Seafood directional cutting method based on image recognition

Abstract

The application discloses a kind of based on image recognition's seafood directional cutting method, it is related to image recognition technical field, the application is by obtaining the polarization image group and multispectral image of the target area of seafood to be cut, according to polarization image group generates polarization degree image and according to polarization degree image generates candidate anchor point set, according to the design of multispectral image and candidate anchor point set generates target organ boundary anchor point set.This design, first by the spatial coordinate position of polarization degree less than neighborhood polarization degree representative value as candidate anchor point;On the basis of output candidate anchor point set, only keep the candidate anchor point of local optimal ratio greater than or equal to the average reflectivity ratio corresponding to selected single-band image as target organ boundary anchor point.By the cascade combination of primary screening and secondary screening, alleviate the problem that prior art cannot extract high confidence feature points on target organ boundary from scattered light interference.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method for directional cutting of seafood based on image recognition. Background Technology

[0002] In the field of automated seafood processing, image recognition-based visual guidance technology has been gradually applied to cutting path planning and execution. Existing technologies often employ the following methods to obtain the location information of target organs within seafood: First, polarization imaging technology is used to acquire multi-angle polarization images of the target area and calculate the degree of polarization, suppressing interference from scattered light from the water or tissue surface to enhance the contrast between the target and the background, thereby extracting the target contour. Second, multispectral or hyperspectral imaging technology is used to acquire spectral reflectance information across multiple bands, and target detection and boundary extraction are achieved by analyzing the spectral differences between the target and surrounding tissues. Some solutions attempt to fuse polarization and spectral information, improving the accuracy of target recognition through feature combination. These technologies have been applied in underwater target detection and marine organism detection, demonstrating certain effectiveness. Specifically, for fish after decapitation and evisceration, the blood lining that needs to be removed during fillet processing, the residual bone spurs of the vertebrae that need to be avoided during slicing, and the peritoneum remaining on the abdominal wall after evisceration are all target organs covered by mantle tissue and requiring localization during processing.

[0003] Currently, existing technologies for locating and avoiding internal target organs in seafood processing suffer from the following problems: the target organ is completely covered by a semi-transparent mantle tissue with strong scattering properties, and most of its boundary area is obscured by scattered light from the mantle, with only local fragments visible in the optical image. While polarization imaging can suppress scattered light, polarization information alone cannot distinguish the target organ boundary from pseudo-boundaries caused by factors such as variations in mantle thickness, uneven tissue density distribution, and surface condition differences, resulting in numerous noise points in the extraction results. Although multispectral imaging can capture the spectral absorption characteristics of the target organ in specific wavelength bands, global spectral analysis is easily affected by uneven illumination and individual tissue differences, making it difficult to reliably pinpoint weak boundaries that are partially obscured. Given that only part of the target organ boundary is visible, existing technologies fail to provide a stable and accurate method for extracting high-confidence feature points on the target organ boundary from scattered light interference, thus limiting the accuracy and safety of subsequent cutting path planning. Summary of the Invention

[0004] To address the technical problems existing in the background art, this invention proposes a seafood directional cutting method based on image recognition.

[0005] This invention proposes a method for directional cutting of seafood based on image recognition, comprising the following steps: S1. Acquire polarization image group and multispectral image of the target area of ​​the seafood to be cut; The target area of ​​the seafood to be cut includes the mantle tissue and the target organs covered by the mantle tissue; Mantle tissues exhibit light scattering properties in the near-infrared or visible light bands; As an explanation, since the mantle tissue exhibits light scattering characteristics in the near-infrared or visible light bands, in polarized image groups and multispectral images, the boundary of the target organ is partially obscured by the light scattering of the mantle tissue, appearing as discontinuous visible boundary segments. The boundary of the target organ is only partially visible, making it impossible to observe the target organ completely through single-view optical images. S2. Based on the polarization image set, generate the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut, forming a polarization degree image; based on the polarization degree image, generate a set of candidate anchor points. S3. Generate a set of target organ boundary anchor points based on the multispectral image and the candidate anchor point set; S4. Based on the set of boundary anchor points of the target organ, use the B-spline curve fitting algorithm or radial basis function interpolation method to perform boundary fitting and obtain the continuous boundary curve corresponding to the target organ. S5. Obtain the tool positioning accuracy parameters and the wall thickness of the target organ during cutting. Use the sum of the tool positioning accuracy parameters and the wall thickness as a safety distance. Shift the continuous boundary curve equidistantly along the side away from the target organ by this safety distance to generate the cutting path.

