A sauce braised beef color uniformity determination method based on image recognition

Abstract

The present application relates to food detection technology field, disclose a kind of sauce halogen beef color uniformity determination method based on image recognition, method includes: the color feature of original sample image is characterized, obtains benchmark color feature vector;Benchmark color feature vector is reconstructed to color space, obtains the multi-dimensional feature space of original sample image;Pixel point segmentation is carried out to multi-dimensional feature space, obtains the texture pixel set and background pixel set of multi-dimensional feature space;Color feature statistics is carried out to texture pixel set and background pixel set, and demarcation uniformity determination threshold surface;Pixel point feature vector in multi-dimensional feature space is compared with the space mapping of uniformity determination threshold surface, obtains the qualified pixel point and defective pixel point of original sample image;The color uniformity of sauce halogen beef is comprehensively judged, obtains the color uniformity evaluation index of sauce halogen beef;The present application can improve the efficiency of a kind of sauce halogen beef color uniformity determination based on image recognition.

Description

Technical Field

[0001] This invention relates to the field of food testing technology, and in particular to a method for determining the color uniformity of braised beef based on image recognition. Background Technology

[0002] The color uniformity of braised beef is one of the core indicators for its quality inspection. Currently, the industry still mainly relies on manual visual inspection to determine the color uniformity of braised beef. This method lacks a unified quantitative standard and depends entirely on the color perception ability and experience of the inspectors. It is easily affected by factors such as subjective judgment and visual fatigue. Not only can it not accurately analyze and quantify the color characteristics of the beef surface, but it can also lead to individual differences and batch deviations in the judgment results. Furthermore, manual inspection takes a long time to process a single sample, resulting in low inspection efficiency. It is difficult to meet the needs of large-scale, streamlined production and inspection of braised beef, and it cannot provide accurate color data support for optimizing the production process.

[0003] Existing image recognition-based methods for determining the color uniformity of braised beef only perform basic color channel extraction and simple pixel feature analysis on the acquired beef images. They fail to incorporate the texture of the braised beef for precise pixel segmentation, easily leading to confusion between beef texture pixels and background pixels. Furthermore, they lack multi-dimensional spatial reconstruction of color features, failing to deeply explore the spatial distribution patterns of color features in the image. The determination of the uniformity threshold lacks scientific feature distribution boundary support, and the pixel discrimination results are not corrected for neighborhood consistency or analyzed for spatial distribution clustering. This results in low accuracy in identifying color defect areas, a high rate of false negatives and false positives, and the judgment results only reflect the surface color proportions, failing to accurately characterize the actual uniformity of the beef color. The accuracy and reliability of the detection do not meet industrial-grade quality inspection requirements. Therefore, improving the standardization, quantification, accuracy, and efficiency of color uniformity determination for braised beef has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method for determining the color uniformity of braised beef based on image recognition, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for determining the color uniformity of braised beef based on image recognition, comprising:

[0006] S1. Acquire original sample images of braised beef under a standard light source, and characterize the color features of the original sample images to obtain the reference color feature vector of the original sample images;

[0007] S2. Reconstruct the color space of the reference color feature vector to obtain the multidimensional feature space of the original sample image;

[0008] S3. Based on the beef texture direction of the original sample image, the multidimensional feature space is segmented into pixels to obtain the texture pixel set and background pixel set of the multidimensional feature space.

[0009] S4. Perform color feature statistics on the texture pixel set and the background pixel set, and delineate the uniformity judgment threshold surface based on the feature distribution boundary in the multidimensional feature space according to the statistical results.

[0010] S5. Perform spatial mapping comparison between the pixel feature vectors in the multidimensional feature space and the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image.

[0011] S6. Based on the ratio and spatial distribution of the qualified and defective pixels, the color uniformity of the braised beef is comprehensively judged to obtain the color uniformity evaluation index of the braised beef.

[0012] In a preferred embodiment, the process of acquiring original sample images of braised beef under a standard light source and performing color feature characterization on the original sample images to obtain a reference color feature vector for the original sample images includes:

[0013] Original sample images of braised beef are collected, and color feature deconstruction is performed on the original sample images to obtain the pixel color components of the original sample images. The pixel color components include red channel components, green channel components and blue channel components.

[0014] The color components of the pixels are mapped to equal interval quantization intervals, and the mapping results are frequency-collected to obtain the color frequency distribution sequence of the original sample image.

[0015] Weighted feature aggregation is performed on the color frequency distribution sequence to obtain the color feature normalization value of the original sample image;

[0016] The color feature values ​​are vectorized and arranged according to the channel order of the color components of the pixels to obtain the reference color feature vector of the original sample image.

[0017] In a preferred embodiment, the step of reconstructing the reference color feature vector into a color space to obtain the multidimensional feature space of the original sample image includes:

[0018] The dimensional feature values ​​of the reference color feature vector are mapped to coordinates to obtain the initial feature coordinate point set of the original sample image;

[0019] Density clustering is performed on the initial set of feature coordinate points to obtain feature clusters of the original sample image;

[0020] The cluster centroids of the feature clusters are extracted to obtain the cluster centroid points of the feature clusters;

[0021] The feature space coordinate system of the original sample image is constructed by using the direction vector of the centroid of the cluster as the spatial basis vector of the original sample image.

[0022] Based on the feature space coordinate system, the initial feature coordinate point set is spatially relocated to obtain the multidimensional feature space of the original sample image.

[0023] In a preferred embodiment, the step of spatially relocating the initial set of feature coordinate points based on the feature space coordinate system to obtain the multidimensional feature space of the original sample image includes:

[0024] Based on the feature space coordinate system, the initial coordinate points in the initial feature coordinate point set are orthogonally projected and decomposed to obtain the projected coordinate values ​​of the initial coordinate points.

[0025] Based on the spatial neighborhood relationship between pixels in the original sample image, spatial consistency constraints are applied to the projected coordinate values ​​to obtain the corrected projected coordinate values ​​of the initial coordinate points.

[0026] The corrected projected coordinate values ​​are vectorized and assembled to obtain the repositioning coordinate vector of the initial coordinate point;

[0027] The repositioning coordinate vectors are arranged in a spatial structure to obtain the multidimensional feature space of the original sample image.

[0028] In a preferred embodiment, the step of segmenting the multidimensional feature space based on the beef texture direction of the original sample image to obtain a texture pixel set and a background pixel set in the multidimensional feature space includes:

[0029] The original sample image is subjected to texture direction analysis to obtain the texture direction features of the original pixels in the original sample image.

[0030] The texture direction features are mapped to the corresponding pixels in the multidimensional feature space to obtain the feature pixels in the multidimensional feature space.

[0031] Based on the texture direction features, the feature pixels are clustered according to direction consistency to obtain the texture candidate clusters in the multidimensional feature space;

[0032] The texture candidate cluster is expanded by neighborhood growth to obtain the texture pixel set of the multidimensional feature space;

[0033] The feature pixels outside the texture pixel set in the multidimensional feature space are classified into background domains to obtain the background pixel set.

