Sea shrimp cake forming quality detection method and system based on image recognition

Abstract

The invention discloses a sea shrimp cake forming quality detection method based on image recognition. Aiming at the problems of low manual sampling inspection efficiency, poor precision and defective product elimination lag, the method comprises the following steps: firstly, triggering a structured light 3D camera through a laser grating sensor to synchronously collect color images and depth data of shrimp cakes to generate a three-dimensional point cloud; and then shape error detection and weight error detection are synchronously executed: the diameter error is calculated through projection fitting of a circle after point cloud depth threshold segmentation and outlier filtering, the volume is calculated through layer-by-layer fitting of the circle of the point cloud according to thickness slices in weight detection, and the weight is dynamically calibrated in combination with a temperature compensation density model. And finally, if the diameter or weight error exceeds the limit, calculating the starting time of the vacuum chuck according to the conveying belt speed, the system delay and the mechanism position, and dynamically positioning and adsorbing and removing unqualified products. According to the invention, 100% online detection under the full speed of a production line is realized, the single processing time is less than 0.2 second, the product percent of pass is obvious, and the labor cost is obviously reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of food processing automation, and in particular to a method and system for detecting the molding quality of shrimp cakes based on image recognition. Background Art

[0002] In the current industrial production of shrimp cakes, molding quality inspection is highly dependent on manual sampling and offline weighing, which has significant defects: due to the soft texture of shrimp cakes and easy deformation, manual visual inspection is subject to subjective bias, and the insufficient sampling rate leads to a high rate of missed defective products; at the same time, traditional weight detection requires manual sampling and inspection, which affects the production process, and existing machine vision solutions (such as 2DRGB detection) are affected by surface reflection interference, resulting in large diameter measurement errors, and contact weighing mechanisms are prone to damage to the product structure. In addition, the density of shrimp cakes fluctuates with the temperature of the raw materials. The traditional volume-weight model does not incorporate temperature change compensation, resulting in large weight errors. The above problems cause delays in the removal of defective products and increase the risk of contamination in subsequent processes. There is an urgent need to develop a non-contact, high-precision, and fully automated online quality control solution. Summary of the Invention

[0003] The present invention is proposed to solve the above-mentioned problems, and its purpose is to provide a method and system for detecting the molding quality of shrimp cakes based on image recognition.

[0004] The present invention provides a shrimp cake molding quality detection method based on image recognition, which has the following characteristics and comprises the following steps: step 1, triggering acquisition: using a laser grating sensor to sense the round shrimp cakes on the conveyor belt in real time, when the shrimp cakes enter the field of view, triggering a structured light 3D camera to synchronously collect RGB images and depth data of the shrimp cakes, and generating three-dimensional point cloud data; step 2, parallel processing: extracting the appearance shape parameters of the shrimp cakes based on the three-dimensional point cloud data, calculating the diameter error, and performing shape error detection; at the same time, calculating the volume based on the point cloud data slices, and calculating the weight error in combination with a density adaptive model, and performing weight error detection; step 3, comprehensive judgment: if the diameter error is ≤1mm and the weight error is ≤1.0g, it is judged to be qualified; otherwise, it is judged to be unqualified and the unqualified shrimp cakes are recorded; step 4, rejecting unqualified products: calculating the time t when the shrimp cakes reach the vacuum suction cup based on the conveyor belt speed, the system delay time and the rejection mechanism coordinates. a , at t a The vacuum rejection system is always activated to remove unqualified shrimp cakes.

[0005] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following features: wherein, in step 1, the conveying speed of the conveyor belt is 200 mm / s; the parameters of the structured light 3D camera include: installation height 500 mm, resolution 1280×720; depth accuracy ±0.05 mm@0.5 m, using a blue LED light source with a wavelength of 459 nm; the generated point cloud data set is P={p i =(x i ,y i ,z i )|i=1,2,...,n}.

[0006] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following features: wherein, in step 2, shape error detection specifically includes the following steps: step 2-1, point cloud preprocessing: based on depth threshold Segment the point cloud data of the shrimp cakes and remove the points with a distance from the mean value greater than 3σ from the segmented shrimp cake point cloud subset. Step 2-2: Projection and fitting: Orthogonally project the preprocessed point cloud onto the XY plane, fit a circle using the least squares method, and calculate the center (a, b) and radius r. Step 2-3: Calculate the diameter error: ΔD shape =|rr design |, where r design is the standard radius, requiring ΔD shape ≤1mm.