[0006] Preferably, in S2, based on the polarization image group, the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is generated, as follows: The polarization image group includes images at multiple polarization angles; pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut correspond to the same row and column positions in images at multiple polarization angles; Obtain the pixel grayscale value of each pixel in an image with multiple polarization angles; For a group of polarized images, the pixel grayscale value of the same spatial coordinate position in images at multiple polarization angles is used as the light intensity response value of that spatial coordinate position at different polarization angles; For the same spatial coordinate position at different polarization angles, the Stokes vector of that spatial coordinate position is obtained by using the Fourier transform algorithm; the degree of polarization corresponding to that spatial coordinate position is extracted from the Stokes vector of that spatial coordinate position. The polarization degree corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is obtained to form a polarization degree image.

[0007] Preferably, in S2, a set of candidate anchor points is generated based on the polarization degree image, as follows: Obtain the boundary of the polarization degree image, and expand it outward by a fixed width based on the boundary of the polarization degree image; use the area obtained by expanding it outward by a fixed width based on the boundary of the polarization degree image as the expanded area of ​​the polarization degree image. The polarization degree value is assigned to each spatial coordinate position in the extended region of the polarization degree image by using the mirror filling technique, so as to obtain the polarization degree corresponding to each spatial coordinate position in the extended region of the polarization degree image. The polarization degree corresponding to each spatial coordinate position in the target area of ​​the seafood to be cut is combined with the polarization degree corresponding to each spatial coordinate position in the extended area of ​​the polarization degree image to obtain the extended polarization degree image. Each spatial coordinate within the target area for cutting the seafood: In the extended polarization image, a first neighborhood window of fixed size is defined with the spatial coordinate position as the center; Obtain the degree of polarization corresponding to all spatial coordinate positions within the first neighborhood window, sum the degrees of polarization corresponding to all spatial coordinate positions within the first neighborhood window, divide the sum by the number of spatial coordinate positions within the first neighborhood window to obtain the mean degree of polarization corresponding to the first neighborhood window, and use the mean degree of polarization as the representative value of the neighborhood polarization corresponding to the first neighborhood window; if the degree of polarization corresponding to the spatial coordinate position is less than the representative value of the neighborhood polarization, then the spatial coordinate position is used as a candidate anchor point; As an explanation, when expanding outward by a fixed width based on the boundary of the polarization degree image, the fixed width is not less than half the side length of the first neighborhood window; Obtain all candidate anchor points and form a candidate anchor point set.

[0008] Preferably, in S3, a set of target organ boundary anchor points is generated based on the multispectral image and the candidate anchor point set, as follows: Based on the multispectral image, generate the spectral reflectance value of the same spatial coordinate position in each single-band image within the target area of ​​the seafood to be cut; For any candidate anchor point in the candidate anchor point set, obtain the spectral reflectance value of that candidate anchor point in each single-band image; Obtain the first neighborhood window corresponding to the candidate anchor point; within the first neighborhood window, obtain the spatial coordinates of all the seafood within the target area to be cut, and use all the spatial coordinates of all the seafood within the target area to be cut as the target spatial locations; obtain all the target spatial locations within the first neighborhood window to form a set of target spatial locations. Based on the set of target spatial locations, the full-band average reflectance corresponding to each target spatial location is generated in the multispectral image; Generate a selected single-band image based on the full-band average reflectance corresponding to each target spatial location; Generate an expanded image of the selected single-band image based on the selected single-band image; Based on the expanded single-band image, the target organ boundary anchor points are identified in the candidate anchor point set to form a target organ boundary anchor point set.

[0009] Preferably, in step S3, based on the multispectral image, the spectral reflectance values ​​of the same spatial coordinate location within the target area of ​​the seafood to be cut are generated in each single-band image, as follows: Multispectral images consist of multiple single-band images, each corresponding to a different center wavelength; The pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut, corresponding to the same row and column positions in multiple single-band images; Obtain the spectral reflectance value of each pixel in multiple single-band images; Obtain the spectral reflectance value of the same spatial coordinate position within the target area of ​​the seafood to be cut in each single-band image.

[0010] Preferably, in S3, based on the set of target spatial locations, the average reflectance across all bands corresponding to each target spatial location is generated in the multispectral image, as follows: In a multispectral image, for any target spatial location in the target spatial location set, the spectral reflectance value of that target spatial location in each single-band image is obtained; the spectral reflectance values ​​of that target spatial location in each single-band image are accumulated and then divided by the number of single-band images to obtain the full-band average reflectance corresponding to that target spatial location.

