtain a one-dimensional distance waveform of the contour segment sequence;

[0099] The one-dimensional distance waveform is decomposed into a waveform spectrum to obtain the frequency domain amplitude distribution sequence of the contour segment sequence;

[0100] The main lobe features are extracted from the frequency domain amplitude distribution sequence to obtain the contour feature parameters of the slice contour image.

[0101] The step of performing waveform spectral decomposition on the one-dimensional distance waveform to obtain the frequency domain amplitude distribution sequence of the contour segment sequence includes:

[0102] Amplitude analysis is performed on the time-series sampling points of the one-dimensional distance waveform to obtain the amplitude sequence and amplitude sequence length of the one-dimensional distance waveform;

[0103] The amplitude sequence is subjected to discrete cosine spectral mapping to obtain the frequency domain mapped spectral coefficients of the amplitude sequence. The calculation formula for the frequency domain mapped spectral coefficients is as follows:

[0104] ;

[0105] in, Indicates the magnitude sequence of the first The amplitude of each sampling point Indicates the length of the amplitude sequence. This represents the frequency domain index of the frequency domain mapping spectral coefficients. Indicates the first The frequency domain mapping spectral coefficients. Represents the cosine function;

[0106] The frequency domain mapping spectral coefficients are rearranged and recombined in frequency order to obtain the frequency domain amplitude distribution sequence of the contour segment sequence.

[0107] Using contour abrupt change points as segmentation anchors, the specific positions of each contour abrupt change point are located sequentially along the complete closed contour of the slice contour image. A segmentation operation is performed at the contour position corresponding to each contour abrupt change point, dividing the complete and continuous slice contour into multiple independent contour segments. All contour segments are arranged and combined in an orderly manner according to the original extension direction of the slice contour to form a complete and continuous set of segments, thus obtaining the contour segment sequence of the slice contour image.

[0108] The center position of the slice contour image is selected as a fixed radial reference center point. For each contour segment in the contour segment sequence, the straight-line distance values ​​from each point on the contour to the radial reference center point are collected point by point. All distance values ​​of each contour segment are arranged in the order of the contour points and converted into a continuous oscillating waveform. The waveforms corresponding to all contour segments are connected into a single complete waveform to obtain the one-dimensional distance waveform of the contour segment sequence.

[0109] According to the temporal arrangement of the one-dimensional distance waveform, the waveform amplitude value corresponding to each temporal sampling point on the waveform is extracted one by one. All the extracted amplitude values ​​are combined into a complete numerical sequence according to the temporal order. At the same time, the total number of all sampling points contained in the numerical sequence is counted to obtain the amplitude sequence and amplitude sequence length of the one-dimensional distance waveform.

[0110] Each amplitude value in the amplitude sequence is processed by domain transformation according to the rules of discrete cosine transform, converting the amplitude value in the spatial domain into characteristic coefficients in the frequency domain. By transforming point by point, a complete set of frequency domain characteristic coefficients is generated, thus obtaining the frequency domain mapping spectrum coefficients of the amplitude sequence.

[0111] The amplitude values ​​at corresponding sampling positions in the amplitude sequence are derived from the amplitude values ​​extracted from each sampling point after amplitude analysis of the time-series sampling points of the one-dimensional distance waveform. The total number of samples in the amplitude sequence is derived from the total number of sampling points obtained after amplitude analysis of the time-series sampling points of the one-dimensional distance waveform. The frequency domain index of the frequency domain mapping spectral coefficients corresponds to the amplitude sequence length. The cosine function is used to sequentially traverse all frequency domain components and assign a unique identifier position to each frequency domain mapping spectral coefficient. It is used to perform the spatial domain to frequency domain conversion operation and is the core operational rule for realizing domain conversion in the discrete cosine spectrum mapping process. This operation is used to perform the discrete cosine spectrum mapping operation, converting the amplitude sequence in the spatial domain into the mapping spectral coefficients in the frequency domain, realizing the slice contour morphology. In the quantization conversion operation from the spatial domain to the frequency domain, the amplitude of each sampling point in the amplitude sequence is multiplied by the corresponding cosine term, and then all multiplication results are accumulated to obtain the frequency domain mapping spectral coefficients of the corresponding frequency domain components. Each frequency domain mapping spectral coefficient corresponds to a specific frequency domain component, used to characterize the morphological features of the slice contour at that frequency, providing basic data for subsequent frequency sequence arrangement and reorganization. After the operation is completed, the generated frequency domain mapping spectral coefficients completely retain the morphological feature information of the slice contour, realizing the standardized frequency domain representation of the slice contour morphology.