[0007] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following features: wherein, in step 2-1, the pretreatment process includes the following steps:

[0008] Step 2-1-1, background segmentation: based on depth threshold The point cloud data of shrimp cakes is segmented, and the segmentation formula is:

[0009]

[0010] The point cloud of the shrimp cake after background segmentation is as follows:

[0011] S={p i ∈(x i ,y i ,z i )|i=1,2,...,M} (2)

[0012] Where S is the segmented shrimp cake point cloud subset, z iis the coordinate of the i-th point in the point cloud, i.e., the height coordinate, H is the base height of the shrimp cake, and δ is the thickness tolerance of the shrimp cake. Step 2-1-2, outlier filtering: remove points whose distance from the mean is greater than 3σ in the shrimp cake point cloud subset segmented by S, where σ is the standard deviation of the point cloud density. The outlier filtering formula is:

[0013] S filtered ={p i ∈S||p i -μ||≤3σ} (3)

[0014] Where μ is the mean center of the point cloud density distribution, σ is the standard deviation of the point cloud density, and ||·|| is the Euclidean distance.

[0015] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following features: wherein, in step 2-2, the projection and fitting process includes the following steps: Step 2-2-1, projection extraction: orthogonally projecting the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set:

[0016] C={q i ∈(x i ,y i )|i=1,2,...,nn} (4)

[0017] Step 2-2-2, circle fitting: Use the least squares method to fit the circle, set the center to (a, b), the radius to r, and the coordinates of the point cloud to (x i ,y i ), minimize the objective function:

[0018]

[0019] Using variable substitution:

[0020] c=a 2 +b 2 -r 2 (6)

[0021] The objective equation is transformed into:

[0022]

[0023] Construct a system of linear equations:

[0024]

[0025] Abbreviated as: AX=B,

[0026] Solve the equation using the least squares method:

[0027] X=(A T A)-1 A T B (9)

[0028] The solution is a, b, c;

[0029] Step 2-2-3, calculate the center and radius:

[0030]

[0031] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following features: wherein, in step 2, the weight error detection process includes the following steps: Step 2-4, for the acquired point cloud dataset S filtered ={q i ∈S||q i -μ||≤3σ}, determine the number of point cloud slices K and the slice thickness according to the reference height H of the shrimp cake The height interval of the kth slice is in:

[0032]

[0033] Where k = 1, 2, 3, ..., K; Step 2-5, extract the point cloud data set of the kth slice Projection extraction: Orthogonally project the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set. Use the least squares method to fit a circle and obtain the radius r. k , considering each slice as a thin cylindrical disk, the volume is:

[0034]

[0035] The total cumulative volume is:

[0036]

[0037] The shrimp cake density adaptive model compensates for density changes based on the temperature of the shrimp cake raw materials:

[0038] ρ=ρ0(1+αΔT) (15)

[0039] Where ρ0 is the base density of shrimp cakes, calculated based on raw materials, temperature coefficient α = -0.001 / °C, ΔT = 25°C-T. The weight calculation formula is:

[0040] m test =ρ·V (16)

[0041] The qualification judgment formula is:

[0042] |m test -mset |≤1g (17)

[0043] Where m test is the calculated weight, m set It is the set weight.

[0044] The shrimp cake forming quality detection method based on image recognition provided by the present invention may also have the following characteristics: wherein, in step 4, the time t when the shrimp cake reaches the vacuum suction cup a for:

[0045]

[0046] In t a Always start the vacuum system to remove shrimp cakes. The diameter of the vacuum suction cup should cover more than 80% of the diameter of the shrimp cake. a is the predicted time when the shrimp cakes reach the rejection point, X a is the X-axis coordinate of the rejection mechanism in the conveyor belt coordinate system, X 虾 is the original X-axis coordinate of the shrimp cake in the conveyor belt coordinate system, v is the speed of the conveyor belt along the X-axis, t d System processing latency is the time it takes to compute images and point clouds.

