A broiler chicken online grading and intelligent cutting system based on machine vision

Abstract

The present application relates to meat product processing technical field, especially to a kind of broiler online grading and intelligent segmentation system based on machine vision, including online grading vision unit, segmentation feature extraction unit, multi-axis cutting execution unit, pressure sensing unit and optimization control unit, the surface image of broiler is obtained by vision unit and is graded online, then its multi-view profile is analyzed by feature extraction unit to calculate the density distribution index representing trunk structure characteristics;Optimization control unit integrates grading results and structure characteristics, plans initial cutting parameters and expected pressure interval;During cutting process, through real-time feedback of pressure sensor, sawing path and speed are dynamically adjusted, so that cutting force is maintained in the optimal interval.The present application realizes self-adaptive accurate segmentation to individual difference, effectively improves cutting quality and yield.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of meat product processing, and particularly relates to a broiler chicken online grading and intelligent cutting system based on machine vision. BACKGROUND

[0002] In the poultry slaughtering and processing industry, automatic cutting of broiler chickens is a key link to improve production efficiency and product standardization. At present, the automatic bone sawing equipment commonly used in the industry relies on pre-set mechanical programs and fixed cutting paths. Such equipment usually drives cutting tools (such as disc saws and band saws) to cut the broiler chicken body suspended on the conveying line through pre-set spatial coordinates and feeding speed. Some more advanced systems will introduce basic visual recognition technology to judge the approximate size or orientation of the broiler chicken, so as to trigger different fixed cutting program libraries to cope with size differences within a certain range. These technology paradigms have realized the leap from pure manual work to mechanization and automation, and basically meet the needs of continuous production. However, the core logic is essentially "open-loop", that is, the cutting execution process strictly follows the pre-set instructions, and lacks the ability of online perception and real-time response to the individual specific physical properties of the processing object.

[0003] Based on the above existing technology, the main technical problems of broiler chicken automatic cutting are concentrated in the stability of cutting quality and raw material utilization. First of all, due to the significant differences in body structure, bone hardness and muscle fat distribution of broiler chickens caused by different breeds, feeding cycles and feeding conditions, the fixed cutting path and parameters cannot be adapted to them. This directly leads to phenomena such as saw blade deviation, bone fragmentation or uneven cutting surface in the cutting process, causing excessive bone residue to pollute the meat, rough cuts to affect the appearance, and even mistakenly cutting to damage the integrity of high-priced parts (such as complete chicken breasts), which seriously affects the product yield and commodity grade. Therefore, the existing equipment, while pursuing high efficiency, is difficult to meet the processing requirements of high precision and low loss, becoming a technical bottleneck restricting the further development of the poultry processing industry towards refinement and high value.

[0004] The Chinese patent publication No. CN117084285A discloses a chicken processing chicken wing segmentation device, which comprises a workbench, an installation groove is formed in the middle of the workbench, a group of drive shafts are installed inside the installation groove, each drive shaft is provided with a conveying belt for conveying materials, a drive motor is fixedly installed outside the workbench, and the output end of the drive motor is fixedly connected with one of the drive shafts. In the present application, in order to realize continuous automatic feeding operation, the drive motor drives the drive shaft to rotate, and the conveying belt transfers the materials above to one side, so that the present application can realize automatic continuous feeding and discharging. By transferring the materials to the corresponding position, under the positioning cooperation of the positioning assembly and the limiting assembly, the automatically fixed continuous chicken wing parts can be quickly segmented by the segmentation assembly, and then the chicken wings and the main body are classified and collected. Therefore, the chicken processing chicken wing segmentation device has the following problems: it lacks real-time perception of the characteristics of broilers, which leads to the fact that fixed cutting parameters cannot adapt to individual differences, resulting in unstable cutting quality. SUMMARY

[0005] To this end, the present application provides a broiler online grading and intelligent segmentation system based on machine vision to overcome the problem in the prior art that the lack of real-time perception of the characteristics of broilers leads to the fact that fixed cutting parameters cannot adapt to individual differences, resulting in unstable cutting quality.

[0006] To achieve the above-mentioned purpose, the present application provides a broiler online grading and intelligent segmentation system based on machine vision, which comprises:

[0007] An online grading vision unit, which comprises an image acquisition module for acquiring surface images of broilers on a hanging conveying line and a first feature calculation module for obtaining a grading signal corresponding to a preset commercial grade based on the surface images;

[0008] A segmentation feature extraction unit, which is arranged at a subsequent station of the online grading vision unit, comprises a multi-view acquisition module for acquiring multi-view profile images of the broilers, and a second feature calculation module for calculating the multi-view profile images to obtain a trunk density distribution index for representing the trunk weight distribution and the skeletal center of gravity position of the broilers;

[0009] A multi-axis cutting execution unit, which comprises a hanging conveying module for clamping and moving the broilers, and a multi-axis linkage bone sawing module for sawing the broilers;

[0010] A pressure sensing unit, which is arranged on the multi-axis linkage bone sawing module, is used to detect the contact pressure feedback value between the saw blade and the broiler tissue in the sawing process in real time;

[0011] An optimization control unit is connected with the online grading vision unit, the segmentation feature extraction unit, the multi-axis cutting execution unit and the pressure sensing unit respectively, to determine a desired pressure interval according to the obtained initial sternum rigidity offset and initial lower saw offset angle, and to adjust the cutting path and cutting feed speed of the multi-axis linkage saw bone module according to the obtained contact pressure feedback value during single sawing process.

[0012] Further, the first feature calculation module processes the surface image, obtains contour geometric parameters representing the overall size of the broiler chicken through contour fitting, and determines a defect feature region in the epidermis region through image segmentation;

[0013] The contour geometric parameter quantitative value is calculated according to the contour geometric parameters, and a defect representation value is calculated according to the defect feature region;

[0014] The grading signal is determined according to the comparison result of the contour geometric parameter quantitative value and the first preset grading threshold, and the comparison result of the defect representation value and the second preset grading threshold.