[0112] As the frequency domain index of the frequency domain mapping spectral coefficients increases, the amplitude of the corresponding frequency domain mapping spectral coefficients shows a gradual decreasing trend. The low-frequency components correspond to the overall morphological features of the slice contour with highly concentrated energy. The high-frequency components correspond to the local details and morphological changes of the slice contour with a relatively low energy proportion, used to characterize the subtle changes in the contour. The smoother the overall shape of the slice contour, the higher the amplitude of the low-frequency components. The lower the amplitude of the high-frequency components, the more local changes in the slice contour. The higher the amplitude of the high-frequency components, the more stable the amplitude of the low-frequency components. The amplitude distribution trend of the frequency domain mapping spectral coefficients directly reflects the morphological feature distribution of the slice contour, providing a reliable basis for subsequent main lobe feature extraction.

[0113] Arrange all frequency domain mapping spectral coefficients in ascending order of frequency, sort and regularize them one by one, integrate the scattered frequency domain mapping spectral coefficients into a continuous and ordered sequence structure, maintain the integrity and regularity of frequency domain features, and obtain the frequency domain amplitude distribution sequence of the contour segment sequence.

[0114] In the frequency domain amplitude distribution sequence, the main lobe region with the most concentrated energy is located, and the amplitude distribution pattern, distribution trend and core distribution features within the main lobe region are extracted. All extracted core features are integrated and summarized to form a standardized parameter set that can characterize the overall frequency domain features of the contour, thus obtaining the contour feature parameters of the slice contour image.

[0115] The beneficial effects are that the contour abrupt change points are used as segmentation anchor points to complete the accurate segmentation of the slice contour. The contour shape is transformed into a standardized one-dimensional distance waveform through radial distance encoding. After amplitude analysis, discrete cosine spectrum mapping and frequency sequence arrangement and recombination, a regular frequency domain amplitude distribution sequence is obtained. Then, contour feature parameters are generated by extracting the main lobe features, realizing the accurate quantitative conversion of the slice contour shape from the spatial domain to the frequency domain. The complex contour shape is transformed into stable and unified contour feature parameters, providing accurate and reliable feature data support for subsequent inertial principal axis orientation, and improving the stability and standardization of contour feature analysis.

[0116] P4. Based on the contour feature parameters, perform inertial principal axis orientation on the slice contour image, and perform symmetry constraint correction on the initial principal axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice.

[0117] In this embodiment of the invention, the step of orienting the slice contour image by inertial principal axis based on the contour feature parameters, and correcting the initial principal axis direction after orientation by symmetry constraint to obtain the spatial attitude reference axis of the target konjac slice, includes:

[0118] Using the contour feature parameters as weighting factors, a weighted field tensor is constructed on the slice contour image to obtain the weighted gray-level distribution field of the slice contour image;

[0119] The initial principal axis pointing orientation of the slice contour image is obtained by performing inertial principal axis positioning on the weighted gray-level distribution field.

[0120] Based on the initial principal axis pointing direction, the slice contour image is divided into mirror half-regions to obtain the left and right contour half-regions of the initial principal axis pointing direction.

[0121] A topological deviation measurement is performed on the left and right contour halves to obtain the bilateral contour difference value of the initial principal axis pointing direction.

[0122] Based on the difference value of the two-sided contour, the initial principal axis pointing direction is finely adjusted by symmetry constraint, and the finely adjusted principal axis pointing direction is calibrated as the spatial attitude reference axis of the target konjac slice.

[0123] The step of performing inertial principal axis positioning on the weighted grayscale distribution field to obtain the initial principal axis pointing orientation of the slice contour image includes:

[0124] The gray-level centroid of the weighted gray-level distribution field is calibrated to obtain the gray-level centroid coordinates of the weighted gray-level distribution field;

[0125] Using the gray-level centroid coordinates as the origin, the moment field tensor of the weighted gray-level distribution field is extracted to obtain the inertial moment tensor matrix of the weighted gray-level distribution field;

[0126] The eigenor orientation of the moment of inertia tensor matrix is ​​identified to obtain the eigenvector direction corresponding to the largest eigenvalue of the moment of inertia tensor matrix.