[0047] The shrimp cake forming quality inspection method based on image recognition provided by the present invention may also have the following characteristics: wherein, in step 4, the specific process of rejecting unqualified products is: during the continuous movement of the conveyor belt, a vacuum suction cup mechanism is used to achieve vacuum rejection during the conveyance of the shrimp cakes. The main factors include horizontal force and vertical force, wherein the vertical force includes overcoming gravity and acceleration inertia, and is expressed as:

[0048]

[0049] Where v1 is the vertical speed of the shrimp cake during vacuum removal, which is preset to 0.1-0.2m / s, Δt v is the vertical acceleration time of the shrimp culling process, generally 0.1s, g is the acceleration due to gravity, and the horizontal force is:

[0050]

[0051] Where Δt h Generally 0.1s; total suction force, including safety factor is:

[0052]

[0053] Where S f Use 1.5-2.0; vacuum parameter is expressed as vacuum degree P total :

[0054]

[0055] Where A is the area of ​​the suction cup in the vacuum rejection system, which is generally between 0.8 and 1.0 times the area of ​​the shrimp cake; μ is the safety factor, which is 0.6; and the air flow rate is:

[0056]

[0057] Where, ρ a is the air density, λ is the safety factor of 1.5, ρ is the density of shrimp cakes, is the friction coefficient between the shrimp cake and the conveyor belt.

[0058] The shrimp cake molding quality detection method based on image recognition provided by the present invention may also have the following characteristics: wherein, in step 4, at t a Within the time, the vacuum suction cup rejection mechanism should be positioned just above the predicted position of the center of the shrimp cake to be rejected, and ensure that the vacuum suction cup rejection mechanism is 3-5mm away from the surface of the shrimp cake.

[0059] The present invention also provides a shrimp cake forming quality detection system based on image recognition, which has the following characteristics: a trigger acquisition module, which uses a laser grating sensor to sense the round shrimp cakes on the conveyor belt in real time. When the shrimp cakes enter the field of view, the structured light 3D camera is triggered to synchronously collect the RGB image and depth data of the shrimp cakes to generate three-dimensional point cloud data; a parallel processing module, which extracts the appearance shape parameters of the shrimp cakes based on the three-dimensional point cloud data, calculates the diameter error, and performs shape error detection. At the same time, the volume is calculated based on the point cloud data slices, and the weight error is calculated in combination with the density adaptive model to perform weight error detection; a comprehensive judgment module, if the diameter error is ≤1mm and the weight error is ≤1.0g, it is judged to be qualified, otherwise it is judged to be unqualified and the unqualified shrimp cakes are recorded; a module for rejecting unqualified products, which calculates the time t when the shrimp cakes reach the vacuum suction cup based on the conveyor belt speed, system delay time and rejection mechanism coordinates a , at t a Always activate the vacuum rejection system to remove unqualified shrimp cakes. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the shrimp cake molding quality detection method based on image recognition in this embodiment. DETAILED DESCRIPTION

[0061] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0062] This embodiment provides a method for inspecting the quality of shrimp cakes using image recognition. This method is applicable to round shrimp cakes. This embodiment uses a structured light 3D camera to trigger the acquisition of RGB and depth data from shrimp cakes on a conveyor belt, generating point cloud data. By analyzing this point cloud data, the appearance and weight parameters of the shrimp cakes are determined to determine whether the shrimp cakes meet the product's dimensional and weight accuracy requirements. A vacuum rejection system is then used to remove shrimp cakes that do not meet these accuracy requirements from the conveyor belt.

[0063] Figure 1 This is a flow chart of the shrimp cake molding quality detection method based on image recognition in this embodiment.

[0064] like Figure 1 As shown, the shrimp cake molding quality detection method based on image recognition in this embodiment includes the following steps:

[0065] Step S1, triggering acquisition: The laser grating sensor is used to sense the round shrimp cakes on the conveyor belt in real time. When the shrimp cakes enter the field of view, the structured light 3D camera is triggered to synchronously collect the RGB image and depth data of the shrimp cakes to generate three-dimensional point cloud data.

[0066] In this embodiment, the conveyor belt has a conveying speed of 200 mm / s. The parameters of the structured light 3D camera are: installation height 500 mm, resolution 1280×720, depth accuracy ±0.05 mm @ 0.5 m, using a blue LED light source with a wavelength of 459 nm to enhance edge contrast.