[0015] Further, the second feature calculation module processes the multi-view contour image to obtain a three-dimensional contour model of the broiler chicken body;

[0016] Based on the three-dimensional contour model, a first projection area of the thoracic region and a second projection area of the leg crotch region are calculated respectively, and a first feature parameter is determined according to the ratio of the first projection area to the second projection area;

[0017] Based on the three-dimensional contour model, a second feature parameter is determined according to the width-height ratio of the broiler chicken body contour;

[0018] The body density distribution index is obtained based on the first feature parameter and the second feature parameter.

[0019] Further, the optimization control unit determines a cutting quality target library corresponding to different commercial grade broiler chickens required to define cutting accuracy and loss rate based on the grading signal, and calculates the initial sternum rigidity offset and the initial lower saw offset angle according to the quality coefficient determined by the body density distribution index and the cutting quality target library; the base pressure interval median value and the base pressure interval half-width are determined based on the initial sternum rigidity offset, and the base pressure interval median value and the base pressure interval half-width are corrected based on the quality coefficient to obtain the actual pressure interval median value and the actual pressure interval half-width of the desired pressure interval.

[0020] Further, the optimization control unit calculates the pressure deviation value of the contact pressure feedback value and the actual pressure interval median value of the desired pressure interval;

[0021] The direction of adjustment of the sawing feed speed is determined based on the sign of the pressure deviation value; wherein, when the deviation value is positive, the sawing feed speed is decreased, and when the deviation value is negative, the sawing feed speed is increased.

[0022] Based on the absolute value of the deviation, the adjustment range of the sawing feed speed and the compensation adjustment amount for the downward saw offset angle are determined; wherein, both the adjustment range and the compensation adjustment amount are positively correlated with the absolute value of the deviation.

[0023] Furthermore, the optimization control unit calculates the initial compensation angle based on the absolute value of the pressure deviation; and corrects the initial compensation angle based on the quality coefficient to determine the compensation adjustment amount.

[0024] Furthermore, the trunk density distribution index is positively correlated with the first feature parameter and negatively correlated with the second feature parameter.

[0025] Furthermore, the initial sternal stiffness offset is positively correlated with the trunk density distribution index, and the initial downward saw offset angle is positively correlated with the trunk density distribution index.

[0026] Furthermore, the image acquisition module is positioned above and to the side of the suspended conveyor line to acquire surface images that include the complete outer contour of the broiler chicken and the main body surface.

[0027] Furthermore, the multi-view acquisition module includes at least one first camera disposed on one side of the suspended conveyor line, and at least one second camera disposed at a non-zero angle with the optical axis of the first camera, for synchronously acquiring the multi-view contour image of the broiler chicken.

[0028] The second feature calculation module is connected to the multi-view acquisition module to receive and process the multi-view contour image.

[0029] Compared with existing technologies, the beneficial effects of this invention are that it constructs a collaborative system comprising an online hierarchical vision unit, a segmentation feature extraction unit, a multi-axis cutting execution unit, a pressure sensing unit, and an optimization control unit, thereby achieving closed-loop control of the entire process from visual perception, feature extraction, cutting planning to real-time dynamic adjustment. This system can adaptively plan and adjust cutting parameters online based on individual differences in broilers, overcoming the technical shortcomings of existing technologies where fixed cutting programs cannot adapt to individual characteristics, resulting in large fluctuations in cutting quality and low yield. This significantly improves the cutting accuracy, cross-sectional quality, and overall yield of broilers with different body shapes and structures.

[0030] Furthermore, this invention acquires surface images and calculates contour geometric parameters and defect characterization values ​​through online grading visual units, and determines commercial grades by combining preset grading thresholds. This enables rapid, objective, and automated grading of broiler body size and appearance quality, providing accurate front-end input for subsequent differentiated cutting quality control, and effectively replacing the subjectivity and inconsistency of manual grading.

[0031] Furthermore, this invention obtains multi-view contour images and reconstructs a three-dimensional model by segmenting feature extraction units, calculates feature parameters reflecting the ratio of the thorax and leg regions and the trunk width-to-height ratio, and then fuses them to obtain the trunk density distribution index, which can effectively characterize the weight distribution and skeletal structure characteristics of individual broilers, providing a quantitative basis for predicting cutting mechanical behavior.

[0032] Furthermore, this invention dynamically determines the initial cutting parameters and desired pressure range by optimizing the integrated grading signal and trunk density distribution index of the control unit, and adjusts the sawing path and feed speed in conjunction with the real-time feedback from the pressure sensing unit during the cutting process, so that the cutting process can adapt to the real-time changes in tissue resistance and maintain the cutting force in the optimal range.

[0033] Furthermore, this invention combines a multi-axis linkage bone saw module with a pressure sensing unit and an optimization control unit to achieve coordinated dynamic adjustment of the cutting path and cutting force. The pressure deviation is not only used to adjust the feed speed to control the magnitude of the force, but its absolute value is also used to trigger compensation correction of the downward saw offset angle, thereby simultaneously optimizing force control and trajectory control in the cutting process and improving the adaptability to cutting irregular parts of the bone.

[0034] Furthermore, this invention modulates the desired pressure range and angle compensation amount based on the quality coefficient determined by the commercial grade, enabling the system to implement differentiated process control strategies for broilers of different value grades. While ensuring the cutting precision of high-grade products, it also takes into account the processing efficiency of low-grade products, achieving an optimal balance between economic benefits and process requirements. Attached Figure Description

[0035] Figure 1 This is a connection block diagram of the machine vision-based online grading and intelligent segmentation system for broilers according to the present invention.

[0036] Figure 2 This is a schematic diagram of the first and second projected areas of a broiler chicken in the machine vision-based online grading and intelligent segmentation system of the present invention.

[0037] Figure 3 This is a logic block diagram of the machine vision-based online grading and intelligent segmentation system for broilers, used in this invention to determine the initial sawing offset angle and the desired pressure range.

[0038] Figure 4 This is a logic block diagram of the pressure feedback-based dynamic adjustment in a single cut of the machine vision-based online grading and intelligent segmentation system for broilers in this invention.

