Food detection sampling device

Abstract

This invention relates to the field of food testing, specifically disclosing a food testing sampling device, including a shovel body. The upper surface of the shovel body has multiple fixed openings, each containing a liftable support platform. The upper surface of the support platform has an arc-shaped groove. All platforms are connected by a bracket, with a traction rope and a spring at the bottom of the bracket. A weighing sensor is embedded in the bottom of the arc-shaped groove. This invention lowers the platform to a first depth position by pulling the traction rope, forming a deep groove to accommodate multiple samples. After releasing the traction rope, the platform rises to a second depth position under the action of the spring. The limited volume of the arc-shaped groove pushes individual samples upwards and automatically removes excess samples. Simultaneously, the weighing sensor collects the weight of each sample as the platform rises. Combined with a camera on a limit stop bar, this acquires sample images, achieving integrated automatic screening, weighing, and image acquisition of individual samples during the shoveling process, significantly improving the sampling and testing efficiency of granular foods.

Description

Technical Field

[0001] This invention belongs to the field of food testing, specifically a food testing sampling device. Background Technology

[0002] Food testing is a crucial step in ensuring food safety. For granular foods (such as grains, beans, nuts, and coffee beans), it is usually necessary to collect representative samples from each batch and perform weight measurement, appearance inspection, and component analysis on individual particles. Current sampling methods often involve randomly scooping a certain amount of sample from the material to be tested using a shovel or sampling spoon, then transferring the sample to a container or testing table where testing personnel manually sort out individual particles for weighing and observation, or using auxiliary tools such as sieves for sieving.

[0003] However, this traditional sampling and pretreatment method has significant efficiency bottlenecks: the samples obtained by shoveling often contain multiple or even dozens of particles, and testing personnel must spend a lot of time separating individual particles from the mixed samples. Furthermore, manual sorting is prone to particle confusion, omissions, or surface contamination, affecting the accuracy and consistency of subsequent test results. Especially in scenarios requiring high-throughput testing (such as testing dozens to hundreds of samples at a time), the workload of manual sorting is extremely heavy, severely restricting testing efficiency. Therefore, there is an urgent need for a device that can automatically complete the screening and positioning of multiple samples into single samples during the sampling process, reducing manual intervention and improving the automation level of food testing. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a food testing and sampling device to solve the problems of low testing efficiency and easy contamination caused by the need for manual sorting of individual particles after shoveling samples in the prior art.

[0005] This invention provides a food testing and sampling device, comprising:

[0006] The shovel body has a handle connected to its rear end, and the upper surface of the shovel body is the sample receiving surface.

[0007] Multiple holes are formed on the upper surface of the shovel body;

[0008] A depth adjustment mechanism, disposed within the shovel body, includes: a liftable support platform, the number of which corresponds to the number of holes, each support platform having an arc-shaped groove on its upper surface for supporting a single sample; a bracket for fixedly connecting all the support platforms; a traction rope fixedly connected to the bottom of the bracket, the traction rope extending to the outside of the shovel body and close to the handle; and a spring fixed to the bottom of the bracket.

[0009] The weighing sensor is fixed on the arc-shaped groove and moves up and down together with the support platform;

[0010] The depth adjustment mechanism is configured to have a first depth position and a second depth position:

[0011] The first depth position is: when the traction rope is pulled, the traction rope overcomes the elastic force of the spring and drives the bracket and all the support platforms to move down synchronously, so that the support platforms are located at a lower position in the hole, and the bottom of the arc-shaped groove is lower than the upper surface of the shovel body, forming a deep groove that can accommodate multiple samples;

[0012] The second depth position is: when the traction rope is released, under the action of the spring's restoring force, the bracket and all the support platforms rise synchronously, the bottom of the arc-shaped groove is higher than the upper surface of the shovel body, so that during the rising process only one sample embedded in the arc-shaped groove is lifted up, and the excess samples automatically slide off, and finally only one sample is retained in each hole and the sample protrudes from the upper surface of the shovel body.

[0013] Preferably, the holes are arranged in an array along the length and width directions on the upper surface of the shovel body.