[0067] The point cloud dataset generated in this embodiment is P={p i =(x i ,y i ,z i )|i=1,2,...,n}.

[0068] Step S2, parallel processing: extract the appearance shape parameters of the shrimp cake based on the three-dimensional point cloud data, calculate the diameter error, and perform shape error detection. At the same time, calculate the volume based on the point cloud data slices, and calculate the weight error in combination with the density adaptive model to perform weight error detection.

[0069] In this embodiment, shape error detection specifically includes the following steps:

[0070] Step S2-1, point cloud preprocessing: based on depth threshold Segment the point cloud data of shrimp cakes and remove points whose distance from the mean is greater than 3σ from the segmented shrimp cake point cloud subset. The specific process is as follows:

[0071] Step S2-1-1, background segmentation: based on depth threshold The point cloud data of shrimp cakes is segmented, and the segmentation formula is:

[0072]

[0073] The point cloud of the shrimp cake after background segmentation is as follows:

[0074] S={p i ∈(x i ,y i ,z i )|i=1,2,...,M} (2)

[0075] Where S is the segmented shrimp cake point cloud subset, z i is the coordinate of the i-th point in the point cloud, i.e., the height coordinate, H is the base height of the shrimp cake, and δ is the thickness tolerance of the shrimp cake.

[0076] Step S2-1-2, outlier filtering: remove points whose distance from the mean is greater than 3σ in the shrimp cake point cloud subset segmented by S, where σ is the standard deviation of the point cloud density. The outlier filtering formula is:

[0077] S filtered ={p i ∈S||p i -μ||≤3σ} (3)

[0078] Where μ is the mean center of the point cloud density distribution, σ is the standard deviation of the point cloud density, and ||·|| is the Euclidean distance.

[0079] Step S2-2, projection and fitting: orthogonally project the pre-processed point cloud onto the XY plane, fit a circle using the least squares method, and calculate the circle center (a, b) and radius r. The specific process is as follows:

[0080] Step S2-2-1, projection extraction: Orthogonally project the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set:

[0081] C={q i ∈(x i ,y i )|i=1,2,...,nn} (4)

[0082] Step S2-2-2, circle fitting: Use the least squares method to fit a circle, set the center to (a, b), the radius to r, and the coordinates of the point cloud to (x i ,y i ), minimize the objective function:

[0083]

[0084] Using variable substitution:

[0085] c=a 2 +b 2 -r 2 (6)

[0086] The objective equation is transformed into:

[0087]

[0088] Construct a system of linear equations:

[0089]

[0090] Abbreviated as: AX=B.

[0091] Use the least squares method to connect the equations:

[0092] X=(A T A) -1 A T B (9)

[0093] The solution is a, b, c;

[0094] Step S2-2-3, calculate the center and radius of the circle:

[0095]

[0096] Step S2-3, diameter error calculation: ΔD shape =|rr design |, where r design is the standard radius, requiring ΔD shape ≤1mm.

[0097] In this embodiment, the process of weight error detection includes the following steps:

[0098] Step S2-4, for the acquired point cloud dataset S filtered ={q i ∈S||q i -μ||≤3σ}, determine the number of point cloud slices K and the slice thickness according to the reference height H of the shrimp cake The height interval of the kth slice is in:

[0099]

[0100] Where k = 1, 2, 3, ..., K.

[0101] Step S2-5, extract the point cloud data set of the kth slice Projection extraction: Orthogonally project the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set. Use the least squares method to fit a circle and obtain the radius r. k .

[0102] Consider each slice as a thin cylindrical disk with a volume of:

[0103]

[0104] The total cumulative volume is:

[0105]

[0106] The shrimp cake density adaptive model compensates for density changes based on the temperature of the shrimp cake raw materials:

[0107] ρ=ρ0(1+αΔT) (15)

[0108] Where ρ0 is the reference density of shrimp cakes, which is calculated based on the raw materials, the temperature coefficient α = -0.001 / °C, and ΔT = 25°C-T.

[0109] The weight calculation formula is:

[0110] m test =ρ·V (16)

[0111] The qualification judgment formula is:

[0112] |m test -m set |≤1g (17)

[0113] Where m test is the calculated weight, m set It is the set weight.