[0039] In the figure, 1 - first projected area; 2 - second projected area. Detailed Implementation

[0040] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0041] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0042] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate the direction or positional relationship, are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0043] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0044] Please see Figure 1 The diagram shown is a connection block diagram of the machine vision-based online grading and intelligent segmentation system for broilers according to the present invention. The present invention provides a machine vision-based online grading and intelligent segmentation system for broilers, comprising:

[0045] The online grading vision unit includes an image acquisition module for acquiring surface images of broiler chickens on a suspended conveyor line and a first feature calculation module for obtaining a grading signal corresponding to a preset commercial grade based on the surface images.

[0046] Specifically, the image acquisition module is located above and to the side of the suspended conveyor line to acquire surface images that include the complete outer contour of the broiler chicken and the main body surface.

[0047] In one specific embodiment, the image acquisition module is configured to have three area-array industrial cameras evenly spaced along the conveying direction directly above the overhead conveyor, with their optical axes perpendicular to the conveying plane, to capture the back and side profiles of the broiler chickens; and a line-array industrial camera positioned at a 45-degree angle to the conveying direction on one side of the conveyor to scan the side profile of the broiler chickens as they pass. All cameras are equipped with a ring-shaped LED white light source to eliminate ambient light interference and provide uniform illumination. Preferably, the area-array industrial cameras have a resolution of at least 5 megapixels and a frame rate of at least 30fps; the line-array industrial cameras have a line scanning frequency of at least 10kHz. The triggering of the image acquisition module is synchronized with the encoder signal of the overhead conveyor to ensure image capture when the broiler chickens reach a preset position.

[0048] It is understandable that the above implementation uses a combination of area scan and line scan cameras to economically and reliably capture images of the complete outer contour and main body surfaces (such as the chest, back, and legs) of the broiler chicken from the top and sides without any blind spots. The area scan camera group at the top is responsible for quickly acquiring the overall top view contour and back features, while the line scan cameras on the sides continuously scan as the broiler chicken moves to acquire its side view contour. The combination of the two provides complete two-dimensional projection information for subsequent hierarchical calculations.

[0049] Please see Figure 3 As shown, it is a logic block diagram of a machine vision-based online grading and intelligent segmentation system for broilers to determine the initial sawing offset angle and the desired pressure range.

[0050] Specifically, the first feature calculation module processes the surface image, obtains the contour geometric parameters that characterize the overall size of the broiler chicken through contour fitting, and determines the defect feature region in the epidermis region through image segmentation.

[0051] The contour geometric parameters are quantized based on the contour geometric parameters, and the defect characterization value is calculated based on the defect feature region.

[0052] The grading signal is determined based on the comparison results of the quantized value of the contour geometric parameters and the first preset grading threshold, and the comparison results of the defect characterization value and the second preset grading threshold.

[0053] In one specific embodiment, contour fitting and parameter extraction are performed. For the binarized image after denoising and background segmentation, a contour tracking algorithm (such as the Suzuki85 algorithm) is used to extract the maximum connected component contour of the broiler body. Based on this contour, its minimum bounding rectangle is calculated. The quantized value Qg of the contour geometric parameter is defined as being determined by the aspect ratio of the bounding rectangle and the projected area of ​​the contour, and the calculation formula is as follows:

[0054] Q g =k1×R+k2×log10 (A / A) ref );

[0055] Among them, Q g R represents the quantized values ​​of the contour geometric parameters, dimensionless; R is the ratio of the longer side to the shorter side of the smallest bounding rectangle (aspect ratio), dimensionless; A is the projected area represented by the total number of contour pixels, in pixels; A ref The reference area is expressed in pixels and is calibrated based on the average projected area of ​​standard broiler chickens; k1 is the aspect ratio weighting coefficient, dimensionless, ranging from 0.6 to 0.8, preferably 0.7, calibrated based on statistical analysis of the contribution of aspect ratio to weight grading in historical data; k2 is the area weighting coefficient, dimensionless, ranging from 0.2 to 0.4, preferably 0.3, calibrated based on statistical analysis of the contribution of area to body size grading. 10 It is a base-10 logarithmic function used to compress the dynamic range of area data, making it numerically compatible with the aspect ratio parameter.

[0056] Understandably, the formula obtains a comprehensive quantitative value for body size by weightedly combining the logarithms of the aspect ratio and area. The aspect ratio reflects the leanness or elongation of the broiler, while the area reflects its overall size. The aspect ratio is given a higher weight because, at similar weights, length and width are more sensitive characteristics for distinguishing breeds and rearing stages. The logarithmic operation prevents an excessively large area value from dominating the weighted calculation, ensuring that both characteristics effectively participate in the grading process.

[0057] Secondly, defect feature region segmentation and characterization are performed. The original RGB image is converted to the HSV color space, and the saturation (S) and lightness (V) channels are extracted. For suspected defect regions (such as bruises appearing as dark red, and feather residue appearing as bright, discolored areas), adaptive thresholding combined with morphological opening and closing operations is used for extraction, resulting in a binarized defect mask image. The defect characterization value Qd is defined as:

[0058] Q d =A d / A×100%;

[0059] Among them, Q d This is a defect characterization value, in units of %; A d A represents the total number of pixels in all defect areas of the defect mask image, in pixels; A represents the outline projection area of ​​the aforementioned broiler chicken body, in pixels. d This represents the percentage of the defective area, expressed as a percentage (%). A higher value indicates a more severe epidermal defect.

[0060] Understandably, this formula uses the percentage of the defective area relative to the main body area of ​​the broiler chicken to quantify the severity of the defect. This is an objective and repeatable metric. Compared to simply determining whether a defect exists, the area percentage can better distinguish between minor and serious defects, providing a basis for refined grading.

[0061] Finally, the grading signal is determined. The system has a first preset grading threshold and a second preset grading threshold pre-stored. In this implementation, the first preset grading threshold includes a first-level contour grading threshold and a second-level contour grading threshold, which are used to divide the business level into three levels. The grading logic is as follows:

[0062] If Q g ≥Tg2 and Q d If ≤Td, then the graded signal corresponds to the highest commercial grade (e.g., Grade A).