[0014] Preferably, the opening shape of the hole is circular, square, or regular hexagonal.

[0015] Preferably, the radius of curvature of the arc-shaped groove matches the average radius of the sample to be tested, and its cross-section in the depth direction is circular or elliptical.

[0016] Preferably, at the first depth position, the bottom of the arc-shaped groove is 3-5 mm lower than the upper surface of the shovel body; at the second depth position, the bottom of the arc-shaped groove is 0.5-2.0 mm higher than the upper surface of the shovel body.

[0017] Preferably, a limit stop is provided on the outer side of the shovel body near the handle, and a camera is provided on the limit stop, with the lens of the camera facing the hole area on the upper surface of the shovel body.

[0018] Preferably, the limiting stop bar also integrates a fill light, which is a ring-shaped LED light group arranged around the lens of the camera.

[0019] The invention also provides a food testing and sampling system, comprising:

[0020] Such as the food testing and sampling device mentioned above;

[0021] The image acquisition module includes a camera and its supplementary light mounted on the shovel body;

[0022] The control module is electrically connected to the stroke detection component of the depth adjustment mechanism, the weighing sensor, and the image acquisition module, respectively.

[0023] The control module is configured as follows:

[0024] When the depth adjustment mechanism is detected to be at the second depth position, the weighing sensor is triggered to collect the weight data of the sample in each hole. And trigger the camera to acquire image data of the upper surface of the shovel;

[0025] The sampling results are output by performing a fusion analysis based on the weight data and the image data.

[0026] The fusion analysis includes the following steps:

[0027] Step S1: Preprocess the image data to segment the region of interest corresponding to each hole;

[0028] Step S2: Perform instance segmentation algorithm on each region of interest to identify the individual samples exposed after being pushed up, and calculate the image feature parameters of each sample. The image feature parameters include at least the projected area. equivalent diameter Roundness Color feature vector and texture feature vector ;

[0029] Step S3: Record the weight data for each sample. Align with the corresponding image feature parameters;

[0030] Step S4: Calculate the weight-projected area ratio for each sample. Calculate the mean of all samples. and standard deviation ,like If so, it is marked as a weight-area mismatch anomaly;

[0031] Step S5: Convert the feature vector of each sample Input a pre-defined classifier model, and output the class label and confidence score of the sample;

[0032] Step S6: Based on the results of steps S4 and S5, determine whether each sample is qualified or abnormal, and generate a sampling result report.

[0033] Preferably, the classifier model is any one of support vector machine, random forest, gradient boosting tree or multilayer perceptron neural network, and is pre-trained using a training sample set labeled with weight, area, roundness, color, texture and category labels.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] By setting up a liftable support platform and a traction rope and spring linked to it, the holes on the shovel can form a deep groove at the first depth position to accommodate multiple samples. At the second depth position, the limited volume of the arc-shaped groove is used to push a single sample upward and automatically remove excess samples, thus achieving accurate screening and positioning of single samples. No manual sorting or additional screening devices are required, which significantly simplifies the sampling operation process.

[0036] Meanwhile, the bottom of the platform integrates a weighing sensor that rises and falls with it. After the sample is pushed up and stably exposed on the upper surface of the shovel, the weight data is collected immediately, avoiding the errors and time-consuming process caused by transferring the sample and weighing it again after traditional sampling.

[0037] With the camera mounted on the limit stop bar, the system can clearly capture the complete surface image of each sample from a fixed perspective. Combined with the fusion analysis algorithm built into the control module, the weight data of each hole is matched and comprehensively judged with the image features of the corresponding sample, such as area, roundness, color, and texture. Abnormal samples are automatically identified and the overall pass rate is calculated, realizing the integration and automation of sampling, single-particle separation, weighing and visual inspection.

[0038] This invention is particularly suitable for rapid on-site testing of granular foods. While ensuring sampling consistency and testing accuracy, it significantly reduces the intensity of manual operation and the requirements for operational skills, providing an efficient and reliable technical means for food quality and safety supervision. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0040] Figure 2 This is a schematic diagram of the depth adjustment mechanism of the present invention;

[0041] Figure 3 This is a schematic diagram illustrating the switching between the first depth position and the second depth position in Embodiment 1 of the present invention;

[0042] Figure 4 This is a schematic diagram of the installation structure of the limiting stop bar in Embodiment 2 of the present invention.