[0127] The direction of the feature vector is mapped to the image coordinate system of the slice contour image to obtain the initial principal axis pointing direction of the slice contour image.

[0128] Contour feature parameters are used as weighting factors and matched one by one to each pixel of the slice contour image. The original gray value of each pixel is weighted and adjusted according to the magnitude of the contour feature parameters to highlight the gray value of key feature areas of the contour and weaken the gray value interference of non-feature areas. A complete field tensor structure is constructed based on the weighted gray value information. The weighted gray values ​​of all pixels are integrated with the tensor information to form a continuous and unified distribution field, thus obtaining the weighted gray value distribution field of the slice contour image.

[0129] Iterate through all pixels in the weighted grayscale distribution field one by one, record the weighted grayscale value of each pixel and its two-dimensional spatial position in the image, perform comprehensive calculation on the weighted grayscale values ​​and spatial positions of all pixels to determine the mass center position of the entire weighted grayscale distribution field, and accurately calibrate the spatial coordinates of the center position to obtain the grayscale centroid coordinates of the weighted grayscale distribution field.

[0130] Using the gray-scale centroid coordinates as the origin of the entire analysis process, the horizontal and vertical offsets of each pixel point relative to the origin, as well as the corresponding weighted gray-scale information, are extracted within the weighted gray-scale distribution field. Based on this information, a tensor structure is constructed to describe the inertial characteristics of the distribution field. All tensor data are combined according to fixed rules to form a complete matrix form, thus obtaining the inertial moment tensor matrix of the weighted gray-scale distribution field.

[0131] The moment of inertia tensor matrix is ​​subjected to eigenvalue decomposition, which extracts all eigenvalues ​​and their corresponding eigenvectors one by one. The magnitudes of all eigenvalues ​​are compared, the eigenvalue with the largest value is identified, and the extension direction of the eigenvector associated with the largest eigenvalue in space is determined, thus obtaining the direction of the eigenvector corresponding to the largest eigenvalue of the moment of inertia tensor matrix.

[0132] The direction of the eigenvector corresponding to the largest eigenvalue is precisely transformed according to the image coordinate system rules of the slice contour image. The abstract eigenvector direction is mapped to the actual pixel space of the slice contour image, and the specific direction and angle of the direction in the image are determined to obtain the initial principal axis direction of the slice contour image.

[0133] Using the initial principal axis as the symmetrical dividing line, the slice contour image is divided into two symmetrical regions along the complete extension path of the initial principal axis. All konjac slice contour information contained in the two regions is extracted to form the left and right contour halves of the initial principal axis.

[0134] The contour points of the left and right contour halves are matched one by one according to their symmetrical positions. The contour shape and position deviation of each matched point are compared. The deviation information of all matched points is summarized and comprehensively calculated to form a numerical result that reflects the degree of symmetry of the two contours, and the difference value of the two contours in the direction of the initial principal axis is obtained.

[0135] Using the difference value of the two-sided contour as the adjustment basis, the initial main axis pointing direction is continuously deflected by small angles. After each adjustment, the difference value of the two-sided contour is recalculated to continuously reduce the degree of deviation of the two-sided contour until the difference value of the two-sided contour reaches a stable minimum state. The main axis pointing direction at this time is finally calibrated to obtain the spatial attitude reference axis of the target konjac slice.

[0136] The beneficial effects are that constructing a weighted grayscale distribution field with contour feature parameters as weights can enhance the expression of the core features of the contour and improve the feature recognition of the grayscale distribution field. By calibrating the grayscale centroid, a precise analysis origin is determined. The inertial principal axis is initially accurately located through moment field tensor extraction and intrinsic orientation identification. Then, the symmetrical comparison of the two-sided contours is achieved through mirror half-region division. The orientation deviation of the initial principal axis is quantified by relying on topological deviation measurement. The orientation error of the initial principal axis is completely eliminated through symmetric constraint fine-tuning. Finally, a stable and accurate spatial attitude reference axis is obtained, which provides a unified and standard attitude reference for subsequent radial offset analysis. This avoids detection errors caused by attitude positioning deviations from the root cause and comprehensively improves the orientation accuracy and stability of the detection results of irregular konjac slices.