[0034] In a preferred embodiment, the step of performing color feature statistics on the texture pixel set and the background pixel set, and defining a uniformity determination threshold surface based on the feature distribution boundary in the multidimensional feature space according to the statistical results, includes:

[0035] Based on the coordinate positions in the multidimensional feature space, feature vector quantization is performed on the texture pixel set and the background pixel set to obtain the texture feature value of the texture pixel set and the background feature value of the background pixel set;

[0036] The distribution centers of the texture feature values ​​and the background feature values ​​are located to obtain the texture distribution center point and the background distribution center point of the texture pixel set and the background pixel set in the multidimensional feature space.

[0037] The texture distribution center point and the background distribution center point are spatially divided and positioned to obtain spatially equidistant points between the texture distribution center point and the background distribution center point;

[0038] Centered on the spatial equidistant points, a boundary surface is fitted to the distribution boundary region of the texture feature values ​​and the background feature values ​​in the multidimensional feature space to obtain the uniformity determination threshold surface of the multidimensional feature space.

[0039] In a preferred embodiment, the step of fitting a boundary surface to the distribution boundary region of the texture feature values ​​and the background feature values ​​in the multidimensional feature space, centered on the spatially equidistant points, to obtain the uniformity threshold surface of the multidimensional feature space, includes:

[0040] Convex hull surfaces are constructed on the spatial points of the texture feature values ​​in the multidimensional feature space to obtain the texture distribution convex hull surface of the texture feature values.

[0041] An envelope surface is generated for the spatial points of the background feature values ​​in the multidimensional feature space to obtain the background distribution envelope surface of the background feature values.

[0042] The boundary lines of the spatial intersection region of the texture distribution convex hull surface and the background distribution envelope surface in the multidimensional feature space are extracted to obtain the distribution boundary contour lines of the texture feature values ​​and the background feature values.

[0043] Centered on the equidistant points in space, the distribution boundary contour is extended to obtain the uniformity determination threshold surface of the multidimensional feature space.

[0044] In a preferred embodiment, the step of spatially mapping and comparing the pixel feature vectors in the multidimensional feature space with the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image includes:

[0045] The feature vectors of the pixels in the multidimensional feature space are parsed to obtain the feature vectors to be discriminated for the pixels.

[0046] Based on the uniformity threshold surface, the spatial orientation of the feature vector to be judged is determined to obtain the preliminary discrimination label of the pixel.

[0047] The preliminary discrimination label is subjected to neighborhood consistency analysis, and the preliminary discrimination label is corrected and iterated according to the analysis results to obtain the final discrimination label of the pixel.

[0048] Based on the final discrimination label, the original pixels of the original sample image are identified and divided to obtain the qualified pixels and defective pixels of the original sample image.

[0049] In a preferred embodiment, the step of performing neighborhood consistency analysis on the preliminary discrimination label and correcting and iterating the preliminary discrimination label based on the analysis results to obtain the final discrimination label of the pixel includes:

[0050] The neighborhood range of the pixel in the multidimensional feature space is defined to obtain the spatial neighborhood pixels of the pixel.

[0051] The frequency of tags in the spatial neighborhood pixels is statistically analyzed to obtain the percentage of qualified pixels and the percentage of defective pixels in the spatial neighborhood pixels.

[0052] Based on the proportion of qualified frequencies and the proportion of defective frequencies, a confidence value is assigned to the preliminary discrimination label to obtain the label confidence coefficient of the pixel.

[0053] Based on the label confidence coefficient, the preliminary discrimination label is weighted and corrected to obtain the corrected discrimination label of the pixel;

[0054] The modified discrimination label is iteratively converged to obtain the final discrimination label of the pixel.

[0055] In a preferred embodiment, the step of comprehensively judging the color uniformity of the braised beef based on the ratio and spatial distribution aggregation of the qualified pixels and the defective pixels to obtain the color uniformity evaluation index of the braised beef includes:

[0056] Connectivity components are labeled for the qualified pixels and the defective pixels to obtain the qualified and defective connected components of the original sample image;

[0057] The defective connected component and the qualified connected component are analyzed for topological attributes to obtain the topological structure feature values ​​of the defective connected component and the qualified connected component. The topological structure feature values ​​include the number and area of ​​the qualified connected component, the number and area of ​​the defective connected component, and the spatial distance between the qualified connected component and the defective connected component.

[0058] Based on the topological feature values, the coupling relationship between the qualified connected components and the defective connected components is quantified to obtain the region crossover coefficient between the qualified and defective connected components. The formula for calculating the region crossover coefficient is as follows:

[0059] ;

[0060] in, This represents the interlacing coefficient of the region. This indicates the number of qualified connected components. This indicates the number of the defective connected components. Indicates the first The area of ​​each of the qualified connected regions. Indicates the first The area of ​​each of the aforementioned defective connected regions. Indicates the first The qualified connected components mentioned above and the first The square of the spatial distance between each of the defective connected domains;

[0061] Based on the spatial distribution topological structure feature value and the regional interlacing coefficient, the color uniformity of the original sample image is comprehensively scored to obtain the color uniformity evaluation index of the braised beef.

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

[0063] 1. This method acquires images of braised beef under standard light sources and accurately represents its color features. It constructs a multi-dimensional feature space through color space reconstruction, deeply exploring the spatial distribution patterns of color features in the image. Simultaneously, it combines the beef texture direction to achieve precise pixel segmentation, effectively distinguishing texture pixel sets from background pixel sets. Based on this, a uniformity judgment threshold surface is scientifically defined, and neighborhood consistency correction iterations are performed on pixel discrimination labels. This achieves high-precision identification of qualified and defective pixels, making the analysis of braised beef color features more scientific and accurate, and significantly improving the accuracy and reliability of color pixel discrimination.

[0064] 2. This method comprehensively determines color uniformity based on the ratio of qualified to defective pixels and their spatial distribution clustering. It quantifies the region overlap through connected component labeling and topological attribute parsing, achieving a digital and standardized comprehensive score for the color uniformity of braised beef and outputting accurate color uniformity evaluation indicators. The entire determination process is automated using image recognition and artificial intelligence technologies, eliminating the need for subjective human intervention and significantly improving the overall efficiency of color uniformity determination for braised beef. Furthermore, the quantified determination results provide precise data support for process optimization in the production of braised beef, adapting to the needs of large-scale industrial production testing and making the color uniformity determination results more valuable for reference and application. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating a method for determining the color uniformity of braised beef based on image recognition, according to an embodiment of the present invention.

[0066] 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

[0067] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0068] This application provides a method for determining the color uniformity of braised beef based on image recognition. The executing entity of this method includes, but is not limited to, at least one electronic device that can be configured to execute the method provided in this application, such as a server or a terminal. In other words, the method for determining the color uniformity of braised beef based on image recognition 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.