[0063] If Tg1≤Q g <Tg2 and Q d If ≤Td, then the graded signal corresponds to the intermediate business level (such as Grade B).

[0064] If Q g <Tg1 or Q d If the value is greater than Td, then the graded signal corresponds to a lower business grade (such as Grade C).

[0065] Wherein, Tg1 is the first-level contour grading threshold, dimensionless, with a value range of 1.5 to 3, preferably 2.2; Tg2 is the second-level contour grading threshold, dimensionless, with a value range of 3.5 to 5.5, preferably 4.5; and Tg1 < Tg2; Tg1 and Tg2 are determined by cluster analysis of the contour geometric parameter quantification value Qg of a large number of samples according to the target market for different body sizes of broiler chickens; Td is the second preset grading threshold, in percentage, with a value range of usually 0.5% to 2.0%, preferably 1%, which is determined according to the maximum tolerance of consumers or downstream processors for skin defects.

[0066] Understandably, the quality of broiler chickens is primarily determined by their body size and skin integrity. The quantified geometric parameters obtained through contour fitting comprehensively reflect the size and shape of the broiler chicken, serving as an indirect indicator of meat production potential. Defect characterization values ​​obtained through color space analysis and image segmentation directly quantify skin problems affecting product appearance and processing losses. By comparing these two calculated quantified values ​​with preset thresholds based on market rules or process requirements, the system simulates the judgment logic of experienced graders, mapping continuous visual features to discrete, commercially meaningful grading signals. This avoids subjectivity, ensures a high degree of consistency in grading standards across different batches and time periods, and structures the visual information, providing clear decision-making input for downstream intelligent segmentation systems.

[0067] The segmentation feature extraction unit is set at the subsequent station of the online hierarchical vision unit. It includes a multi-view acquisition module for acquiring multi-view contour images of the broiler chicken, and a second feature calculation module for calculating the trunk density distribution index, which characterizes the trunk weight distribution and the position of the skeletal center of gravity of the broiler chicken, from the multi-view contour images.

[0068] Specifically, the multi-view acquisition module includes at least one first camera disposed on one side of the suspended conveyor line, and at least one second camera disposed at a non-zero angle to the optical axis of the first camera, for synchronously acquiring the multi-view contour image of the broiler chicken.

[0069] The second feature calculation module is connected to the multi-view acquisition module to receive and process the multi-view contour image.

[0070] In a specific embodiment, the multi-view acquisition module is implemented as follows: Two area array industrial cameras are installed at the segmentation feature extraction station of the suspended conveyor line. The first camera is installed on one side of the conveyor line, with its optical axis horizontal and perpendicular to the conveyor line, and its center aligned with the midpoint of the broiler suspension point on the conveyor line. The second camera is installed diagonally above the conveyor line, and its installation position simultaneously satisfies the following two conditions: on the horizontal plane, the optical axis of the second camera forms an angle with the direction of the conveyor line, the angle range being 25° to 35°, preferably 30°; on the vertical plane, the optical axis of the second camera forms a downward angle with the horizontal plane, the downward angle range being 40° to 50°, preferably 45°; both cameras are equipped with an infrared LED array light source with a wavelength of 850nm, and the light source operates in pulse mode with a pulse width of 100μm. The trigger signals for the cameras and light sources are provided by the Z-phase signal of the encoder of the conveyor line servo motor, which corresponds to the preset position of the broiler reaching the center point of the station. Preferably, both cameras have a resolution of 5 megapixels, a global shutter, and a frame rate of 60 frames per second. At a working distance of 500nm, the illuminance of the light source is not less than 1000lx.

[0071] Understandably, the first camera provides a side-view projection perpendicular to the conveying direction, enabling the acquisition of the broiler's lateral height, body length, and lateral profile. The second camera, mounted at a specific horizontal and downward angle, simultaneously acquires the broiler's back profile, part of its forechest profile, and the top-view profile of the leg-hip connection area. This perspective compensates for the lack of depth information in a purely side-view view. Hard synchronization is achieved via encoder signals, ensuring simultaneous exposure of both cameras at precisely aligned broiler positions. The resulting two images have a definite spatial correspondence, providing a foundation for subsequent contour information fusion based on parallax or triangulation principles. Using an infrared light source reduces interference from the visible light environment, and the pulse operating mode eliminates motion blur. The specific installation angle range has been verified to effectively cover the main characteristic areas of the broiler's body within a typical chicken processing plant's installation space while avoiding obstruction by its own mechanical structure.

[0072] Specifically, the second feature calculation module processes the multi-view contour image to obtain a three-dimensional contour model of the broiler's torso.

[0073] Based on the three-dimensional contour model, the first projected area of ​​the chest region and the second projected area of ​​the leg and hip region are calculated respectively, and the first feature parameter is determined according to the ratio of the first projected area to the second projected area.

[0074] Based on the three-dimensional contour model, the second feature parameter is determined according to the aspect ratio of the broiler body contour;

[0075] The trunk density distribution index is obtained based on the first feature parameter and the second feature parameter.

[0076] Specifically, the trunk density distribution index is positively correlated with the first feature parameter and negatively correlated with the second feature parameter.

[0077] In a specific embodiment, the process of the second feature calculation module processing and calculating the multi-view contour image is as follows:

[0078] First, based on images simultaneously acquired from a first and second camera, epipolar constraints and triangulation principles from binocular vision are used to match corresponding points on the broiler torso contour in the two images. The 3D coordinates of these points in the camera coordinate system are then calculated to generate a 3D point cloud of the torso surface. Subsequently, the Poisson surface reconstruction algorithm is used to convert this 3D point cloud into a continuous 3D triangular mesh model, i.e., the 3D contour model of the broiler torso.