[0137] P5. Perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0138] In this embodiment of the invention, the step of performing radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct a deformation distribution map of the edge points relative to the spatial attitude reference axis includes:

[0139] By performing normal field radiation on the spatial attitude reference axis, a cluster of normal rays of the spatial attitude reference axis is obtained;

[0140] The edge points of the slice contour image are projected onto the normal ray cluster to obtain the radial landing position of the edge points along the normal ray cluster;

[0141] The radial offset amplitude of the radial landing point position is measured to obtain the radial offset distance of the edge point relative to the spatial attitude reference axis;

[0142] Based on the traversal order of the edge points, the radial offset distance is continuously expanded and constructed to obtain the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0143] Along the entire extended axis of the spatial attitude reference axis, select each point on the axis uniformly from the starting position to the ending position. Using each selected point as the emission center, emit rays synchronously to the left and right sides perpendicular to the spatial attitude reference axis. Control all rays to maintain a uniform radiation interval, so that the rays completely cover the entire edge area of ​​the slice contour image. All emitted perpendicular rays are connected to form a dense and continuous set of radiation rays, resulting in the normal ray cluster of the spatial attitude reference axis.

[0144] Extract all continuously arranged edge points on the slice contour image. For each independent edge point, locate the normal ray closest to that edge point. Project the edge point perpendicularly onto this normal ray to determine the perpendicular intersection point of the edge point and the normal ray. This point is the precise stopping position of the edge point on the normal ray. The projection points of all edge points together constitute a complete set of projection positions, and the radial landing position of the edge point along the cluster of normal rays is obtained.

[0145] The starting point of the measurement is the point on the spatial attitude reference axis from which the corresponding normal ray is emitted, and the ending point of the measurement is the radial landing point after the edge point is projected. The straight line length between the two points is directly measured. This length value directly represents the actual magnitude of the edge point's deviation from the spatial attitude reference axis. A corresponding length value is generated for each edge point to obtain the radial offset distance of the edge point relative to the spatial attitude reference axis.

[0146] Following the original traversal order of the edge points in the slice contour image, the radial offset distance values ​​corresponding to all edge points are sequentially connected and combined to transform the discrete offset distance values ​​into a continuous distribution pattern, fully presenting the offset size and overall distribution pattern of each edge point, forming a visual map that can intuitively display the contour deformation, and obtaining the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0147] The beneficial effects are that a cluster of normal rays with full coverage and no omissions is constructed by radiating the normal field of the spatial attitude reference axis, providing a stable reference carrier for edge point projection. The radial landing position of each edge point is accurately determined by the vertical projection of the edge points, ensuring the consistency of the reference for offset measurement. The actual deformation degree of each edge point is quantified by accurately measuring the offset amplitude. Then, the deformation distribution map is continuously constructed according to the traversal order of the edge points. It can completely capture all the global and local deformation information of the slice contour, and intuitively and clearly present the deformation amplitude and distribution law of the slice contour. It provides comprehensive, accurate and visualized deformation data support for subsequent deviation pattern recognition, and greatly improves the detail capture capability and data reliability of irregularity detection.

[0148] P6. Perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

[0149] In this embodiment of the invention, the step of performing deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice includes:

[0150] The deformation distribution map is encoded at the amplitude level to obtain the offset level distribution map of the deformation distribution map;

[0151] High-density domain mining is performed on the offset level distribution map to obtain high offset clusters of the deformation distribution map spectrum;

[0152] The high-offset clusters are subjected to morphological feature capture to obtain the maximum radial span within the clusters and the inter-cluster distribution spacing of the high-offset clusters.

[0153] Based on the maximum radial span within the cluster and the distribution spacing between the clusters, the topological configuration of the high-offset clusters is determined to obtain the irregularity category of the target konjac slice.

[0154] The radial offset amplitude information corresponding to all edge points in the deformation distribution map is traversed. All radial offset amplitudes are classified in an orderly manner according to the preset fixed amplitude division rules. A unique code identifier is assigned to the radial offset amplitude of each level. The code identifier is accurately matched and bound to the corresponding edge point position in the deformation distribution map. The discrete offset amplitude is transformed into a layered and visualized graphic distribution form, which fully presents the distribution position and distribution state of different offset amplitudes on the slice contour, and obtains the offset level distribution map of the deformation distribution map.

[0155] Locate all points corresponding to high coding levels in the offset level distribution map, track the spatial distribution of these high coding level points, identify the range of areas where points are continuously concentrated, independently delineate and integrate each continuous concentrated high coding level point to form a complete set of regions aggregated from high offset points, accurately lock the core area with prominent deformation on the slice outline, and obtain the high offset cluster of the deformation distribution map.