[0069] Reference Figure 1 The diagram shown is a flowchart illustrating a method for determining the color uniformity of braised beef based on image recognition, according to an embodiment of the present invention. In this embodiment, the method for determining the color uniformity of braised beef based on image recognition includes:

[0070] S1. Acquire original sample images of braised beef under a standard light source, and characterize the color features of the original sample images to obtain the reference color feature vector of the original sample images;

[0071] In this embodiment of the invention, the step of acquiring an original sample image of braised beef under a standard light source and performing color feature characterization on the original sample image to obtain a reference color feature vector of the original sample image includes:

[0072] Original sample images of braised beef are collected, and color feature deconstruction is performed on the original sample images to obtain the pixel color components of the original sample images. The pixel color components include red channel components, green channel components and blue channel components.

[0073] The color components of the pixels are mapped to equal interval quantization intervals, and the mapping results are frequency-collected to obtain the color frequency distribution sequence of the original sample image.

[0074] Weighted feature aggregation is performed on the color frequency distribution sequence to obtain the color feature normalization value of the original sample image;

[0075] The color feature values ​​are vectorized and arranged according to the channel order of the color components of the pixels to obtain the reference color feature vector of the original sample image.

[0076] Images of braised beef samples were acquired under constant illumination using standard light sources to ensure that the original sample images were clear and free from lighting deviations and distortions. After acquisition, pixel-by-pixel color feature deconstruction was performed on the original sample images. The overall color information of each pixel in the image was split into its corresponding red, green, and blue channel components. After the color information of all pixels in the image was split, the complete set of red, green, and blue channel components corresponding to all pixels in the original sample image was obtained, which is the pixel color component of the original sample image.

[0077] For the pixel color components of the extracted original sample image, equidistant quantization interval mapping is performed on the red, green, and blue channel components respectively. The value range of each color channel component is divided into multiple equidistant quantization intervals according to a unified partitioning rule. Then, the red, green, and blue channel components of each pixel are matched and mapped to the corresponding quantization intervals. After completing the interval mapping of all pixel color channel components, a frequency aggregation operation is performed on the mapping results. The number of pixels contained in each quantization interval is counted one by one. Then, according to the fixed order of red, green, and blue channels, the pixel frequency data corresponding to different quantization intervals under each color channel are sorted in sequence. These frequency data are then integrated in an orderly manner according to the arrangement order of each quantization interval to form an ordered frequency data set, that is, the color frequency distribution sequence of the original sample image.

[0078] A weighted feature aggregation operation is performed on the color frequency distribution sequence of the obtained original sample image. The frequency data of each quantization interval corresponding to the red, green and blue channels in the color frequency distribution sequence are fused and integrated. The frequency feature information scattered in the three color channels is uniformly aggregated and processed, and multiple scattered frequency feature data are merged into a unified feature value that can comprehensively reflect the overall frequency distribution characteristics of the three color channels of the original sample image. This unified feature value obtained after weighted feature aggregation is the color feature normalization value of the original sample image.

[0079] For the color feature rounded values ​​of the obtained original sample image, a vectorization operation is performed strictly according to the channel order of the pixel color components, that is, the order of red channel components, green channel components, and blue channel components. The color feature rounded values ​​that comprehensively reflect the characteristics of each color channel are transformed into an ordered vector form according to the predetermined channel order. The color feature information of each color channel is integrated into each dimension of the vector to form an ordered vector that can completely and systematically represent the overall color characteristics of the original sample image. This ordered vector is the reference color feature vector of the original sample image.

[0080] The beneficial effects of this implementation process are as follows: Starting with the standardized acquisition of original sample images of braised beef, the three-channel color components are accurately extracted through pixel-by-pixel color feature deconstruction. Then, the color features are quantized and statistically analyzed through equidistant quantization interval mapping and frequency aggregation. Subsequently, multi-channel color features are effectively integrated through weighted feature aggregation. Finally, the baseline color feature vector is obtained by vectorizing the vectors according to a fixed channel order. The entire process achieves accurate, systematic, and standardized representation of the color features of the original sample image. The obtained baseline color feature vector can completely, accurately, and comprehensively reflect the overall color features of the braised beef sample image, providing an accurate, reliable, and unified color feature foundation for subsequent color space reconstruction. At the same time, the standardized operation process ensures that the color feature representation results have high consistency and stability, effectively avoiding information deviations in the color feature extraction process and ensuring the accuracy of subsequent color uniformity determination.

[0081] S2. Reconstruct the color space of the reference color feature vector to obtain the multidimensional feature space of the original sample image;

[0082] In this embodiment of the invention, the step of reconstructing the reference color feature vector into a color space to obtain the multidimensional feature space of the original sample image includes:

[0083] The dimensional feature values ​​of the reference color feature vector are mapped to coordinates to obtain the initial feature coordinate point set of the original sample image;

[0084] Density clustering is performed on the initial set of feature coordinate points to obtain feature clusters of the original sample image;

[0085] The cluster centroids of the feature clusters are extracted to obtain the cluster centroid points of the feature clusters;

[0086] The feature space coordinate system of the original sample image is constructed by using the direction vector of the centroid of the cluster as the spatial basis vector of the original sample image.

[0087] Based on the feature space coordinate system, the initial feature coordinate point set is spatially relocated to obtain the multidimensional feature space of the original sample image.

[0088] The step of relocating the initial feature coordinate point set based on the feature space coordinate system to obtain the multidimensional feature space of the original sample image includes:

[0089] Based on the feature space coordinate system, the initial coordinate points in the initial feature coordinate point set are orthogonally projected and decomposed to obtain the projected coordinate values ​​of the initial coordinate points.

[0090] Based on the spatial neighborhood relationship between pixels in the original sample image, spatial consistency constraints are applied to the projected coordinate values ​​to obtain the corrected projected coordinate values ​​of the initial coordinate points.

[0091] The corrected projected coordinate values ​​are vectorized and assembled to obtain the repositioning coordinate vector of the initial coordinate point;

[0092] The repositioning coordinate vectors are arranged in a spatial structure to obtain the multidimensional feature space of the original sample image.

[0093] Coordinate mapping is performed on the feature values ​​of each dimension contained in the reference color feature vector. The corresponding spatial coordinate dimension is matched for each feature value, and the feature value of each dimension is directly converted into the coordinate value under the corresponding coordinate dimension. The single reference color feature vector is transformed into a unique coordinate point in space. All coordinate points transformed from the reference color feature vector are integrated to form a complete set of coordinate points. This set is the initial feature coordinate point set of the original sample image.

[0094] A comprehensive analysis of the spatial distribution density of all coordinate points in the initial feature coordinate point set is performed. Groups of coordinate points that are spatially close to each other and whose overall distribution density meets a unified judgment standard are identified one by one. Each group of coordinate points that meets the density judgment standard is independently classified and integrated. Each group of coordinate points that has been classified and integrated forms an independent cluster. All independent clusters are integrated together to form the feature clusters of the original sample image.

[0095] For each independent feature cluster, a precise spatial location analysis is performed. The specific position values ​​of all coordinate points in each feature cluster in each spatial coordinate dimension are sorted out, the mean of the position values ​​in each spatial coordinate dimension is calculated, and the spatial coordinate point corresponding to the mean is determined as the center position point of the corresponding feature cluster. Each feature cluster corresponds to a unique center position point, which is the cluster centroid. The extraction and integration of the cluster centroids corresponding to all feature clusters are completed.