[0011] Preferably, in S3, a selected single-band image is generated based on the full-band average reflectance corresponding to each target spatial location, as follows: In the target spatial location set, obtain the full-band average reflectance corresponding to all target spatial locations to form a full-band average reflectance set; in the full-band average reflectance set, select the full-band average reflectance with the largest value as the selected full-band average reflectance; extract the target spatial location corresponding to the selected full-band average reflectance from the target spatial location set as the selected target spatial location; obtain the spectral reflectance value of the selected target spatial location in each single-band image; For each single-band image, the ratio of the spectral reflectance value of the candidate anchor point in the single-band image to the spectral reflectance value of the selected target spatial location in the single-band image is obtained, and used as the local reflectance ratio of the candidate anchor point in the single-band image. Obtain the local reflectance ratio of all candidate anchor points in the single-band image. After summing the local reflectance ratios of all candidate anchor points in the single-band image, divide the sum by the number of candidate anchor points in the candidate anchor point set to obtain the average reflectance ratio of the single-band image. Obtain the average reflectance ratio of all single-band images, select the center wavelength corresponding to the single-band image with the largest average reflectance ratio as the optimal band, and select the single-band image with the largest average reflectance ratio as the selected single-band image.

[0012] Preferably, in S3, an extended selected single-band image is generated based on the selected single-band image, as follows: Obtain the boundary of the selected single-band image, and expand it outward by a fixed width based on the boundary of the selected single-band image; use the area obtained by expanding it outward by a fixed width based on the boundary of the selected single-band image as the band image expansion area; As an explanation, when expanding a fixed width outward from the boundary of the selected single-band image, the same fixed width is used as when expanding a fixed width outward from the boundary of the polarization degree image.

[0013] The mirror fill technique is used to assign a spectral reflectance value to each spatial coordinate position within the extended region of the band image, thus obtaining the spectral reflectance value corresponding to each spatial coordinate position within the extended region of the band image. The spectral reflectance values ​​corresponding to the spatial coordinates of each location within the target area of ​​the seafood to be cut in the selected single-band image are combined with the spectral reflectance values ​​corresponding to the spatial coordinates of each location within the extended area of ​​the band image to obtain the extended selected single-band image.

[0014] Preferably, in S3, based on the expanded selected single-band image, the target organ boundary anchor points are identified in the candidate anchor point set to form a target organ boundary anchor point set, as follows: For any candidate anchor point in the candidate anchor point set: In the expanded selected single-band image, a second neighborhood window of fixed size is defined with the candidate anchor point as the center; As an explanation, when expanding outward by a fixed width based on the boundary of the selected single-band image, the fixed width is not less than half the side length of the second neighborhood window; Obtain the spectral reflectance values ​​corresponding to all spatial coordinate positions within the second neighborhood window to form a set of spectral reflectance values; in the set of spectral reflectance values, take the spectral reflectance value with the largest value as the local mantle reference reflectance corresponding to the candidate anchor point; Obtain the spectral reflectance value of the candidate anchor point in the extended optimal single-band image, and use it as the candidate reflectance; take the ratio of the local mantle reference reflectance to the candidate reflectance as the local optimal ratio of the candidate anchor point. If the local optimal ratio of the candidate anchor point is greater than or equal to the average reflectance ratio corresponding to the selected single-band image, then the candidate anchor point is used as the boundary anchor point of the target organ. As an explanation, the comparison between the local optimal ratio of the candidate anchor point and the average reflectance ratio corresponding to the selected single-band image is because, in reality, the boundary of the target organ is located at the interface of two different biological tissues. The tissues on both sides of this interface have inherent differences in composition or structure. These inherent differences are reflected in the selected single-band image as a step change in spectral reflectance, ensuring that the local reflectance ratio at the boundary of the target organ is at least not lower than the average level of the local reflectance ratios of all candidate anchor points in that band. In contrast, the decrease in spectral reflectance in the boundary region of non-target organs is caused by local thickness changes or surface state fluctuations of the mantle tissue itself. The spectral reflectance gradient in this type of region is lower than the spectral reflectance gradient at the tissue interface, and its corresponding local reflectance ratio will be lower than the average level.

[0015] Obtain all target organ boundary anchor points to form a target organ boundary anchor point set.