[0156] For each independent high-offset cluster, the radial distribution coordinate information of all points within the cluster is extracted, the farthest interval between radial coordinates within the cluster is determined, and the center coordinate information of different high-offset clusters is extracted. The interval between the center coordinates of any two adjacent high-offset clusters is determined. The farthest interval between each high-offset cluster and the interval between clusters are recorded respectively, so as to obtain the maximum radial span within the high-offset cluster and the distribution spacing between clusters.

[0157] Using the maximum radial span within a cluster and the inter-cluster distribution spacing as the core criteria, the spatial arrangement and overall morphological characteristics of high-offset clusters are analyzed. The topological configuration features obtained from the analysis are precisely matched and compared with the preset standard configurations of various irregular slices. Based on the matching configuration standards, the irregular type corresponding to the target konjac slice is determined, and the final determination of the irregular category is completed, resulting in the determination result of the irregular category of the target konjac slice.

[0158] The beneficial effects are that by encoding the deformation distribution map at the amplitude level, the abstract offset amplitude can be transformed into an intuitive, layered offset level distribution map, clearly distinguishing different degrees of deformation areas. Through high-density domain mining, high offset clustering areas on the slice outline can be accurately located, quickly locking the core location of irregular defects. By capturing the maximum radial span within the high offset clusters and the distribution spacing between clusters, the core morphological features of irregular defects can be completely extracted. Based on topological configuration discrimination, the automatic and accurate matching and judgment of irregular categories can be achieved without manual intervention throughout the process. This significantly improves the automation level and judgment accuracy of konjac irregular slice detection, ensuring the objectivity and consistency of detection results, and meeting the needs of efficient and accurate irregular detection in industrial continuous production.

[0159] like Figure 2 The diagram shown is a functional block diagram of a konjac irregular slice detection system provided in an embodiment of the present invention.

[0160] The konjac irregular slice detection system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the konjac irregular slice detection system 100 may include a foreground reconstruction module 101, a mutation point identification module 102, a frequency domain quantization module 103, a principal axis orientation module 104, a radial offset analysis module 105, and a deviation pattern recognition module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0161] In this embodiment, the functions of each module / unit are as follows:

[0162] The foreground reconstruction module 101 is used to acquire initial image data of the target konjac slice under variable angle light source excitation, and to perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice.

[0163] The mutation point identification module 102 is used to perform edge curvature analysis on the slice contour image to identify the contour mutation points of the slice contour image.

[0164] The frequency domain quantization module 103 is used to perform contour frequency domain quantization on the slice contour image based on the contour abrupt change point to obtain the contour feature parameters of the slice contour image.

[0165] The main axis orientation module 104 is used to perform inertial main axis orientation on the slice contour image based on the contour feature parameters, and to perform symmetry constraint correction on the initial main axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice.

[0166] The radial offset analysis module 105 is used to perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis, so as to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis.

[0167] The deviation pattern recognition module 106 is used to perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

[0168] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0169] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0170] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0171] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0172] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for detecting irregularly shaped pieces of konjac, characterized by, The method includes: P1. Obtain the initial image data of the target konjac slice under the excitation of a variable angle light source, and perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice. P2. Perform edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image; P3. Based on the contour abrupt change points, perform contour frequency domain quantization on the slice contour image to obtain the contour feature parameters of the slice contour image, including: Using the contour abrupt change points as segmentation anchor points, the slice contour image is segmented by anchor point segmentation to obtain the contour segment sequence of the slice contour image; Radial distance encoding is performed on the contour segment sequence to obtain a one-dimensional distance waveform of the contour segment sequence; The one-dimensional distance waveform is decomposed into a waveform spectrum to obtain the frequency domain amplitude distribution sequence of the contour segment sequence; The main lobe features of the frequency domain amplitude distribution sequence are extracted to obtain the contour feature parameters of the slice contour image; P4. Based on the contour feature parameters, perform inertial principal axis orientation on the slice contour image, and perform symmetry constraint correction on the initial principal axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice, including: Using the contour feature parameters as weighting factors, a weighted field tensor is constructed on the slice contour image to obtain the weighted gray-level distribution field of the slice contour image; Contour feature parameters are used as weighting factors and matched one by one to each pixel of the slice contour image. The original gray value of each pixel is weighted and adjusted according to the magnitude of the contour feature parameters to highlight the gray value of the key feature area of ​​the contour and weaken the gray value interference of non-feature area. Based on the weighted gray value information, a complete field tensor structure is constructed. The weighted gray values ​​of all pixels are integrated with the tensor information to form a continuous and unified distribution field, thus obtaining the weighted gray value distribution field of the slice contour image. The initial principal axis pointing orientation of the slice contour image is obtained by performing inertial principal axis positioning on the weighted gray-level distribution field. Based on the initial principal axis pointing direction, the slice contour image is divided into mirror half-regions to obtain the left and right contour half-regions of the initial principal axis pointing direction. A topological deviation measurement is performed on the left and right contour halves to obtain the bilateral contour difference value of the initial principal axis pointing direction. Based on the difference value of the two-sided contour, the initial principal axis pointing direction is finely adjusted by symmetry constraint, and the finely adjusted principal axis pointing direction is calibrated as the spatial attitude reference axis of the target konjac slice. P5. Perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis. P6. Perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