[0096] A systematic analysis of the spatial orientation of all extracted cluster centroids is performed to determine the spatial orientation vector pointing from the origin of the spatial coordinate system to each cluster centroid. These spatial orientation vectors are used as the basis vectors for building the coordinate system. Based on the spatial orientation and dimensional characteristics of these basis vectors, a spatial coordinate system matching the dimension of the reference color feature vector is built. The completed coordinate system is the feature spatial coordinate system of the original sample image.

[0097] Based on the constructed feature space coordinate system, each initial coordinate point in the initial feature coordinate point set is projected onto each coordinate axis of the coordinate system. According to the orthogonal projection decomposition method, the overall spatial position of each initial coordinate point is decomposed into position components in each coordinate axis direction. The specific coordinate values ​​corresponding to the position components in each coordinate axis direction are accurately extracted. These values ​​are the projection coordinate values ​​of the initial coordinate point on the corresponding coordinate axis. The orthogonal projection decomposition of each initial coordinate point on all coordinate axes is completed and its projection coordinate values ​​are integrated.

[0098] The spatial neighborhood relationships between all pixels in the original sample image are comprehensively analyzed to clarify the specific range of adjacent pixels corresponding to each pixel. The spatial neighborhood relationship of the pixel is accurately mapped to the initial coordinate point corresponding to the initial feature coordinate point set. The projected coordinate value of each initial coordinate point is verified one by one. If there is a spatial deviation between the projected coordinate value and the projected coordinate value of the adjacent initial coordinate point, the projected coordinate value is adjusted and corrected according to the overall spatial distribution law of the projected coordinate values ​​in the neighborhood. The corrected projected coordinate value is the corrected projected coordinate value of the initial coordinate point.

[0099] The corrected projected coordinate values ​​obtained by each initial coordinate point on each coordinate axis of the feature space coordinate system are arranged in an orderly manner according to the coordinate axis arrangement order of the coordinate system. The series of corrected projected coordinate values ​​after arrangement are combined into an ordered vector form. This vector can accurately represent the new spatial position of the initial coordinate point in the feature space coordinate system. The ordered vector formed by this combination is the relocation coordinate vector of the initial coordinate point.

[0100] A comprehensive analysis of the spatial location features of the repositioning coordinate vectors corresponding to all initial coordinate points in the initial feature coordinate point set is performed. Based on the spatial features of each repositioning coordinate vector and their spatial relationships, these repositioning coordinate vectors are arranged in an orderly manner in the feature space coordinate system, so that each repositioning coordinate vector corresponds to a unique spatial position in the feature space coordinate system. All repositioning coordinate vectors and their corresponding spatial positions together constitute a structured multidimensional space, which is the multidimensional feature space of the original sample image.

[0101] The beneficial effects of this implementation process are that it completes a systematic color space reconstruction of the baseline color feature vector, transforms vector features into spatial coordinate point features through coordinate mapping, accurately divides feature clusters that fit the color feature distribution using density clustering, and constructs a feature space coordinate system based on the direction vector of the cluster centroid point, which is more suitable for the spatial distribution characteristics of the color features of the braised beef image. Subsequently, accurate spatial correction of the initial coordinate points is achieved through orthogonal projection decomposition and spatial consistency constraints, so that the repositioned coordinate vector can accurately reflect the spatial position of the color features. Finally, through the multi-dimensional feature space formed by spatial structured arrangement, the color features of the braised beef image are transformed from vector form to multi-dimensional spatial form, deeply explore the spatial correlation rules between color features, and fully and accurately present the spatial distribution state of the color features of the original sample image. This provides an accurate and scientific multi-dimensional spatial feature foundation for subsequent pixel segmentation based on the beef texture direction, effectively improving the accuracy and systematicness of subsequent color feature analysis.

[0102] S3. Based on the beef texture direction of the original sample image, the multidimensional feature space is segmented into pixels to obtain the texture pixel set and background pixel set of the multidimensional feature space.

[0103] In this embodiment of the invention, the step of segmenting the multidimensional feature space based on the beef texture direction of the original sample image to obtain the texture pixel set and background pixel set of the multidimensional feature space includes:

[0104] The original sample image is subjected to texture direction analysis to obtain the texture direction features of the original pixels in the original sample image.

[0105] The texture direction features are mapped to the corresponding pixels in the multidimensional feature space to obtain the feature pixels in the multidimensional feature space.

[0106] Based on the texture direction features, the feature pixels are clustered according to direction consistency to obtain the texture candidate clusters in the multidimensional feature space;

[0107] The texture candidate cluster is expanded by neighborhood growth to obtain the texture pixel set of the multidimensional feature space;

[0108] The feature pixels outside the texture pixel set in the multidimensional feature space are classified into background domains to obtain the background pixel set.

[0109] For each original pixel in the original sample image, the color and grayscale distribution of surrounding pixels are analyzed, the arrangement and extension trend of surrounding pixels are sorted out, the extension direction of beef texture corresponding to each original pixel is identified, the extension direction is standardized and characterized, and the standardized texture direction characterization results of all original pixels are integrated to obtain the texture direction features of the original pixels in the original sample image.

[0110] A one-to-one correspondence is established between the original pixels of the original sample image and the pixels in the multidimensional feature space. Based on this fixed correspondence, the texture direction feature of each original pixel is accurately assigned to the corresponding pixel in the multidimensional feature space. Each pixel in the multidimensional feature space is matched with a unique texture direction feature. All multidimensional feature space pixels with texture direction features are integrated to obtain the feature pixels of the multidimensional feature space.

[0111] A comprehensive and one-by-one comparative analysis of the texture direction features of all feature pixels in the multidimensional feature space is performed to identify feature pixels with completely consistent texture direction features and those extending in the same direction. These feature pixels with consistent texture direction features are then grouped and integrated to form multiple independent feature pixel groups. Each independent group composed of feature pixels with consistent direction features is a texture candidate cluster in the multidimensional feature space.

[0112] For each texture candidate cluster, a comprehensive analysis of the neighborhood range of all feature pixels in the multidimensional feature space is performed to clarify the specific range of adjacent feature pixels in the multidimensional feature space for each feature pixel. Adjacent feature pixels that are consistent with the texture direction features of the feature pixels in the texture candidate cluster are included into the texture candidate cluster one by one. The neighborhood search and pixel inclusion operation is continuously performed until there are no adjacent feature pixels that meet the texture direction feature requirements to be included. After the neighborhood growth and expansion operation of all texture candidate clusters is completed, all expanded texture candidate clusters are integrated as a whole to obtain the texture pixel set in the multidimensional feature space.

[0113] A comprehensive analysis of all feature pixels in the multidimensional feature space is conducted. Each feature pixel is assigned a specific category, and its inclusion in the texture pixel set is confirmed. All feature pixels not included in the texture pixel set are then uniformly categorized and organized. These feature pixels not assigned to the texture pixel set are collectively defined as background pixels. All background pixels are then integrated to form a complete pixel set, which is the background pixel set of the multidimensional feature space.