[0079] Next, define the regions required for feature calculation on the 3D contour model. Place the model in a virtual coordinate system, with the chicken's head facing the positive X-axis and its back facing upwards along the positive Z-axis. Define the thoracic region as the area enclosed by the projected contours of the body segments corresponding to the last cervical vertebra and the last rib in the XZ plane (i.e., the sagittal plane). Define the leg and hip regions as the areas enclosed by the projected contours of the two body segments corresponding to the hip joint and the knee joint in the XZ plane. Calculate the pixel area of ​​the thoracic region within the projected contours of the XZ plane, and denote the first projected area as A. c (Unit: pixels). Calculate the pixel area of ​​the hip region within the XZ plane projection contour, and denote the second projection area 2 as A. l (Unit: pixels). The first feature parameter F1 is then calculated using the following formula:

[0080] F1=A c / A l ;

[0081] Where F1 is a dimensionless ratio.

[0082] Then, calculate the second feature parameter. Obtain the projection of the 3D contour model onto the XY plane (i.e., the coronal plane) and calculate its minimum bounding rectangle. The width (along the Y-axis) of this rectangle is denoted as W, and the height (along the Z-axis) is denoted as H. The second feature parameter F2 is calculated using the following formula:

[0083] F2 = W / H;

[0084] Where F2 is the dimensionless aspect ratio.

[0085] Finally, based on the first feature parameter F1 and the second feature parameter F2, the trunk density distribution index I is calculated. d The calculation formula is:

[0086] I d =α×ln(F1+1)-β×F2;

[0087] Among them, I d The trunk density distribution index is a dimensionless, comprehensive quantitative value. ln is the natural logarithm function. α is the weighting coefficient of the first characteristic parameter, dimensionless, ranging from 1.5 to 2.5, preferably 2, and is determined based on the correlation statistical analysis of F1 and the weight ratio of broiler breast muscles in historical data. β is the weighting coefficient of the second characteristic parameter, dimensionless, ranging from 0.8 to 1.2, preferably 1, and is determined based on the correlation statistical analysis of F2 and the height of the broiler's center of gravity (affecting cutting stability) in historical data. "+1" is added to ensure the logarithm is positive.

[0088] Understandably, this formula synthesizes a comprehensive index by weighted combination of two independent morphological features. Taking the logarithm of the first feature parameter F1 (chest-to-leg area ratio) linearizes its potential exponential growth relationship (such as the nonlinear relationship between pectoral muscle development and area), making it more stable in calculation. The coefficients α and β determine the contribution weights of the two features in the final index, and their calibration is based on the historical statistical correlation strength with the target physical characteristics (meat yield, center of gravity). This calculation merges two geometric parameters with different physical meanings (one reflecting the relative development of the anterior and posterior parts, and the other reflecting the overall width and flatness) into a scalar, aiming to summarize the main morphological factors influencing the cutting strategy with a single numerical value.

[0089] Understandably, the morphological characteristics of broiler chickens are correlated with the distribution of their internal muscles (primarily pectoral muscles) and bones. A relatively larger projected area of ​​the thorax region in the sagittal plane generally indicates more developed pectoral muscles, potentially a higher weight percentage, and a more prominent sternal structure, which affects skeletal resistance and meat distribution during cutting. The projected area of ​​the leg and hip regions reflects the weight of the hind limb muscles. The ratio of these two (F1), as a relative indicator, can partially offset the influence of absolute body size, focusing more on the weight distribution ratio between the fore and posterior sections of the trunk. On the other hand, the width-to-height ratio of the trunk in the coronal plane (F2) describes the breadth and flatness of the body. A wider body shape usually means a lower center of gravity and a more laterally extended skeletal structure, which mechanically may mean different spatial orientations and stress stability of the bones along the cutting path. Introducing F2 into the exponent through a negative correlation suggests that, for the same F1, a wider and flatter body shape (larger F2) may correspond to a relatively different skeletal spatial layout or center of gravity, and its "density distribution" characteristics need to be adjusted. Therefore, by combining F1 and F2 with specific weights, the resulting trunk density distribution index I is obtained. d The aim is to comprehensively predict the differences between individual broiler chickens and standard models in terms of bone stiffness distribution, muscle weight distribution, and body stability from a geometric morphological perspective, so as to provide a fused and more representative morphological input for subsequent prediction of cutting mechanical parameters (sternal stiffness offset).

[0090] The multi-axis cutting execution unit includes a suspension conveying module for clamping and moving broilers, and a multi-axis linkage bone sawing module for sawing broilers.

[0091] In one specific embodiment, the suspended conveyor module includes two parallel annular chain guides, a servo motor driving the chain to move synchronously, and hooks for attaching the chicken's feet, with the hooks fixed to the chain at equal intervals. The servo motor is equipped with a high-resolution absolute encoder for feedback and control of the chain displacement. Preferably, the chain pitch is 25mm to 30mm, the hook spacing is 200mm, and the servo motor positioning accuracy is not less than ±0.1mm.

[0092] The multi-axis linkage saw bone module is fixedly mounted on a frame on one side of the suspended conveyor module, located at the cutting station. This module includes a three-axis Cartesian robot, whose X-axis (horizontal transverse), Y-axis (horizontal longitudinal), and Z-axis (vertical direction) are all driven by ball screws and their respective servo motors. A disc saw blade driven by a variable frequency motor is connected to the end of the Z-axis via a floating mount. The floating mount integrates linear guides and a return spring, allowing the saw blade to produce limited elastic displacement in the direction perpendicular to its cutting plane (i.e., the feed pressure direction). The repeatability of the three-axis Cartesian robot is no less than ±0.05 mm.

[0093] A pressure sensing unit is installed on the multi-axis linkage saw bone module to detect the contact pressure feedback value between the saw blade and the broiler tissue in real time during the sawing process.

[0094] In one specific embodiment, the pressure sensing unit employs a spoke-type load cell, which is directly mounted between the floating mounting base and the Z-axis end connecting plate. The sensor has a range of 0 to 200 Newtons, with a comprehensive error not exceeding ±0.1% of full scale. The sensor detects and outputs in real time the force signal generated by the contact between the saw blade and the broiler tissue, along the direction of the saw blade feed pressure. This signal is converted by a transmitter into a standard 4–20 mA analog current signal, which is the contact pressure feedback value.