2. The method of claim 1, wherein the step of detecting the shape of the konjak slice is performed by using a camera. The process of acquiring initial image data of the target konjac slice under variable-angle light source excitation and performing morphological reconstruction on the slice foreground region in the initial image data to obtain the slice contour image of the target konjac slice includes: Acquire initial image data of the target konjac slices under illumination from light sources at different angles; The pixels in the initial image data are encoded with texture features to construct a local texture feature map of the initial image data; Morphological dilation reconstruction is performed on the local texture feature map to obtain a slice foreground mask of the local texture feature map; Based on the slice foreground mask, background suppression mapping is performed on the initial image data to obtain a slice grayscale image of the initial image data; The grayscale image of the slice is reconstructed using contour features to obtain the slice contour image of the target konjac slice.

3. The method of claim 1, wherein the step of detecting the shape of the konjak slice is performed by using a camera. The step of performing edge curvature analysis on the slice contour image to identify contour abrupt change points in the slice contour image includes: The contour direction of the slice contour image is interpreted to obtain the edge direction topology map of the slice contour image; The local curvature response value of the topological node is obtained by sampling the curvature of the neighborhood of the topological node in the edge path topology graph. Gradient transition detection is performed on the local curvature response value to obtain the curvature abrupt change label of the topological node; Based on the curvature change label, the topological nodes are merged and identified to obtain the contour change points of the slice contour image.

4. The method of claim 1, wherein the step of detecting the shape of the konjak slice is performed by using a camera. The step of performing waveform spectral decomposition on the one-dimensional distance waveform to obtain the frequency domain amplitude distribution sequence of the contour segment sequence includes: Amplitude analysis is performed on the time-series sampling points of the one-dimensional distance waveform to obtain the amplitude sequence and amplitude sequence length of the one-dimensional distance waveform; The amplitude sequence is subjected to discrete cosine spectral mapping to obtain the frequency domain mapped spectral coefficients of the amplitude sequence. The calculation formula for the frequency domain mapped spectral coefficients is as follows: ； wherein denotes an amplitude of a k-th sample point of the amplitude sequence, denotes a length of the amplitude sequence, denotes a frequency domain index of the frequency domain mapped spectral coefficient, denotes a k-th frequency domain mapped spectral coefficient, denotes a cosine function;​​ The frequency domain mapping spectral coefficients are rearranged and recombined in frequency order to obtain the frequency domain amplitude distribution sequence of the contour segment sequence.

5. The method of claim 1, wherein the step of detecting the shape of the konjak slice is performed by using a camera. The step of performing inertial principal axis positioning on the weighted grayscale distribution field to obtain the initial principal axis pointing orientation of the slice contour image includes: The gray-level centroid of the weighted gray-level distribution field is calibrated to obtain the gray-level centroid coordinates of the weighted gray-level distribution field; Using the gray-level centroid coordinates as the origin, the moment field tensor of the weighted gray-level distribution field is extracted to obtain the inertial moment tensor matrix of the weighted gray-level distribution field; The eigenor orientation of the moment of inertia tensor matrix is ​​identified to obtain the eigenvector direction corresponding to the largest eigenvalue of the moment of inertia tensor matrix. The direction of the feature vector is mapped to the image coordinate system of the slice contour image to obtain the initial principal axis pointing direction of the slice contour image.