[0114] Step S3, comprehensive judgment: if the diameter error is ≤1mm and the weight error is ≤1.0g, it is judged as qualified; otherwise, it is judged as unqualified and the unqualified shrimp cake is recorded.

[0115] Step S4, rejecting unqualified products: Calculate the time t when the shrimp cakes reach the vacuum suction cup based on the conveyor belt speed, system delay time and the coordinates of the rejection mechanism. a , at t a The vacuum rejection system is always activated to remove unqualified shrimp cakes.

[0116] In this embodiment, the time t for the shrimp cake to reach the vacuum suction cup is a for:

[0117]

[0118] In t a Always start the vacuum system to remove shrimp cakes. The diameter of the vacuum suction cup should cover more than 80% of the diameter of the shrimp cake. a is the predicted time when the shrimp cakes reach the rejection point, X a is the X-axis coordinate of the rejection mechanism in the conveyor belt coordinate system, X 虾 is the original X-axis coordinate of the shrimp cake in the conveyor belt coordinate system, v is the speed of the conveyor belt along the X-axis, t d is the system processing delay, i.e., the time it takes to compute the image and point clouds. a Within the time, the vacuum suction cup rejection mechanism should be positioned just above the predicted position of the center of the shrimp cake to be rejected, and ensure that the vacuum suction cup rejection mechanism is 3-5mm away from the surface of the shrimp cake.

[0119] The specific process of rejecting unqualified products is as follows:

[0120] During the continuous movement of the conveyor belt, a vacuum suction cup mechanism is used to achieve vacuum removal during the conveying process of shrimp cakes. The main factors include horizontal force and vertical force. The vertical force includes overcoming gravity and accelerating inertia, which can be expressed as:

[0121]

[0122] Where v1 is the vertical speed of the shrimp cake during vacuum removal, which is preset to 0.1-0.2m / s, Δt v is the vertical acceleration time of the shrimp culling process, which is generally 0.1s, and g is the acceleration due to gravity.

[0123] The horizontal force is:

[0124]

[0125] Where Δt h Generally it is 0.1s.

[0126] The total suction force, including safety factor, is:

[0127]

[0128] Where S f Use 1.5-2.0.

[0129] The vacuum parameter is expressed as vacuum degree P total :

[0130]

[0131] Where A is the area of ​​the suction cup in the vacuum rejection system, which is generally between 0.8 and 1.0 times the area of ​​the shrimp cake, and μ is the safety factor, which is 0.6.

[0132] The air flow rate is:

[0133]

[0134] Where, ρ a is the air density, λ is the safety factor of 1.5, ρ is the density of shrimp cakes, is the friction coefficient between the shrimp cake and the conveyor belt.

[0135] This embodiment also provides a shrimp cake molding quality detection system based on image recognition, comprising:

[0136] The acquisition module is triggered to sense the round shrimp cakes on the conveyor belt in real time through the laser grating sensor, and the three-dimensional point cloud data is generated using the method in step S1 of this embodiment.

[0137] The parallel processing module performs shape error detection and weight error detection using the method in step S2 of this embodiment.

[0138] The comprehensive determination module uses the method in step S3 of this embodiment to determine qualified and unqualified shrimp cakes.

[0139] The unqualified product rejection module uses the method in step S3 of this embodiment to reject unqualified shrimp cakes.

[0140] Functions and effects of the present invention:

[0141] This invention combines structured light 3D point cloud reconstruction with parallel processing technology to achieve simultaneous, high-precision dynamic detection of shrimp cake appearance and weight. Using a blue light (459nm) structured light camera and depth threshold segmentation (±0.05mm accuracy), diameter detection error is reduced to ≤1mm; combined with a point cloud slicing volume algorithm and a temperature-compensated density model, weight error is controlled to ≤1.0g. A laser grating triggering mechanism and multi-threaded collaboration reduce the inspection cycle for a single shrimp cake to less than 0.2 seconds, supporting 100% inspection at a production line speed of 200mm / s. The vacuum rejection mechanism, dynamically positioned using a spatiotemporal prediction model, significantly reduces rejection delay errors, saving labor costs, and improving raw material utilization through non-contact detection, comprehensively optimizing quality and production costs.