[0114] The beneficial effects are that this implementation process uses the actual texture direction of braised beef as the core basis for pixel segmentation. First, the texture direction features of the original pixels are extracted through precise texture direction analysis. Then, through one-to-one mapping, the pixels in the multi-dimensional feature space are endowed with corresponding texture feature attributes. Accurate screening of texture-related feature pixels is achieved by using directional consistency clustering. Subsequent neighborhood growth expansion makes the collection of texture pixels more comprehensive and closer to the actual texture distribution of beef. Finally, the clear division between the texture pixel set and the background pixel set is completed by background domain classification. The entire segmentation process closely follows the actual texture features of braised beef, achieving accurate and thorough segmentation of the two types of pixel sets in the multi-dimensional feature space. This effectively avoids the confusion between texture and background pixels, providing an accurate and pure pixel set foundation for subsequent targeted color feature statistics of texture pixel sets, and greatly improving the targeting and accuracy of subsequent color feature statistics.

[0115] S4. Perform color feature statistics on the texture pixel set and the background pixel set, and delineate the uniformity judgment threshold surface based on the feature distribution boundary in the multidimensional feature space according to the statistical results.

[0116] In this embodiment of the invention, the step of performing color feature statistics on the texture pixel set and the background pixel set, and delineating the uniformity determination threshold surface based on the feature distribution boundary in the multidimensional feature space according to the statistical results, includes:

[0117] Based on the coordinate positions in the multidimensional feature space, feature vector quantization is performed on the texture pixel set and the background pixel set to obtain the texture feature value of the texture pixel set and the background feature value of the background pixel set;

[0118] The distribution centers of the texture feature values ​​and the background feature values ​​are located to obtain the texture distribution center point and the background distribution center point of the texture pixel set and the background pixel set in the multidimensional feature space.

[0119] The texture distribution center point and the background distribution center point are spatially divided and positioned to obtain spatially equidistant points between the texture distribution center point and the background distribution center point;

[0120] Centered on the spatial equidistant points, a boundary surface is fitted to the distribution boundary region of the texture feature values ​​and the background feature values ​​in the multidimensional feature space to obtain the uniformity determination threshold surface of the multidimensional feature space.

[0121] The step of fitting a boundary surface to the distribution boundary region of the texture feature values ​​and the background feature values ​​in the multidimensional feature space, centered on the spatial equidistant points, to obtain the uniformity judgment threshold surface of the multidimensional feature space, includes:

[0122] Convex hull surfaces are constructed on the spatial points of the texture feature values ​​in the multidimensional feature space to obtain the texture distribution convex hull surface of the texture feature values.

[0123] An envelope surface is generated for the spatial points of the background feature values ​​in the multidimensional feature space to obtain the background distribution envelope surface of the background feature values.

[0124] The boundary lines of the spatial intersection region of the texture distribution convex hull surface and the background distribution envelope surface in the multidimensional feature space are extracted to obtain the distribution boundary contour lines of the texture feature values ​​and the background feature values.

[0125] Centered on the equidistant points in space, the distribution boundary contour is extended to obtain the uniformity determination threshold surface of the multidimensional feature space.

[0126] Based on the coordinate system of the multi-dimensional feature space, the specific coordinate position of each pixel in the texture pixel set and the background pixel set in this space is sorted out. Feature vector quantization is carried out in combination with the color feature information of each pixel. The spatial coordinate features and color features of each pixel are fused into a single quantized value. The quantized values ​​of all pixels in the texture pixel set are integrated to form a complete set, which is the texture feature value of the texture pixel set. The same method is used to complete the feature vector quantization and value integration of the background pixel set to obtain the background feature value of the background pixel set.

[0127] By identifying all spatial points corresponding to texture feature values ​​in the multidimensional feature space and analyzing their distribution across various coordinate dimensions, we can determine the coordinate point that represents the overall center position of this set of spatial points. This coordinate point is the texture distribution center point of the texture pixel set in the multidimensional feature space. Following the same analysis method, we can determine the overall center position of all spatial points corresponding to background feature values ​​to obtain the background distribution center point of the background pixel set in the multidimensional feature space.

[0128] Identify the spatial connection between the texture distribution center point and the background distribution center point in the multidimensional feature space, analyze the coordinate change pattern of this connection in each coordinate dimension, calculate the midpoint position of this spatial connection, and ensure that the coordinate point corresponding to this midpoint position is equidistant from both the texture distribution center point and the background distribution center point. This coordinate point is the spatial equidistant point between the texture distribution center point and the background distribution center point.

[0129] By sorting out all spatial points corresponding to texture feature values ​​in the multidimensional feature space, accurately identifying all points located at the outer boundary, and connecting them spatially according to the natural spatial distribution of the outer boundary points, a closed spatial surface that can completely enclose all corresponding spatial points of the texture feature values ​​is formed. This closed spatial surface is the texture distribution convex hull surface of the texture feature values.

[0130] We identify all spatial points corresponding to the background feature values ​​in the multidimensional feature space, analyze the overall spatial distribution contour features of these points, and construct a continuous spatial surface that closely fits the outer distribution state of all points based on these contour features. This continuous spatial surface can completely cover all corresponding spatial points of the background feature values. This continuous spatial surface is the background distribution envelope surface of the background feature values.

[0131] The texture distribution convex hull surface and the background distribution envelope surface are spatially superimposed and matched in a multi-dimensional feature space to accurately identify the specific areas where the two surfaces intersect and penetrate each other in space. The outer edge lines of the intersection area are extracted. These lines can clearly delineate the distribution boundaries of the spatial points corresponding to the texture feature values ​​and the background feature values. All outer edge lines are integrated to form a complete set of lines, which is the distribution boundary contour line of the texture feature values ​​and the background feature values.

[0132] Using spatial equidistant points as core reference points, and extending the framework based on the distribution boundary contour line, combined with the dimensional features of the multidimensional feature space, continuous and smooth surface extensions are made in all spatial directions of the distribution boundary contour line. The extended surface forms a complete spatial boundary surface in the multidimensional feature space. This surface can accurately define the distribution boundary between texture feature values ​​and background feature values. This extended complete spatial boundary surface is the uniformity judgment threshold surface of the multidimensional feature space.

[0133] The beneficial effects are that the implementation process accurately statistically and quantitatively analyzes the color features of texture and background pixel sets around a multi-dimensional feature space. By locating the distribution center, it accurately captures the core spatial distribution of the two types of feature values. The determination of spatial equidistant points provides a scientific central benchmark for the construction of the uniformity judgment threshold surface. The construction of the texture distribution convex hull surface and the background distribution envelope surface can accurately match the actual spatial distribution of the two types of feature values, thereby enabling the extracted distribution boundary contour lines to have high accuracy. Finally, the uniformity judgment threshold surface formed by extending from the spatial equidistant points can accurately match the actual distribution boundaries of texture and background feature values ​​in the multi-dimensional feature space. This provides a scientific, accurate, and objective judgment benchmark for the subsequent mapping and comparison of pixel feature vectors, effectively ensuring the accuracy and rationality of the subsequent work of distinguishing qualified and defective pixels.

[0134] S5. Perform spatial mapping comparison between the pixel feature vectors in the multidimensional feature space and the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image.