[0016] The image recognition-based directional cutting method for seafood proposed in this invention has the following beneficial technical effects: This application presents a design for acquiring a set of polarized images and a multispectral image of the target area of ​​the seafood to be cut, generating a polarization degree image based on the set of polarized images, generating a set of candidate anchor points based on the polarization degree image, and generating a set of target organ boundary anchor points based on the multispectral image and the set of candidate anchor points. In this design, the polarization degree corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is first determined to be less than the representative value of the polarization degree of the neighborhood corresponding to that spatial coordinate position. Spatial coordinate positions with polarization degrees less than the representative value of the neighborhood polarization degree are selected as candidate anchor points, thereby narrowing the range of areas where target organ boundaries may exist from the entire image to a set of candidate anchor points. Based on the output set of candidate anchor points, for any candidate anchor point, the spectral reflectance value of the candidate anchor point in the expanded selected single-band image is used as the candidate reflectance. The spectral reflectance value with the largest value in the second neighborhood window corresponding to the candidate anchor point is used as the local mantle reference reflectance corresponding to the candidate anchor point. The ratio of the local mantle reference reflectance to the candidate reflectance is used as the local optimal ratio of the candidate anchor point. The local optimal ratio of the candidate anchor point is compared with the average reflectance ratio corresponding to the selected single-band image. Only candidate anchor points with a local optimal ratio greater than or equal to the average reflectance ratio corresponding to the selected single-band image are retained as target organ boundary anchor points. By combining primary and secondary screening, under conditions where the mantle tissue exhibits light scattering characteristics in the near-infrared or visible light bands, the boundary of the target organ is partially obscured by the light scattering of the mantle tissue, presenting as discontinuous visible boundary segments, and the boundary of the target organ is only partially visible, making it impossible to observe the target organ completely through single-view optical images, anchor points belonging to the boundary of the target organ can be extracted more stably and accurately. This alleviates the problem that existing technologies fail to extract high-confidence feature points on the boundary of the target organ from the interference of scattered light. Attached Figure Description