[0142] At the same time, in the present invention, the parallel execution of shape error detection and weight error detection shortens the processing time of a single shrimp cake to within 0.2 seconds, meeting the requirements of continuous production dynamic detection of product quality.

[0143] The above embodiments are preferred examples of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A method for detecting the quality of shrimp cakes based on image recognition, characterized in that: The following steps are involved: Step 1: Trigger acquisition: The laser grating sensor senses the round shrimp cakes on the conveyor belt in real time. When the shrimp cake enters the field of view, the structured light 3D camera is triggered to synchronously capture the RGB image and depth data of the shrimp cake, generating 3D point cloud data. Step 2, parallel processing: extracting the appearance shape parameters of the shrimp cake based on the three-dimensional point cloud data, calculating the diameter error, and performing shape error detection. At the same time, calculating the volume based on the point cloud data slices, calculating the weight error in combination with the density adaptive model, and performing weight error detection; Step 3, comprehensive judgment: If the diameter error is ≤1mm and the weight error is ≤1.0g, it is judged as qualified; otherwise, it is judged as unqualified and the unqualified shrimp cake is recorded; Step 4: Reject unqualified products: Calculate the time t when the shrimp cakes reach the vacuum suction cup based on the conveyor belt speed, system delay time and the coordinates of the rejection mechanism. a , at t a The vacuum rejection system is always activated to remove unqualified shrimp cakes.

2. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 1, the conveying speed of the conveyor belt is 200 mm / s. The parameters of the structured light 3D camera include: Installation height 500mm, resolution 1280×720, Depth accuracy ±0.05mm@0.5m, using blue LED light source, wavelength 459nm, The generated point cloud dataset is P = {p i =(x i ,y i ,z i )|i=1,2,...,n}.

3. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 2, the shape error detection specifically includes the following steps: Step 2-1, point cloud preprocessing: based on depth threshold Segment the point cloud data of the shrimp cakes and remove the points whose distance from the mean is greater than 3σ in the segmented shrimp cake point cloud subsets. Step 2-2, projection and fitting: orthogonally project the preprocessed point cloud onto the XY plane, fit a circle using the least squares method, and calculate the center (a, b) and radius r; Step 2-3, diameter error calculation: ΔD shape =|rr design |, where r design is the standard radius, requiring ΔD shape ≤1mm.

4. The method for detecting the quality of shrimp cakes based on image recognition according to claim 3, wherein: in, In step 2-1, the preprocessing process includes the following steps: Step 2-1-1, background segmentation: based on depth threshold The point cloud data of shrimp cakes is segmented, and the segmentation formula is: The point cloud of the shrimp cake after background segmentation is as follows: S={p i ∈(x i ,y i ,z i )|i=1,2,...,M} (2) Where S is the segmented shrimp cake point cloud subset, z i is the coordinate of the i-th point in the point cloud, i.e., the height coordinate, H is the base height of the shrimp cake, and δ is the thickness tolerance of the shrimp cake; Step 2-1-2, outlier filtering: remove points whose distance from the mean is greater than 3σ in the shrimp cake point cloud subset segmented by S, where σ is the standard deviation of the point cloud density. The outlier filtering formula is: S filtered ={p i ∈S||p i -μ||≤3σ} (3) Where μ is the mean center of the point cloud density distribution, σ is the standard deviation of the point cloud density, and ||·|| is the Euclidean distance.

5. The method for detecting the quality of shrimp cakes based on image recognition according to claim 3, wherein: in, In step 2-2, the projection and fitting process includes the following steps: Step 2-2-1, Projection Extraction: Orthogonally project the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set: C={q i ∈(x i ,y i )|i=1,2,...,nn} (4) Step 2-2-2, circle fitting: Use the least squares method to fit the circle, set the center to (a, b), the radius to r, and the coordinates of the point cloud to (x i ,y i ), minimize the objective function: Using variable substitution: c=a 2 +b 2 -r 2 (6) The objective equation is transformed into: Construct a system of linear equations: Abbreviated as: AX=B, Solve the equation using the least squares method: X=(A T A) -1 A T B (9) The solution is a, b, c; Step 2-2-3, calculate the center and radius:

6. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 2, the weight error detection process includes the following steps: Steps 2-4, for the acquired point cloud dataset S filtered ={q i ∈S||q i -μ||≤3σ}, determine the number of point cloud slices K and the slice thickness according to the reference height H of the shrimp cake The height interval of the kth slice is in: Where k = 1, 2, 3, ..., K; Step 2-5, extract the point cloud data set of the kth slice Projection extraction: Orthogonally project the 3D point cloud after outlier filtering onto the XY plane to generate a two-dimensional point set. Use the least squares method to fit a circle and obtain the radius r. k , Consider each slice as a thin cylindrical disk with a volume of: The total cumulative volume is: The shrimp cake density adaptive model compensates for density changes based on the temperature of the shrimp cake raw materials: ρ=ρ0(1+αΔT) (15) Where ρ0 is the base density of shrimp cakes, which is calculated based on the raw materials, the temperature coefficient α = -0.001 / °C, ΔT = 25°C-T, The weight calculation formula is: m test =ρ·V (16) The qualification judgment formula is: |m test -m set |≤1g (17) In the formula, m test is the calculated weight, m set It is the set weight.

7. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 4, the time t when the shrimp cakes reach the vacuum suction cup a for: In t a Always start the vacuum system to remove shrimp cakes. The diameter of the vacuum suction cup should cover more than 80% of the diameter of the shrimp cake. a is the predicted time when the shrimp cakes reach the rejection point, X a is the X-axis coordinate of the rejection mechanism in the conveyor belt coordinate system, X 虾 is the original X-axis coordinate of the shrimp cake in the conveyor belt coordinate system, v is the speed of the conveyor belt along the X-axis, t d System processing latency is the time it takes to compute images and point clouds.

8. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 4, the specific process of rejecting unqualified products is as follows: During the continuous movement of the conveyor belt, a vacuum suction cup mechanism is used to achieve vacuum removal during the conveying process of shrimp cakes. The main factors include horizontal force and vertical force. The vertical force includes overcoming gravity and accelerating inertia, which can be expressed as: Where v1 is the vertical speed of the shrimp cake during vacuum removal, which is preset to 0.1-0.2m / s, Δt v is the vertical acceleration time of the shrimp culling process, generally 0.1s, g is the acceleration due to gravity, The horizontal force is: Where Δt h Generally 0.1s, The total suction force, including safety factor, is: Where S f Use 1.5-2.0, The vacuum parameter is expressed as vacuum degree P total : Where A is the area of ​​the suction cup in the vacuum rejection system, which is generally between 0.8 and 1.0 times the area of ​​the shrimp cake, and μ is the safety factor, which is 0.

6. The air flow rate is: Where, ρ a is the air density, λ is the safety factor of 1.5, ρ is the density of shrimp cakes, is the friction coefficient between the shrimp cake and the conveyor belt.

9. The method for detecting the quality of shrimp cakes based on image recognition according to claim 1, wherein: in, In step 4, at t a Within the time, the vacuum suction cup rejection mechanism should be positioned just above the predicted position of the center of the shrimp cake to be rejected, and ensure that the vacuum suction cup rejection mechanism is 3-5mm away from the surface of the shrimp cake.

10. A shrimp cake molding quality detection system based on image recognition, characterized in that: include: The trigger acquisition module uses a laser grating sensor to sense the round shrimp cakes on the conveyor belt in real time. When the shrimp cake enters the field of view, it triggers the structured light 3D camera to synchronously capture the RGB image and depth data of the shrimp cake, generating 3D point cloud data. A parallel processing module extracts the appearance shape parameters of the shrimp cake based on the three-dimensional point cloud data, calculates the diameter error, and performs shape error detection. At the same time, it calculates the volume based on the point cloud data slices, calculates the weight error in combination with the density adaptive model, and performs weight error detection; Comprehensive judgment module: if the diameter error is ≤1mm and the weight error is ≤1.0g, it is judged as qualified; otherwise, it is judged as unqualified and the unqualified shrimp cake is recorded; The rejection module calculates the time t when the shrimp cakes reach the vacuum suction cup based on the conveyor belt speed, system delay time and rejection mechanism coordinates. a , at t a Always activate the vacuum rejection system to remove unqualified shrimp cakes.

Images