[0095] The optimization control unit is connected to the online hierarchical vision unit, the segmentation feature extraction unit, the multi-axis cutting execution unit, and the pressure sensing unit, respectively.

[0096] The cutting quality target library is used to determine the corresponding cutting accuracy and loss rate required for different commercial grades of broilers based on the grading signal, and the initial sternal stiffness offset and initial downward saw offset angle are calculated based on the trunk density distribution index and the quality coefficient determined by the cutting quality target library.

[0097] The desired pressure range is determined based on the initial sternal stiffness offset and the quality coefficient.

[0098] During a single sawing process, the multi-axis linkage sawing module is controlled to start sawing based on the initial downward sawing offset angle, and the contact pressure feedback value is obtained. Based on the comparison result between the contact pressure feedback value and the desired pressure range, the sawing path and sawing feed speed of the multi-axis linkage sawing module are adjusted so that the contact pressure feedback value approaches the desired pressure range.

[0099] Please continue reading. Figure 4As shown, it is a logic block diagram of dynamic adjustment based on pressure feedback in a single cut of a machine vision-based online grading and intelligent segmentation system for broilers according to the present invention.

[0100] Specifically, the optimization control unit determines the median and half-width of the basic pressure range based on the initial sternal stiffness offset, and corrects the median and half-width of the basic pressure range based on the quality coefficient to obtain the median and half-width of the actual pressure range of the desired pressure range.

[0101] Specifically, the initial sternal stiffness offset is positively correlated with the trunk density distribution index, and the initial downward saw offset angle is positively correlated with the trunk density distribution index.

[0102] In a specific embodiment, the optimization control unit determines the desired pressure range based on the initial sternal stiffness offset and the quality coefficient as follows:

[0103] First, the system pre-stores a library of cutting quality targets corresponding to different grading signals (commercial grades). Each target library defines a quality coefficient Qc. In this embodiment, when the grading signal is grade A, the corresponding Qc = 0.8; when the grading signal is grade B, the corresponding Qc = 1; and when the grading signal is grade C, the corresponding Qc = 1.2. This quality coefficient Qc is a dimensionless number, and its value is determined comprehensively through process experiments and economic analysis based on the maximum allowable cutting loss rate and the minimum required cutting surface quality standard for each grade of broiler product.

[0104] Let the initial sternal stiffness offset be ΔS (dimensionless). The midpoint Pmidb of the basic pressure interval and the half-width Wb of the basic pressure interval are calculated using the following formula:

[0105] Pmidb = Pstd + kp × ΔS;

[0106] Wb=Wstd;

[0107] Wherein, Pmidb is the median of the basic pressure range, in Newtons (N); Wb is the half-width of the basic pressure range, in Newtons (N). Pstd is the median of the standard pressure, in Newtons (N), ranging from 30N to 50N, preferably 40N, calibrated based on the average sawing force measured for a standard chicken shape under optimal cutting parameters. kp is the stiffness-pressure conversion factor, in Newtons per unit offset (N / L), ranging from 5N / L to 15N / L, preferably 10N / L, calibrated based on the regression relationship between the predicted sternal stiffness and the actual peak cutting force in historical data. Wstd is the half-width of the standard pressure, in Newtons (N), ranging from 5N to 10N, preferably 7.5N, calibrated based on the allowable range of cutting force fluctuation.

[0108] Understandably, the dimensionless sternal stiffness offset ΔS is linearly converted into a force increment via a coefficient kp and superimposed on the standard pressure median Pstd. The principle is that the predicted increase in bone stiffness requires a higher baseline cutting force to maintain effective sawing. The base half-width Wb is set as a constant Wstd independent of the stiffness offset. This simplifies the model, based on the assumption that the force fluctuation range is primarily determined by inherent factors such as system mechanical clearance and meat tissue homogeneity, rather than the predicted bone stiffness.

[0109] Then, the baseline value is corrected using the quality coefficient Qc to obtain the actual pressure range midpoint Pmid and the actual pressure range half-width W of the desired pressure range:

[0110] Pmid = Qc × Pmidb;

[0111] W = Qc × Wb;

[0112] Wherein, Pmid is the median of the actual pressure range, in Newtons (N); W is the half-width of the actual pressure range, in Newtons (N). Accordingly, the desired pressure range is defined as [Pmid-W, Pmid+W].

[0113] Understandably, the quality factor Qc is used to scale the baseline median and half-width proportionally. When Qc < 1 (corresponding to a high level), the actual interval median decreases and the interval width narrows, meaning the system will control with smaller, more precise cutting forces, aiming to reduce tissue damage and improve cut surface quality. When Qc > 1 (corresponding to a low level), the actual interval median increases and the interval widens, allowing for greater force fluctuations to prioritize cutting efficiency and passage through abnormal bone.

[0114] Understandably, the purpose of this implementation is to combine the classification information (grading signal) acquired by the pre-stage vision system with the morphological prediction information (sternal stiffness offset) to transform it into a quantitative and differentiated preset target for the mechanical parameters of the sawing process. The principle is that different commercial grades inherently represent different quality-cost control requirements. High-grade products require higher yield and appearance integrity, thus requiring more precise and conservative force parameters in cutting control to minimize tissue tearing and bone fragmentation; low-grade products prioritize processing efficiency and yield, tolerating a wider force control range and higher baseline force. The sternal stiffness offset is a prediction of individual bone physical characteristics differences. A higher predicted value means greater resistance during cutting, thus requiring a corresponding increase in the preset median force to adapt to this change and avoid cutting stagnation or accelerated tool wear due to excessively low preset force. By scaling the base pressure range calculated from the stiffness offset using a quality coefficient representing the grade requirements, the final desired pressure range is a dynamic control target that simultaneously considers individual physical characteristic predictions and commercial grade process requirements. It enables the same equipment to adaptively implement different levels of precision in cutting broilers of different values, providing a core setpoint basis for achieving differentiated and economically optimal intelligent segmentation.