[0017] Figure 1 This is a flowchart of a seafood directional cutting method based on image recognition according to the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] like Figure 1 The illustrated method for directional cutting of seafood based on image recognition includes the following steps: S1. Acquire polarization image group and multispectral image of the target area of ​​the seafood to be cut; The target area of ​​the seafood to be cut includes the mantle tissue and the target organs covered by the mantle tissue; Mantle tissues exhibit light scattering properties in the near-infrared or visible light bands; As an explanation, since the mantle tissue exhibits light scattering characteristics in the near-infrared or visible light bands, in polarized image groups and multispectral images, the boundary of the target organ is partially obscured by the light scattering of the mantle tissue, appearing as discontinuous visible boundary segments. The boundary of the target organ is only partially visible, making it impossible to observe the target organ completely through single-view optical images. S2. Based on the polarization image set, generate the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut, forming a polarization degree image; based on the polarization degree image, generate a set of candidate anchor points. In an optional embodiment, in S2, based on the polarization image group, the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is generated as follows: The polarization image group includes images at multiple polarization angles; pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut correspond to the same row and column positions in images at multiple polarization angles; Obtain the pixel grayscale value of each pixel in an image with multiple polarization angles; For a group of polarized images, the pixel grayscale value of the same spatial coordinate position in images at multiple polarization angles is used as the light intensity response value of that spatial coordinate position at different polarization angles; For the same spatial coordinate position at different polarization angles, the Stokes vector of that spatial coordinate position is obtained by using the Fourier transform algorithm; the degree of polarization corresponding to that spatial coordinate position is extracted from the Stokes vector of that spatial coordinate position. Obtain the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut, and form a polarization image; In an optional embodiment, in S2, a set of candidate anchor points is generated based on the polarization image, as follows: Obtain the boundary of the polarization degree image, and expand it outward by a fixed width based on the boundary of the polarization degree image; use the area obtained by expanding it outward by a fixed width based on the boundary of the polarization degree image as the expanded area of ​​the polarization degree image. The polarization degree value is assigned to each spatial coordinate position in the extended region of the polarization degree image by using the mirror filling technique, so as to obtain the polarization degree corresponding to each spatial coordinate position in the extended region of the polarization degree image. The polarization degree corresponding to each spatial coordinate position in the target area of ​​the seafood to be cut is combined with the polarization degree corresponding to each spatial coordinate position in the extended area of ​​the polarization degree image to obtain the extended polarization degree image. Each spatial coordinate within the target area for cutting the seafood: In the extended polarization image, a first neighborhood window of fixed size is defined with the spatial coordinate position as the center; Obtain the degree of polarization corresponding to all spatial coordinate positions within the first neighborhood window, sum the degrees of polarization corresponding to all spatial coordinate positions within the first neighborhood window, divide the sum by the number of spatial coordinate positions within the first neighborhood window to obtain the mean degree of polarization corresponding to the first neighborhood window, and use the mean degree of polarization as the representative value of the neighborhood polarization corresponding to the first neighborhood window; if the degree of polarization corresponding to the spatial coordinate position is less than the representative value of the neighborhood polarization, then the spatial coordinate position is used as a candidate anchor point; As an explanation, when expanding outward by a fixed width based on the boundary of the polarization degree image, the fixed width is not less than half the side length of the first neighborhood window; Obtain all candidate anchor points and form a candidate anchor point set; S3. Generate a set of target organ boundary anchor points based on the multispectral image and the candidate anchor point set; In an optional embodiment, in S3, a set of target organ boundary anchor points is generated based on the multispectral image and the candidate anchor point set, as follows: Based on the multispectral image, generate the spectral reflectance value of the same spatial coordinate position in each single-band image within the target area of ​​the seafood to be cut; In an optional embodiment, in S3, based on the multispectral image, the spectral reflectance values ​​of the same spatial coordinate location within the target area of ​​the seafood to be cut are generated in each single-band image, as follows: Multispectral images consist of multiple single-band images, each corresponding to a different center wavelength; The pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut, corresponding to the same row and column positions in multiple single-band images; Obtain the spectral reflectance value of each pixel in multiple single-band images; Obtain the spectral reflectance value of the same spatial coordinate position within the target area of ​​the seafood to be cut in each single-band image; For any candidate anchor point in the candidate anchor point set, obtain the spectral reflectance value of that candidate anchor point in each single-band image; Obtain the first neighborhood window corresponding to the candidate anchor point; within the first neighborhood window, obtain the spatial coordinates of all the seafood within the target area to be cut, and use all the spatial coordinates of all the seafood within the target area to be cut as the target spatial locations; obtain all the target spatial locations within the first neighborhood window to form a set of target spatial locations. Based on the set of target spatial locations, the full-band average reflectance corresponding to each target spatial location is generated in the multispectral image; In an optional embodiment, in S3, based on the set of target spatial locations, the full-band average reflectance corresponding to each target spatial location is generated in the multispectral image, as follows: In a multispectral image, for any target spatial location in the target spatial location set, the spectral reflectance value of that target spatial location in each single-band image is obtained; the spectral reflectance values ​​of that target spatial location in each single-band image are accumulated and then divided by the number of single-band images to obtain the full-band average reflectance corresponding to that target spatial location; Generate a selected single-band image based on the full-band average reflectance corresponding to each target spatial location; In an optional embodiment, in S3, a selected single-band image is generated based on the full-band average reflectance corresponding to each target spatial location, as follows: In the target spatial location set, obtain the full-band average reflectance corresponding to all target spatial locations to form a full-band average reflectance set; in the full-band average reflectance set, select the full-band average reflectance with the largest value as the selected full-band average reflectance; extract the target spatial location corresponding to the selected full-band average reflectance from the target spatial location set as the selected target spatial location; obtain the spectral reflectance value of the selected target spatial location in each single-band image; For each single-band image, the ratio of the spectral reflectance value of the candidate anchor point in the single-band image to the spectral reflectance value of the selected target spatial location in the single-band image is obtained, and used as the local reflectance ratio of the candidate anchor point in the single-band image. Obtain the local reflectance ratio of all candidate anchor points in the single-band image. After summing the local reflectance ratios of all candidate anchor points in the single-band image, divide the sum by the number of candidate anchor points in the candidate anchor point set to obtain the average reflectance ratio of the single-band image. Obtain the average reflectance ratio of all single-band images, and select the center wavelength corresponding to the single-band image with the largest average reflectance ratio as the optimal band; select the single-band image with the largest average reflectance ratio as the selected single-band image. Generate an expanded image of the selected single-band image based on the selected single-band image; In an optional embodiment, in S3, an extended selected single-band image is generated based on the selected single-band image, as follows: Obtain the boundary of the selected single-band image, and expand it outward by a fixed width based on the boundary of the selected single-band image; use the area obtained by expanding it outward by a fixed width based on the boundary of the selected single-band image as the band image expansion area; As an explanation, when expanding a fixed width outward from the boundary of the selected single-band image, the same fixed width is used as when expanding a fixed width outward from the boundary of the polarization degree image.

[0020] The mirror fill technique is used to assign a spectral reflectance value to each spatial coordinate position within the extended region of the band image, thus obtaining the spectral reflectance value corresponding to each spatial coordinate position within the extended region of the band image. The spectral reflectance values ​​of each spatial coordinate position in the target area of ​​the seafood to be cut in the selected single-band image are combined with the spectral reflectance values ​​of each spatial coordinate position in the extended area of ​​the band image to obtain the extended selected single-band image. Based on the expanded single-band image, the target organ boundary anchor points are identified in the candidate anchor point set to form a target organ boundary anchor point set.