[0135] In this embodiment of the invention, the step of spatially mapping and comparing the pixel feature vectors in the multidimensional feature space with the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image includes:

[0136] The feature vectors of the pixels in the multidimensional feature space are parsed to obtain the feature vectors to be discriminated for the pixels.

[0137] Based on the uniformity threshold surface, the spatial orientation of the feature vector to be judged is determined to obtain the preliminary discrimination label of the pixel.

[0138] The preliminary discrimination label is subjected to neighborhood consistency analysis, and the preliminary discrimination label is corrected and iterated according to the analysis results to obtain the final discrimination label of the pixel.

[0139] Based on the final discrimination label, the original pixels of the original sample image are identified and divided to obtain the qualified pixels and defective pixels of the original sample image.

[0140] The step of performing neighborhood consistency analysis on the preliminary discrimination label and correcting and iterating the preliminary discrimination label based on the analysis results to obtain the final discrimination label of the pixel includes:

[0141] The neighborhood range of the pixel in the multidimensional feature space is defined to obtain the spatial neighborhood pixels of the pixel.

[0142] The frequency of tags in the spatial neighborhood pixels is statistically analyzed to obtain the percentage of qualified pixels and the percentage of defective pixels in the spatial neighborhood pixels.

[0143] Based on the proportion of qualified frequencies and the proportion of defective frequencies, a confidence value is assigned to the preliminary discrimination label to obtain the label confidence coefficient of the pixel.

[0144] Based on the label confidence coefficient, the preliminary discrimination label is weighted and corrected to obtain the corrected discrimination label of the pixel;

[0145] The modified discrimination label is iteratively converged to obtain the final discrimination label of the pixel.

[0146] A comprehensive analysis of spatial coordinate features and color features is performed on each pixel in the multidimensional feature space. All feature information contained in each pixel is extracted and structurally integrated. The integrated feature information is transformed into a vector form that can completely represent the features of the pixel. Each pixel corresponds to a unique vector of this type, which is the pixel's undiscriminated feature vector.

[0147] Each feature vector to be discriminated is precisely mapped to its corresponding position in the multidimensional feature space. The spatial point corresponding to the feature vector to be discriminated in the multidimensional feature space is determined. The specific spatial orientation of the spatial point relative to the uniformity judgment threshold surface is determined. If the spatial point is on the side of the uniformity judgment threshold surface pointing to the texture pixel set, the pixel is marked as qualified. If the spatial point is on the side of the uniformity judgment threshold surface pointing to the background pixel set, the pixel is marked as defective. The labeling result is the preliminary discrimination label of the pixel.

[0148] According to the coordinate distribution rules of the multidimensional feature space, a fixed spatial neighborhood range is defined for each pixel. All other pixels within this range are included in the neighborhood analysis range. All pixels within this range are integrated to form a complete set of pixels, which is the spatial neighborhood pixels of the pixel.

[0149] For each pixel, the preliminary discrimination labels of the corresponding spatial neighbor pixels are counted one by one. The number of pixels labeled as qualified and defective in the neighborhood are recorded respectively. The ratio of the number of qualified labels and the number of defective labels to the total number of spatial neighbor pixels is calculated. The values ​​obtained are the proportion of qualified frequency and the proportion of defective frequency in the spatial neighbor pixels, respectively.

[0150] Based on the initial identification label type of the pixel itself, and combined with the corresponding percentage of qualified and defective frequencies, a corresponding numerical confidence index is assigned to the initial identification label. If the initial identification label is qualified, the higher the percentage of qualified frequencies, the higher the confidence index value. If the initial identification label is defective, the higher the percentage of defective frequencies, the higher the confidence index value. This numerical confidence index is the label confidence coefficient of the pixel.

[0151] The initial label is adjusted and corrected based on the specific value of the label confidence coefficient. If the label confidence coefficient reaches the predetermined judgment standard, the original initial label is directly retained. If the label confidence coefficient does not reach the predetermined judgment standard, the initial label is corrected to a label type with a higher frequency proportion in the spatial neighborhood pixels. The corrected label result is the corrected label of the pixel.

[0152] The neighborhood range delineation, label frequency statistics, confidence assignment, and weighted correction operations are performed again on the corrected label of the pixel. This series of operations is repeated until the label result of the pixel no longer changes. The stable label result determined at this time is the final label of the pixel.

[0153] A one-to-one correspondence is established between pixels in the multidimensional feature space and original pixels in the original sample image. Based on this correspondence, the final discrimination label of each pixel is accurately assigned to the corresponding original pixel in the original sample image. All original pixels in the original sample image are identified and divided. Original pixels with the final discrimination label of qualified are uniformly integrated to form qualified pixels in the original sample image, and original pixels with the final discrimination label of defect are uniformly integrated to form defective pixels in the original sample image.

[0154] The beneficial effects of this implementation process are as follows: it achieves accurate extraction of pixel features through feature vector analysis of pixels in a multi-dimensional feature space; it completes the initial orientation discrimination of pixels based on the uniformity judgment threshold surface; it then achieves neighborhood consistency analysis of pixel labels through neighborhood range delineation and frequency statistics; it completes the accurate correction of the initial discrimination labels by combining confidence assignment and weighted correction; iterative convergence judgment makes the label results more stable and accurate; and finally, it completes the accurate identification and division of pixels in the original sample image through one-to-one correspondence. The entire process achieves high-precision discrimination between qualified and defective pixels, effectively avoiding the misjudgment and omission problems caused by single pixel discrimination, and allowing the discrimination results to accurately reflect the actual color distribution of the original sample image. This provides an accurate and reliable pixel discrimination basis for the subsequent comprehensive judgment of the color uniformity of braised beef.

[0155] S6. Based on the ratio and spatial distribution of the qualified and defective pixels, the color uniformity of the braised beef is comprehensively judged to obtain the color uniformity evaluation index of the braised beef.

[0156] In this embodiment of the invention, the step of comprehensively judging the color uniformity of the braised beef based on the ratio and spatial distribution aggregation of the qualified pixels and the defective pixels to obtain the color uniformity evaluation index of the braised beef includes:

[0157] Connectivity components are labeled for the qualified pixels and the defective pixels to obtain the qualified and defective connected components of the original sample image;

[0158] The defective connected component and the qualified connected component are analyzed for topological attributes to obtain the topological structure feature values ​​of the defective connected component and the qualified connected component. The topological structure feature values ​​include the number and area of ​​the qualified connected component, the number and area of ​​the defective connected component, and the spatial distance between the qualified connected component and the defective connected component.

[0159] Based on the topological feature values, the coupling relationship between the qualified connected components and the defective connected components is quantified to obtain the region crossover coefficient between the qualified and defective connected components. The formula for calculating the region crossover coefficient is as follows:

[0160] ;

[0161] in, This represents the interlacing coefficient of the region. This indicates the number of qualified connected components. This indicates the number of the defective connected components. Indicates the first The area of ​​each of the qualified connected regions. Indicates the first The area of ​​each of the aforementioned defective connected regions. Indicates the first The qualified connected components mentioned above and the first The square of the spatial distance between each of the defective connected domains;

[0162] Based on the spatial distribution topological structure feature value and the regional interlacing coefficient, the color uniformity of the original sample image is comprehensively scored to obtain the color uniformity evaluation index of the braised beef.