[0043] In the diagram: 1. Shovel body; 2. Hole; 3. Depth adjustment mechanism; 31. Support platform; 32. Arc-shaped groove; 33. Bracket; 34. Traction rope; 35. Spring; 4. Handle; 5. Limit stop bar; 6. Camera. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1: Refer to Figure 1 This embodiment provides a food testing and sampling device, including a shovel body 1, multiple holes 2, a depth adjustment mechanism 3, and a weighing sensor.

[0046] The shovel body 1 is cylindrical in shape, with a handle 4 connected to its rear end. The upper surface of the shovel body 1 is the sample receiving surface, used to support the food sample to be tested during sampling.

[0047] Multiple holes 2 are formed on the upper surface of the shovel body 1. In this embodiment, the multiple holes 2 are arranged in an array along the length and width directions of the shovel body 1. The spacing between adjacent holes 2 is not less than twice the diameter of the hole opening, so as to avoid mutual interference of samples. The opening shape of the holes 2 is preferably circular, but can also be square or regular hexagonal.

[0048] The depth adjustment mechanism 3 is located inside the shovel body 1; refer to Figure 2 The depth adjustment mechanism 3 includes a liftable support platform 31. The support platform 31 is cylindrical or prismatic in shape, and its outer diameter slides in fit with the inner diameter of the hole 2. An arc-shaped groove 32 is provided on the upper surface of the support platform 31. The arc-shaped groove 32 is used to support a single sample. The radius of curvature of the arc-shaped groove 32 matches the average radius of the sample to be tested. Its cross-section in the depth direction is circular or elliptical, so that the sample can be stably embedded in the groove and is not easy to roll off.

[0049] A drive assembly is connected below the support platform 31. In this embodiment, the drive assembly includes a bracket 33 that is fixedly connected to several support platforms 31. A traction rope 34 is fixedly connected to the bottom of the bracket 33. The traction rope 34 extends to the outside of the shovel body 1 and is close to the handle 4. A spring 35 is also fixed to the bottom of the bracket 33.

[0050] The weighing sensor is fixed on the arc-shaped groove 32, specifically embedded inside the support 31, with its upper surface flush with the bottom surface of the arc-shaped groove 32 so as to directly contact the sample. Since the weighing sensor rises and falls together with the support 31, the sensor can accurately measure the weight of the sample regardless of the height of the support 31.

[0051] like Figure 3 As shown, the depth adjustment mechanism 3 is configured to have a first depth position and a second depth position:

[0052] When the user pulls the traction rope 34, the traction rope 34 overcomes the elastic force of the spring 35 and drives the bracket 33 to move downward. At the same time, the bracket 33 compresses the spring 35. The downward movement of the bracket 33 causes all the support platforms 31 to move downward synchronously, so that the support platforms 31 are located at a lower position in the hole 2. At this time, the effective depth of the hole 2 is relatively large, forming the first depth position. At the first depth position, the bottom of the arc-shaped groove 32 is lower than the upper surface of the shovel body 1 by a certain distance (e.g., 3-5 mm). The hole 2 can accommodate multiple samples or even allow the sample to sink completely into the hole. In this state, the user uses the shovel body 1 to scoop up the food sample to be tested. The sample particles fall smoothly into each hole 2 under the action of gravity. Some samples may be piled up in the hole.