[0021] In an optional embodiment, in S3, based on the expanded selected single-band image, target organ boundary anchor points are identified in the candidate anchor point set to form a target organ boundary anchor point set, as follows: For any candidate anchor point in the candidate anchor point set: In the expanded selected single-band image, a second neighborhood window of fixed size is defined with the candidate anchor point as the center; As an explanation, when expanding outward by a fixed width based on the boundary of the selected single-band image, the fixed width is not less than half the side length of the second neighborhood window; Obtain the spectral reflectance values ​​corresponding to all spatial coordinate positions within the second neighborhood window to form a set of spectral reflectance values; in the set of spectral reflectance values, take the spectral reflectance value with the largest value as the local mantle reference reflectance corresponding to the candidate anchor point; Obtain the spectral reflectance value of the candidate anchor point in the extended optimal single-band image, and use it as the candidate reflectance; take the ratio of the local mantle reference reflectance to the candidate reflectance as the local optimal ratio of the candidate anchor point. If the local optimal ratio of the candidate anchor point is greater than or equal to the average reflectance ratio corresponding to the selected single-band image, then the candidate anchor point is used as the boundary anchor point of the target organ. As an explanation, the comparison between the local optimal ratio of the candidate anchor point and the average reflectance ratio corresponding to the selected single-band image is because, in reality, the boundary of the target organ is located at the interface of two different biological tissues. The tissues on both sides of this interface have inherent differences in composition or structure. These inherent differences are reflected in the selected single-band image as a step change in spectral reflectance, ensuring that the local reflectance ratio at the boundary of the target organ is at least not lower than the average level of the local reflectance ratios of all candidate anchor points in that band. In contrast, the decrease in spectral reflectance in the boundary region of non-target organs is caused by local thickness changes or surface state fluctuations of the mantle tissue itself. The spectral reflectance gradient in this type of region is lower than the spectral reflectance gradient at the tissue interface, and its corresponding local reflectance ratio will be lower than the average level.

[0022] Obtain all target organ boundary anchor points to form a target organ boundary anchor point set; S4. Based on the set of boundary anchor points of the target organ, use the B-spline curve fitting algorithm or radial basis function interpolation method to perform boundary fitting and obtain the continuous boundary curve corresponding to the target organ. S5. Obtain the tool positioning accuracy parameters and the wall thickness of the target organ during cutting. Use the sum of the tool positioning accuracy parameters and the wall thickness as a safety distance. Shift the continuous boundary curve equidistantly along the side away from the target organ by this safety distance to generate the cutting path.

[0023] This application presents a design for acquiring a set of polarized images and a multispectral image of the target area of ​​the seafood to be cut, generating a polarization degree image based on the set of polarized images, generating a set of candidate anchor points based on the polarization degree image, and generating a set of target organ boundary anchor points based on the multispectral image and the set of candidate anchor points. In this design, the polarization degree corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is first determined to be less than the representative value of the polarization degree of the neighborhood corresponding to that spatial coordinate position. Spatial coordinate positions with polarization degrees less than the representative value of the neighborhood polarization degree are selected as candidate anchor points, thereby narrowing the range of areas where target organ boundaries may exist from the entire image to a set of candidate anchor points. Based on the output set of candidate anchor points, for any candidate anchor point, the spectral reflectance value of the candidate anchor point in the expanded selected single-band image is used as the candidate reflectance. The spectral reflectance value with the largest value in the second neighborhood window corresponding to the candidate anchor point is used as the local mantle reference reflectance corresponding to the candidate anchor point. The ratio of the local mantle reference reflectance to the candidate reflectance is used as the local optimal ratio of the candidate anchor point. The local optimal ratio of the candidate anchor point is compared with the average reflectance ratio corresponding to the selected single-band image. Only candidate anchor points with a local optimal ratio greater than or equal to the average reflectance ratio corresponding to the selected single-band image are retained as target organ boundary anchor points. By combining primary and secondary screening, under conditions where the mantle tissue exhibits light scattering characteristics in the near-infrared or visible light bands, the boundary of the target organ is partially obscured by the light scattering of the mantle tissue, presenting as discontinuous visible boundary segments, and the boundary of the target organ is only partially visible, making it impossible to observe the target organ completely through single-view optical images, anchor points belonging to the boundary of the target organ can be extracted more stably and accurately. This alleviates the problem that existing technologies fail to extract high-confidence feature points on the boundary of the target organ from the interference of scattered light.

[0024] For clarification, "acquisition" in this application refers to obtaining the required content or data using existing technical means.