[0163] A comprehensive analysis of the spatial connectivity of qualified pixels in the original sample image is performed. Adjacent qualified pixels are grouped into the same independent connected region. Each formed independent connected region of qualified pixels is labeled. All labeled independent connected regions of qualified pixels are integrated to obtain the qualified connected region of the original sample image. The same method is used to analyze the spatial connectivity and label the independent connected regions of defective pixels in the original sample image. All labeled independent connected regions of defective pixels are integrated to obtain the defective connected region of the original sample image.

[0164] The number of qualified connected components in the original sample image is counted one by one, and the total number of qualified connected components is recorded. At the same time, the number of qualified pixels contained in each qualified connected component is counted to represent the area of ​​the corresponding qualified connected component. The same count and area representation operation is performed on all defective connected components, and the total number of defective connected components and the area of ​​each defective connected component are recorded. Then, the spatial positional relationship between each qualified connected component and each defective connected component is analyzed one by one, and the straight-line distance between the geometric center of each qualified connected component and the geometric center of each defective connected component is determined. The number and area of ​​qualified connected components, the number and area of ​​defective connected components, and the spatial distance between qualified and defective connected components are integrated to form a complete set of feature values. This set is the topological feature value of defective and qualified connected components.

[0165] Based on the extracted topological feature values, the area data of all qualified and defective connected domains, as well as the spatial distance data between each set of grid connected domains and defective connected domains, are sorted out. A comprehensive quantitative analysis of the spatial coupling distribution of qualified and defective connected domains is carried out. Combining the area proportion characteristics of each connected domain and the spatial distance distribution characteristics between them, the spatial interlacing correlation degree of qualified and defective connected domains is transformed into a single quantitative value that can accurately characterize the state. This quantitative value is the regional interlacing coefficient of qualified and defective connected domains.

[0166] The number of qualified connected components comes from the total number of qualified pixels in the original sample image after labeling them as connected components. The number of defective connected components comes from the total number of defective pixels in the original sample image after labeling them as connected components. The area of ​​the first qualified connected component comes from the area of ​​the first... The area of ​​a connected component is represented by the number of qualified pixels contained within each qualified connected component. The area of ​​the defective connected region comes from the area of ​​the first defective region. The area of ​​a connected region is represented by the number of defective pixels contained within each defective connected region. The qualified connected components and the first The square of the spatial distance of the first defective connected component comes from determining the first... The geometric center of each qualified connected region and the first The value obtained by multiplying the straight-line distance between the geometric centers of the connected domains of the defects by itself.

[0167] This calculation method is used to quantify the degree of spatial coupling and overlap between qualified and defective connected components. By comprehensively analyzing the area correlation and spatial distance correlation of each pair of qualified and defective connected components, it transforms this spatial overlap relationship into a single quantitative value, accurately representing the degree of overlap between the two types of connected components in spatial distribution. This provides a scientific quantitative basis for the comprehensive judgment of the color uniformity of braised beef, enabling the color uniformity evaluation index to more realistically reflect the spatial distribution uniformity of the color of braised beef.

[0168] The larger the area of ​​the qualified and defective connected components and the closer the spatial distance between them, the larger the value of the region crossover coefficient. The smaller the area of ​​the qualified and defective connected components and the farther the spatial distance between them, the smaller the value of the region crossover coefficient. The larger the value of the region crossover coefficient, the higher the degree of spatial coupling and crossover between the qualified and defective connected components and the worse the spatial uniformity of the color of braised beef. The smaller the value of the region crossover coefficient, the lower the degree of spatial coupling and crossover between the qualified and defective connected components and the better the spatial uniformity of the color of braised beef.

[0169] The data in the topological structure feature values ​​and the region crossover coefficient are used as the core basis for the comprehensive judgment of the color uniformity of braised beef. A clear correspondence between the data and the color uniformity score is established. According to the rule that the higher the proportion of qualified connected regions and the more regular the number distribution, the lower the proportion of defective connected regions and the fewer the number, the more reasonable the spatial distance distribution between qualified and defective connected regions, and the more the region crossover coefficient fits the characteristics of color uniformity, the higher the score is. The data are comprehensively calculated and scored, and the comprehensive score is transformed into a standardized index form. This standardized index is the color uniformity evaluation index of braised beef.

[0170] The beneficial effects are that this implementation process starts with the spatial distribution characteristics of qualified and defective pixels, accurately captures the spatial aggregation state of the two types of pixels through connected component labeling, and realizes the comprehensive quantitative extraction of multi-dimensional topological features of qualified and defective connected components through topological attribute parsing. The region crossover coefficient accurately represents the spatial coupling and crossover state of the two types of connected components. Finally, the topological structure feature value and the region crossover coefficient are combined for comprehensive scoring, realizing the digital and standardized comprehensive judgment of the color uniformity of braised beef. The obtained color uniformity evaluation index can comprehensively and accurately reflect the proportional distribution characteristics and spatial aggregation distribution characteristics of the color of braised beef, truly representing the actual color uniformity of braised beef. At the same time, the quantitative judgment results can provide accurate and specific data support for the process optimization of braised beef production, greatly improving the scientificity, accuracy and practical application value of the color uniformity judgment results.

[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 determining the color uniformity of braised beef based on image recognition, characterized in that, The method includes: S1. Acquire original sample images of braised beef under standard light source, and characterize the color features of the original sample images to obtain the reference color feature vector of the original sample images; S2. Reconstruct the color space of the reference color feature vector to obtain the multidimensional feature space of the original sample image; S3. Based on the beef texture direction of the original sample image, the multidimensional feature space is segmented into pixels to obtain the texture pixel set and background pixel set of the multidimensional feature space. S4. Perform color feature statistics on the texture pixel set and the background pixel set, and delineate the uniformity judgment threshold surface based on the feature distribution boundary in the multidimensional feature space according to the statistical results, including: Based on the coordinate positions in the multidimensional feature space, feature vector quantization is performed on the texture pixel set and the background pixel set to obtain the texture feature value of the texture pixel set and the background feature value of the background pixel set; The distribution centers of the texture feature values ​​and the background feature values ​​are located to obtain the texture distribution center point and the background distribution center point of the texture pixel set and the background pixel set in the multidimensional feature space. The texture distribution center point and the background distribution center point are spatially divided and positioned equally to obtain spatially equidistant points between the texture distribution center point and the background distribution center point; Centered on the equidistant points in space, a boundary surface is fitted to the distribution boundary region of the texture feature values ​​and the background feature values ​​in the multidimensional feature space to obtain the uniformity judgment threshold surface of the multidimensional feature space, including: Convex hull surfaces are constructed on the spatial points of the texture feature values ​​in the multidimensional feature space to obtain the texture distribution convex hull surface of the texture feature values. An envelope surface is generated for the spatial points of the background feature values ​​in the multidimensional feature space to obtain the background distribution envelope surface of the background feature values. The boundary lines of the spatial intersection region of the texture distribution convex hull surface and the background distribution envelope surface in the multidimensional feature space are extracted to obtain the distribution boundary contour lines of the texture feature values ​​and the background feature values. Using the spatial equidistant points as centers, the distribution boundary contour line is extended by a curved surface to obtain the uniformity judgment threshold surface of the multidimensional feature space. S5. Perform spatial mapping comparison between the pixel feature vectors in the multidimensional feature space and the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image. S6. Based on the ratio and spatial distribution of the qualified and defective pixels, the color uniformity of the braised beef is comprehensively judged to obtain the color uniformity evaluation index of the braised beef.