[0115] Specifically, the optimization control unit calculates the pressure deviation between the contact pressure feedback value and the midpoint of the actual pressure range of the desired pressure range;

[0116] The direction of adjustment of the sawing feed speed is determined based on the sign of the pressure deviation value; wherein, when the deviation value is positive, the sawing feed speed is decreased, and when the deviation value is negative, the sawing feed speed is increased.

[0117] Based on the absolute value of the deviation, the adjustment range of the sawing feed speed and the compensation adjustment amount for the downward saw offset angle are determined; wherein, both the adjustment range and the compensation adjustment amount are positively correlated with the absolute value of the deviation.

[0118] Specifically, the optimization control unit calculates the initial compensation angle based on the absolute value of the pressure deviation; and corrects the initial compensation angle based on the quality coefficient to determine the compensation adjustment amount.

[0119] Specifically, the initial sternal stiffness offset is positively correlated with the trunk density distribution index, and the initial downward saw offset angle is positively correlated with the trunk density distribution index.

[0120] In one specific embodiment, the dynamic adjustment of the optimization control unit during a single sawing process is implemented as follows:

[0121] First, determine the pressure deviation value. Let the contact pressure feedback value obtained in the current sampling period be Ffb (in Newtons, N), and the midpoint of the actual pressure range of the desired pressure range be Pmid (in Newtons, N). Then, the pressure deviation value ΔP is calculated using the following formula:

[0122] ΔP = Ffb - Pmid;

[0123] Wherein, ΔP is the pressure deviation value, and the unit is Newton (N).

[0124] Next, adjust the sawing feed speed. Let the current sawing feed speed be Vf (in millimeters per second). The adjustment direction is determined by the sign of ΔP: if ΔP > 0, decrease Vf; if ΔP < 0, increase Vf. The adjustment range ΔV is calculated using the following formula:

[0125] ΔV = -kv × ΔP;

[0126] Where ΔV is the feed rate adjustment amount, in millimeters per second (mm / s). kv is the speed adjustment coefficient, in millimeters per second per Newton (mm / (s·N)), with a value range of 0.05mm / (s·N) to 0.2mm / (s·N), preferably 0.1mm / (s·N), which is calibrated according to the control system response speed and sawing process stability requirements. The negative sign indicates the aforementioned adjustment direction (the adjustment amount is negative when the deviation is positive, i.e., deceleration). The adjusted feed rate Vf′ = Vf + ΔV, and must be limited between the preset minimum and maximum safe feed rates.

[0127] Next, the compensation adjustment amount for the downward saw offset angle is calculated. The calculation of the compensation adjustment amount Δθc is divided into two steps. First, the initial compensation angle Δθinit is calculated based on the absolute value of the pressure deviation:

[0128] Δθinit=kθ×|ΔP|;

[0129] Wherein, Δθinit is the initial compensation angle, in degrees (°). kθ is the angle compensation coefficient, in degrees per Newton (° / N), with a value range of 0.005° / N to 0.02° / N, preferably 0.01° / N, which is determined based on calibration tests of the influence of saw blade geometry and path deviation on cutting force.

[0130] Then, the initial compensation angle is corrected using the quality coefficient Qc to obtain the final compensation adjustment Δθc:

[0131] Δθc = Qc × Δθinit;

[0132] Where Δθc is the compensation adjustment amount, in degrees (°). This adjustment amount will be superimposed on the current sawing path to fine-tune the spatial attitude of the saw blade in real time.

[0133] Understandably, the coefficient kv determines the control sensitivity; an excessively large value will cause oscillations, while a too small value will result in a slow response. In the angle compensation section, the initial compensation angle is proportional to the absolute value of the pressure deviation. The principle is that a large force deviation may mean a significant deviation between the current sawing path and the optimal cutting surface of the bone, requiring spatial path compensation correction by changing the saw blade angle (i.e., the downward saw offset angle). The coefficient kθ determines the gain of the force deviation on the angle correction. Finally, the quality coefficient Qc is used to scale the angle compensation amount, resulting in finer angle correction for high-grade products (smaller Qc) to achieve more precise path control; and larger angle correction for low-grade products (larger Qc) to prioritize quickly overcoming resistance.

[0134] In one specific embodiment, the initial sternal stiffness offset ΔS and the initial downward saw offset angle θinit are based on the trunk density distribution index I. d The determined implementation method is as follows:

[0135] ΔS=ks×(I d -Istd);

[0136] θinit=θstd+kθinit×(I d -Istd);

[0137] Where ΔS is the initial sternal stiffness offset, dimensionless; θinit is the initial downward saw offset angle, in degrees (°). Istd is the reference value of the trunk density distribution index of the standard broiler model, dimensionless, based on a large number of samples. d Average value calibration. ks is the stiffness offset coefficient, with units of 1 / 1 (i.e., dimensionless), and a value range of 0.5 to 1.5, preferably 1, based on I. d The regression relationship between the θstd and bone stiffness parameters obtained through offline biomechanical measurements is calibrated. θstd is the standard downward saw offset angle, in degrees (°), ranging from 0° to 5°, preferably 2°, calibrated based on the optimal cutting path of the standard chicken shape. kθinit is the initial angle coefficient, in degrees per unit exponent (° / 1), ranging from 0.5° / 1 to 2.0° / 1, preferably 1° / 1, calibrated based on I... d The empirical relationship with the optimal bone-avoiding cutting angle is calibrated. In the formula (I) d The -Istd) term represents the exponential difference between the current individual and the standard model. The greater the difference, the greater the predicted stiffness shift and the required angle pre-adjustment.

[0138] Understandably, these two formulas establish a quantitative relationship between visual morphological features and two key pre-defined control parameters. Trunk density distribution index I dAs a comprehensive morphological representation, its deviation from the standard reference value is used to linearly predict deviations in bone physical properties and pre-adjustments to the cutting path. This enables vision-based, forward-looking initialization of cutting parameters.