[0025] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0026] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0027] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0028] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0029] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for directional cutting of seafood based on image recognition, characterized in that, Includes the following steps: S1. Acquire polarization image groups and multispectral images of the target area of ​​the seafood to be cut; The target area of ​​the seafood to be cut includes the mantle tissue and the target organs covered by the mantle tissue; the mantle tissue exhibits light scattering characteristics in the near-infrared or visible light bands. S2. Based on the polarization image set, generate the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut, forming a polarization degree image; based on the polarization degree image, generate a set of candidate anchor points. S3. Generate a set of target organ boundary anchor points based on the multispectral image and the candidate anchor point set; S4. Based on the set of boundary anchor points of the target organ, use the B-spline curve fitting algorithm or radial basis function interpolation method to perform boundary fitting and obtain the continuous boundary curve corresponding to the target organ. S5. Obtain the tool positioning accuracy parameters and the wall thickness of the target organ during cutting. Use the sum of the tool positioning accuracy parameters and the wall thickness as a safety distance. Shift the continuous boundary curve equidistantly along the side away from the target organ by this safety distance to generate the cutting path.

2. The seafood directional cutting method based on image recognition according to claim 1, characterized in that, In S2, based on the polarization image group, the degree of polarization corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is generated, as follows: The polarization image group includes images at multiple polarization angles; pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut correspond to the same row and column positions in images at multiple polarization angles; Obtain the pixel grayscale value of each pixel in an image with multiple polarization angles; For a group of polarized images, the pixel grayscale value of the same spatial coordinate position in images at multiple polarization angles is used as the light intensity response value of that spatial coordinate position at different polarization angles; For the same spatial coordinate location at different polarization angles, the Stokes vector of that spatial coordinate location is obtained by using the Fourier transform algorithm; the degree of polarization corresponding to that spatial coordinate location is extracted from the Stokes vector of that spatial coordinate location. The polarization degree corresponding to each spatial coordinate position within the target area of ​​the seafood to be cut is obtained to form a polarization degree image.

3. The seafood directional cutting method based on image recognition according to claim 2, characterized in that, In S2, a set of candidate anchor points is generated based on the polarization degree image, as follows: Obtain the boundary of the polarization degree image, and expand it outward by a fixed width based on the boundary of the polarization degree image; use the area obtained by expanding it outward by a fixed width based on the boundary of the polarization degree image as the expanded area of ​​the polarization degree image. The polarization degree value is assigned to each spatial coordinate position in the extended region of the polarization degree image by using the mirror filling technique, so as to obtain the polarization degree corresponding to each spatial coordinate position in the extended region of the polarization degree image. The polarization degree corresponding to each spatial coordinate position in the target area of ​​the seafood to be cut is combined with the polarization degree corresponding to each spatial coordinate position in the extended area of ​​the polarization degree image to obtain the extended polarization degree image. Each spatial coordinate within the target area for cutting the seafood: In the extended polarization image, a first neighborhood window of fixed size is defined with the spatial coordinate position as the center; Obtain the degree of polarization corresponding to all spatial coordinate positions within the first neighborhood window, sum the degrees of polarization corresponding to all spatial coordinate positions within the first neighborhood window, divide the sum by the number of spatial coordinate positions within the first neighborhood window to obtain the mean degree of polarization corresponding to the first neighborhood window, and use the mean degree of polarization as the representative value of the neighborhood polarization corresponding to the first neighborhood window; if the degree of polarization corresponding to the spatial coordinate position is less than the representative value of the neighborhood polarization, then the spatial coordinate position is used as a candidate anchor point; Obtain all candidate anchor points and form a candidate anchor point set.

4. The seafood directional cutting method based on image recognition according to claim 3, characterized in that, In S3, based on the multispectral image and the candidate anchor point set, a set of target organ boundary anchor points is generated, as follows: Based on the multispectral image, generate the spectral reflectance value of the same spatial coordinate position in each single-band image within the target area of ​​the seafood to be cut; For any candidate anchor point in the candidate anchor point set, obtain the spectral reflectance value of that candidate anchor point in each single-band image; Obtain the first neighborhood window corresponding to the candidate anchor point; within the first neighborhood window, obtain the spatial coordinates of all the seafood within the target area to be cut, and use all the spatial coordinates of all the seafood within the target area to be cut as the target spatial locations; obtain all the target spatial locations within the first neighborhood window to form a set of target spatial locations. Based on the set of target spatial locations, the full-band average reflectance corresponding to each target spatial location is generated in the multispectral image; Generate a selected single-band image based on the full-band average reflectance corresponding to each target spatial location; Generate an expanded image of the selected single-band image based on the selected single-band image; Based on the expanded single-band image, the target organ boundary anchor points are identified in the candidate anchor point set to form a target organ boundary anchor point set.