2. The method for determining the color uniformity of braised beef based on image recognition as described in claim 1, characterized in that, The process involves acquiring original sample images of braised beef under a standard light source and performing color feature characterization on the original sample images to obtain a baseline color feature vector for the original sample images, including: Original sample images of braised beef are collected, and color feature deconstruction is performed on the original sample images to obtain the pixel color components of the original sample images. The pixel color components include red channel components, green channel components and blue channel components. The color components of the pixels are mapped to equal interval quantization intervals, and the mapping results are frequency-collected to obtain the color frequency distribution sequence of the original sample image. Weighted feature aggregation is performed on the color frequency distribution sequence to obtain the color feature normalization value of the original sample image; The color feature values ​​are vectorized and arranged according to the channel order of the color components of the pixels to obtain the reference color feature vector of the original sample image.

3. The method for determining the color uniformity of braised beef based on image recognition as described in claim 1, characterized in that, The step of reconstructing the reference color feature vector into a color space to obtain the multidimensional feature space of the original sample image includes: The dimensional feature values ​​of the reference color feature vector are mapped to coordinates to obtain the initial feature coordinate point set of the original sample image; Density clustering is performed on the initial set of feature coordinate points to obtain feature clusters of the original sample image; The cluster centroids of the feature clusters are extracted to obtain the cluster centroid points of the feature clusters; The feature space coordinate system of the original sample image is constructed by using the direction vector of the centroid of the cluster as the spatial basis vector of the original sample image. Based on the feature space coordinate system, the initial feature coordinate point set is spatially relocated to obtain the multidimensional feature space of the original sample image.

4. The method for determining the color uniformity of braised beef based on image recognition as described in claim 3, characterized in that, The step of relocating the initial feature coordinate point set based on the feature space coordinate system to obtain the multidimensional feature space of the original sample image includes: Based on the feature space coordinate system, the initial coordinate points in the initial feature coordinate point set are orthogonally projected and decomposed to obtain the projected coordinate values ​​of the initial coordinate points. Based on the spatial neighborhood relationship between pixels in the original sample image, spatial consistency constraints are applied to the projected coordinate values ​​to obtain the corrected projected coordinate values ​​of the initial coordinate points. The corrected projected coordinate values ​​are vectorized and assembled to obtain the repositioning coordinate vector of the initial coordinate point; The repositioning coordinate vectors are arranged in a spatial structure to obtain the multidimensional feature space of the original sample image.

5. The method for determining the color uniformity of braised beef based on image recognition as described in claim 1, characterized in that, The step of segmenting the multidimensional feature space based on the beef texture direction of the original sample image into pixels yields a texture pixel set and a background pixel set in the multidimensional feature space, including: The original sample image is subjected to texture direction analysis to obtain the texture direction features of the original pixels in the original sample image. The texture direction features are mapped to the corresponding pixels in the multidimensional feature space to obtain the feature pixels in the multidimensional feature space. Based on the texture direction features, the feature pixels are clustered according to direction consistency to obtain the texture candidate clusters in the multidimensional feature space; The texture candidate clusters are expanded by neighborhood growth to obtain the texture pixel set of the multidimensional feature space; The feature pixels outside the texture pixel set in the multidimensional feature space are classified into background domains to obtain the background pixel set.

6. The method for determining the color uniformity of braised beef based on image recognition as described in claim 1, characterized in that, The step of spatially mapping and comparing the pixel feature vectors in the multidimensional feature space with the uniformity judgment threshold surface to obtain the qualified and defective pixels of the original sample image includes: The feature vectors of the pixels in the multidimensional feature space are parsed to obtain the feature vectors to be discriminated for the pixels. Based on the uniformity determination threshold surface, the spatial orientation of the feature vector to be determined is determined to obtain the preliminary discrimination label of the pixel. The preliminary discrimination label is subjected to neighborhood consistency analysis, and the preliminary discrimination label is corrected and iterated according to the analysis results to obtain the final discrimination label of the pixel. Based on the final discrimination label, the original pixels of the original sample image are identified and divided to obtain the qualified pixels and defective pixels of the original sample image.

7. The method for determining the color uniformity of braised beef based on image recognition as described in claim 6, characterized in that, The step of performing neighborhood consistency analysis on the preliminary discrimination label and correcting and iterating the preliminary discrimination label based on the analysis results to obtain the final discrimination label of the pixel includes: The neighborhood range of the pixel in the multidimensional feature space is defined to obtain the spatial neighborhood pixels of the pixel. The frequency of tags in the spatial neighborhood pixels is statistically analyzed to obtain the percentage of qualified pixels and the percentage of defective pixels in the spatial neighborhood pixels. Based on the proportion of qualified frequencies and the proportion of defective frequencies, a confidence value is assigned to the preliminary discrimination label to obtain the label confidence coefficient of the pixel. Based on the label confidence coefficient, the preliminary discrimination label is weighted and corrected to obtain the corrected discrimination label of the pixel; The modified discrimination label is iteratively converged to obtain the final discrimination label of the pixel.

8. The method for determining the color uniformity of braised beef based on image recognition as described in claim 1, characterized in that, The color uniformity of the braised beef is comprehensively judged based on the ratio and spatial distribution of qualified and defective pixels, resulting in an evaluation index for the color uniformity of the braised beef, including: Connectivity component labeling is performed on the qualified pixels and the defective pixels to obtain the qualified and defective connected components of the original sample image; The defective connected component and the qualified connected component are analyzed for topological attributes to obtain the topological structure feature values ​​of the defective connected component and the qualified connected component. The topological structure feature values ​​include the number and area of ​​the qualified connected component, the number and area of ​​the defective connected component, and the spatial distance between the qualified connected component and the defective connected component. Based on the topological feature values, the coupling relationship between the qualified connected components and the defective connected components is quantified to obtain the region crossover coefficient between the qualified and defective connected components. The formula for calculating the region crossover coefficient is as follows: ;in, This represents the interlacing coefficient of the region. This indicates the number of qualified connected components. This indicates the number of the defective connected components. Indicates the first The area of ​​each of the qualified connected regions. Indicates the first The area of ​​each of the aforementioned defective connected regions. Indicates the first The qualified connected components mentioned above and the first The square of the spatial distance between each of the defective connected domains; Based on the spatial distribution topological structure feature value and the regional interlacing coefficient, the color uniformity of the original sample image is comprehensively scored to obtain the color uniformity evaluation index of the braised beef.

Images