6. The method of claim 1, wherein the step of detecting the shape of the konjak slice is performed by using a camera. The step of performing radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis to construct a deformation distribution map of the edge points relative to the spatial attitude reference axis includes: By performing normal field radiation on the spatial attitude reference axis, a cluster of normal rays of the spatial attitude reference axis is obtained; The edge points of the slice contour image are projected onto the normal ray cluster to obtain the radial landing position of the edge points along the normal ray cluster; The radial offset amplitude of the radial landing point position is measured to obtain the radial offset distance of the edge point relative to the spatial attitude reference axis; Based on the traversal order of the edge points, the radial offset distance is continuously expanded and constructed to obtain the deformation distribution map of the edge points relative to the spatial attitude reference axis.

7. The method for detecting irregularly shaped konjac slices as described in claim 1, characterized in that, The step of performing deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice includes: The deformation distribution map is encoded at the amplitude level to obtain the offset level distribution map of the deformation distribution map; High-density domain mining is performed on the offset level distribution map to obtain high offset clusters of the deformation distribution map spectrum; The high-offset clusters are subjected to morphological feature capture to obtain the maximum radial span within the clusters and the inter-cluster distribution spacing of the high-offset clusters. Based on the maximum radial span within the cluster and the distribution spacing between the clusters, the topological configuration of the high-offset clusters is determined to obtain the irregularity category of the target konjac slice.

8. A konjac heteromorphic slice detection system, characterized in that, The system for implementing the method for detecting irregular slices of konjac as described in claim 1 includes: The foreground reconstruction module is used to acquire the initial image data of the target konjac slice under the excitation of a variable angle light source, and to perform morphological reconstruction on the slice foreground region in the initial image data to obtain the slice outline image of the target konjac slice. The mutation point identification module is used to perform edge curvature analysis on the slice contour image to identify the contour mutation points of the slice contour image. A frequency domain quantization module is used to perform contour frequency domain quantization on the slice contour image based on the contour abrupt change points to obtain the contour feature parameters of the slice contour image, including: Using the contour abrupt change points as segmentation anchor points, the slice contour image is segmented by anchor point segmentation to obtain the contour segment sequence of the slice contour image; Radial distance encoding is performed on the contour segment sequence to obtain a one-dimensional distance waveform of the contour segment sequence; The one-dimensional distance waveform is decomposed into a waveform spectrum to obtain the frequency domain amplitude distribution sequence of the contour segment sequence; The main lobe features of the frequency domain amplitude distribution sequence are extracted to obtain the contour feature parameters of the slice contour image; The principal axis orientation module is used to perform inertial principal axis orientation on the slice contour image based on the contour feature parameters, and to perform symmetry constraint correction on the initial principal axis direction after orientation to obtain the spatial attitude reference axis of the target konjac slice, including: Using the contour feature parameters as weighting factors, a weighted field tensor is constructed on the slice contour image to obtain the weighted gray-level distribution field of the slice contour image; Contour feature parameters are used as weighting factors and matched one by one to each pixel of the slice contour image. The original gray value of each pixel is weighted and adjusted according to the magnitude of the contour feature parameters to highlight the gray value of the key feature area of ​​the contour and weaken the gray value interference of non-feature area. Based on the weighted gray value information, a complete field tensor structure is constructed. The weighted gray values ​​of all pixels are integrated with the tensor information to form a continuous and unified distribution field, thus obtaining the weighted gray value distribution field of the slice contour image. The initial principal axis pointing orientation of the slice contour image is obtained by performing inertial principal axis positioning on the weighted gray-level distribution field. Based on the initial principal axis pointing direction, the slice contour image is divided into mirror half-regions to obtain the left and right contour half-regions of the initial principal axis pointing direction. A topological deviation measurement is performed on the left and right contour halves to obtain the bilateral contour difference value of the initial principal axis pointing direction. Based on the difference value of the two-sided contour, the initial principal axis pointing direction is finely adjusted by symmetry constraint, and the finely adjusted principal axis pointing direction is calibrated as the spatial attitude reference axis of the target konjac slice. The radial offset analysis module is used to perform radial offset analysis on the edge points of the slice contour image along the normal direction of the spatial attitude reference axis, so as to construct the deformation distribution map of the edge points relative to the spatial attitude reference axis. The deviation pattern recognition module is used to perform deviation pattern recognition on the deformation distribution map to obtain the irregularity category determination result of the target konjac slice.

Images