5. The seafood directional cutting method based on image recognition according to claim 4, characterized in that, In S3, based on the multispectral image, the spectral reflectance values ​​of the same spatial coordinate location within the target area of ​​the seafood to be cut are generated in each single-band image, as follows: Multispectral images consist of multiple single-band images, each corresponding to a different center wavelength; The pixels at the same spatial coordinate position within the target area of ​​the seafood to be cut, corresponding to the same row and column positions in multiple single-band images; Obtain the spectral reflectance value of each pixel in multiple single-band images; Obtain the spectral reflectance value of the same spatial coordinate position within the target area of ​​the seafood to be cut in each single-band image.

6. The seafood directional cutting method based on image recognition according to claim 4, characterized in that, In S3, based on the set of target spatial locations, the average reflectance across all bands corresponding to each target spatial location is generated in the multispectral image, as follows: In a multispectral image, for any target spatial location in the target spatial location set, the spectral reflectance value of that target spatial location in each single-band image is obtained; the spectral reflectance values ​​of that target spatial location in each single-band image are accumulated and then divided by the number of single-band images to obtain the full-band average reflectance corresponding to that target spatial location.

7. The seafood directional cutting method based on image recognition according to claim 4, characterized in that, In S3, a selected single-band image is generated based on the full-band average reflectance corresponding to each target spatial location, as follows: In the target spatial location set, obtain the full-band average reflectance corresponding to all target spatial locations to form a full-band average reflectance set; in the full-band average reflectance set, select the full-band average reflectance with the largest value as the selected full-band average reflectance. From the set of target spatial locations, extract the target spatial location corresponding to the average reflectance of the selected full-band image as the selected target spatial location; obtain the spectral reflectance value of the selected target spatial location in each single-band image; For each single-band image, the ratio of the spectral reflectance value of the candidate anchor point in the single-band image to the spectral reflectance value of the selected target spatial location in the single-band image is obtained, and used as the local reflectance ratio of the candidate anchor point in the single-band image. Obtain the local reflectance ratio of all candidate anchor points in the single-band image. After summing the local reflectance ratios of all candidate anchor points in the single-band image, divide the sum by the number of candidate anchor points in the candidate anchor point set to obtain the average reflectance ratio of the single-band image. Obtain the average reflectance ratio of all single-band images, select the center wavelength corresponding to the single-band image with the largest average reflectance ratio as the optimal band, and select the single-band image with the largest average reflectance ratio as the selected single-band image.

8. The seafood directional cutting method based on image recognition according to claim 4, characterized in that, In S3, an expanded image of the selected single-band image is generated based on the selected single-band image, as follows: Obtain the boundary of the selected single-band image, and expand it outward by a fixed width based on the boundary of the selected single-band image; use the area obtained by expanding it outward by a fixed width based on the boundary of the selected single-band image as the band image expansion area; The mirror fill technique is used to assign a spectral reflectance value to each spatial coordinate position within the extended region of the band image, thus obtaining the spectral reflectance value corresponding to each spatial coordinate position within the extended region of the band image. The spectral reflectance values ​​corresponding to the spatial coordinates of each location within the target area of ​​the seafood to be cut in the selected single-band image are combined with the spectral reflectance values ​​corresponding to the spatial coordinates of each location within the extended area of ​​the band image to obtain the extended selected single-band image.

9. The seafood directional cutting method based on image recognition according to claim 4, characterized in that, In S3, based on the expanded selected single-band image, the target organ boundary anchor points are identified from the candidate anchor point set, forming the target organ boundary anchor point set as follows: For any candidate anchor point in the candidate anchor point set: In the expanded selected single-band image, a second neighborhood window of fixed size is defined with the candidate anchor point as the center; Obtain the spectral reflectance values ​​corresponding to all spatial coordinate positions within the second neighborhood window to form a set of spectral reflectance values; in the set of spectral reflectance values, take the spectral reflectance value with the largest value as the local mantle reference reflectance corresponding to the candidate anchor point; Obtain the spectral reflectance value of the candidate anchor point in the extended optimal single-band image, and use it as the candidate reflectance; take the ratio of the local mantle reference reflectance to the candidate reflectance as the local optimal ratio of the candidate anchor point. If the local optimal ratio of the candidate anchor point is greater than or equal to the average reflectance ratio corresponding to the selected single-band image, then the candidate anchor point is used as the boundary anchor point of the target organ. Obtain all target organ boundary anchor points to form a target organ boundary anchor point set.

Images