[0139] Understandably, the cutting process is a dynamic interaction of forces, and relying solely on initial preset parameters is insufficient to address the inhomogeneities within the tissue and model prediction errors. By measuring the sawing force in real time and comparing it with a dynamic desired pressure range, the system can obtain direct feedback on the process status. The sign of the pressure deviation directly indicates whether the current feed rate is too fast (resulting in excessive force) or too slow (resulting in insufficient force). Therefore, adjusting the feed rate directly based on the sign and magnitude of the deviation is the most direct and effective force control method, regulating the interaction force by changing the contact rate between the saw blade and the material. Simultaneously, abnormal force changes may also stem from an suboptimal saw path, such as cutting directly into harder bone or an unfavorable angle leading to increased friction. Therefore, the absolute value of the pressure deviation is used as the basis for triggering path fine-tuning (downward saw offset angle compensation). The larger the deviation, the greater the degree to which the initial path assumption may deviate from reality, requiring bolder angle corrections to attempt to bring the cutting force back to the normal range. However, the radicalness of this correction needs to be constrained by product grade, and therefore modulated using a quality coefficient. High-grade products require stable cutting and precise paths, so even if the force deviation is large, a relatively conservative angle correction is used to avoid distorting the cutting trajectory due to excessive correction; low-grade products prioritize quickly restoring the normal cutting force, allowing for more extensive angle correction attempts.

[0140] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A machine vision-based online grading and intelligent segmentation system for broiler chickens, characterized in that, include: An online grading vision unit includes an image acquisition module for acquiring surface images of broiler chickens on a suspended conveyor line and a first feature calculation module for obtaining a grading signal corresponding to a preset commercial grade based on the surface images. The first feature calculation module processes the surface images, obtains contour geometric parameters characterizing the overall size of the broiler chickens through contour fitting, and determines defect feature regions in the epidermal region through image segmentation; calculates quantized values ​​of the contour geometric parameters based on the contour geometric parameters, and calculates defect characterization values ​​based on the defect feature regions; and determines the grading signal based on the comparison results of the quantized values ​​of the contour geometric parameters with a first preset grading threshold and the comparison results of the defect characterization values ​​with a second preset grading threshold. The segmentation feature extraction unit, located at a subsequent station of the online hierarchical vision unit, includes a multi-view acquisition module for acquiring multi-view contour images of the broiler chicken, and a second feature calculation module for calculating a trunk density distribution index, which characterizes the weight distribution of the broiler chicken's trunk and the position of its skeletal center of gravity, from the multi-view contour images. The second feature calculation module processes the multi-view contour images to obtain a three-dimensional contour model of the broiler chicken's trunk. Based on the three-dimensional contour model, a first projected area of ​​the thorax region and a second projected area of ​​the leg and hip regions are calculated respectively. A first feature parameter is determined based on the ratio of the first projected area to the second projected area. Based on the three-dimensional contour model, a second feature parameter is determined according to the aspect ratio of the broiler's trunk contour; the trunk density distribution index is obtained based on the first feature parameter and the second feature parameter. The multi-axis cutting execution unit includes a suspension conveying module for clamping and moving broilers, and a multi-axis linkage bone sawing module for sawing broilers. A pressure sensing unit is installed on the multi-axis linkage saw bone module to detect the contact pressure feedback value between the saw blade and the broiler tissue in real time during the sawing process. An optimization control unit, connected to the online grading vision unit, the segmentation feature extraction unit, the multi-axis cutting execution unit, and the pressure sensing unit, determines the desired pressure range based on the obtained initial sternal stiffness offset and initial downward sawing offset angle. During a single sawing operation, it adjusts the sawing path and sawing feed speed of the multi-axis linkage bone sawing module based on the acquired contact pressure feedback value. The optimization control unit determines a corresponding cutting quality target library based on the grading signal, defining the required cutting accuracy and loss rate for different commercial grades of broiler chickens. It calculates the initial sternal stiffness offset and the initial downward sawing offset angle based on the trunk density distribution index and the quality coefficient determined by the cutting quality target library. Based on the initial sternal stiffness offset, it determines the base pressure range... The system calculates the midpoint and half-width of the base pressure range, and corrects these values ​​and half-widths based on the quality coefficient to obtain the midpoint and half-width of the actual pressure range for the desired pressure range. The optimization control unit calculates the pressure deviation between the contact pressure feedback value and the midpoint of the actual pressure range for the desired pressure range. The direction of adjustment for the sawing feed speed is determined based on the sign of the pressure deviation. When the deviation is positive, the sawing feed speed is decreased; when the deviation is negative, the sawing feed speed is increased. The adjustment range of the sawing feed speed and the compensation adjustment amount for the initial downward sawing offset angle are determined based on the absolute value of the deviation. Both the adjustment range and the compensation adjustment amount are positively correlated with the absolute value of the deviation. The optimization control unit calculates the initial compensation angle based on the absolute value of the pressure deviation; and corrects the initial compensation angle based on the quality coefficient to determine the compensation adjustment amount.

2. The machine vision-based online grading and intelligent segmentation system for broiler chickens according to claim 1, characterized in that, The trunk density distribution index is positively correlated with the first feature parameter and negatively correlated with the second feature parameter.

3. The machine vision-based online grading and intelligent segmentation system for broiler chickens according to claim 2, characterized in that, The initial sternal stiffness offset is positively correlated with the trunk density distribution index, and the initial downward saw offset angle is positively correlated with the trunk density distribution index.

4. The machine vision-based online grading and intelligent segmentation system for broiler chickens according to claim 1, characterized in that, The image acquisition module is located above and to the side of the suspended conveyor line to acquire surface images that include the complete outer contour of the broiler chicken and the main body surface.

5. The machine vision-based online grading and intelligent segmentation system for broiler chickens according to claim 1, characterized in that, The multi-view acquisition module includes at least one first camera disposed on one side of the suspended conveyor line, and at least one second camera disposed at a non-zero angle to the optical axis of the first camera, for synchronously acquiring the multi-view contour image of the broiler chicken. The second feature calculation module is connected to the multi-view acquisition module to receive and process the multi-view contour image.

Images