[0053] After sampling is completed and the shovel 1 is removed from the food to be tested, the user slowly releases the traction rope 34. Under the restoring force of the spring 35, the support 33 gradually moves upward, causing the platform 31 to rise synchronously. As the platform 31 rises, the sample located in the hole 2 is gradually pushed upward. Since the upper surface of the platform 31 is only provided with an arc-shaped groove 32, the opening area of ​​the arc-shaped groove 32 is slightly larger than the projected area of ​​a single sample, and the groove depth is only 0.3 to 0.5 times the average particle size of the sample, when the platform 31 rises, only the bottommost sample that is just embedded in the arc-shaped groove 32 can be stably supported by the groove; the rest are stacked on top of it. Samples located above or at the edge of the groove, lacking sufficient lateral support, may slip off the edge of the arc-shaped groove 32 or roll out of the hole 2 during the rise of the support platform. When the spring 35 is fully reset, the support platform 31 rises to the second depth position. At the second depth position, the bottom of the arc-shaped groove 32 is 0.5 to 2.0 mm higher than the upper surface of the shovel body 1. The lower half of the single sample is embedded in the arc-shaped groove 32, and the upper half is completely protruding from the upper surface of the shovel body 1. The sample has completely escaped the constraint of the side wall of the hole 2 and its surface is fully exposed. At this time, only one sample is retained in each hole 2, realizing the automatic screening and positioning of single samples.

[0054] Example 2: This example is a further improvement on Example 1. The similarities will not be repeated here; the differences are as follows:

[0055] Reference Figure 4 A limiting stop bar 5 is also provided on the outer side of the shovel body 1 near the handle 4. The limiting stop bar 5 is a rigid rod, with its first end installed on the side wall of the shovel body 1 (near the handle 4) through a fixing seat, and its second end extending outward to the outer side of the shovel body 1; the limiting stop bar 5 has sufficient height to ensure that the multiple samples pushed up can be successfully photographed.

[0056] A camera 6 is provided on the limiting stop bar 5, and the lens of the camera 6 is facing the hole 2 area on the upper surface of the shovel body 1. Specifically, the camera 6 is fixed to the bottom outer side of the second end of the limiting stop bar 5, and its shooting direction can cover the entire hole array. The resolution of the camera 6 is not less than 5 million pixels, and it is equipped with autofocus and macro shooting functions, which can clearly capture the surface texture, color and contour details of the pushed sample when the support platform 31 is in the second depth position.

[0057] To ensure image quality, the limit lever 5 also integrates a fill light, which is a ring-shaped LED light group arranged around the lens of the camera 6 to provide uniform and shadowless illumination. The brightness of the fill light can be automatically adjusted according to the ambient light or preset by the control module.

[0058] Example 3: This example provides a food testing and sampling system, which includes the food testing and sampling device as described in Example 1 or Example 2, and further integrates an image acquisition module, a control module and a data processing module.

[0059] The system in this embodiment includes: a sampling device body, namely the shovel 1, hole 2, depth adjustment mechanism 3, weighing sensor, and handle 4 as described in Embodiment 1 or Embodiment 2, as well as the limit stop bar 5 and camera 6 (image acquisition module) as described in Embodiment 2.

[0060] The control module (not shown, built into the handle 4 or an external portable controller) is electrically connected to the traction rope drive end of the depth adjustment mechanism 3, the weighing sensor, the camera 6, and the supplementary light, respectively. In actual implementation, the traction rope 34 can be manually operated in conjunction with the position sensor for status detection; or it can be further upgraded to electric drive, that is, a micro motor is used to replace manual rope pulling, and the control module sends a command to drive the motor to rotate forward and backward to realize the lifting and lowering of the support platform 31. This embodiment is illustrated by manually pulling the rope in conjunction with the position sensor.

[0061] The position detection module includes limit sensors (such as photoelectric switches or contact switches) set at the upper and lower ends of the travel of the support 33, which are used to detect whether the platform 31 has reached the first depth position or the second depth position and feed the signal back to the control module.

[0062] Data acquisition and triggering logic:

[0063] When the user pulls the traction rope 34, the bracket 33 moves down to the first depth position, the first limit sensor (lower end) is triggered, and the control module records the current state as "sampling state"; after the user completes the shovel sampling, the traction rope 34 is slowly released, and the bracket 33 rises under the action of the spring 35. When it rises to the second depth position, the second limit sensor (upper end) is triggered, the control module determines that the sampling is completed, and prepares to collect data.

[0064] After receiving the second depth position trigger signal, the control module delays for 0.3 seconds (to wait for mechanical vibration to decay) and then performs the following operations:

[0065] Trigger weighing data acquisition: Read the output value of the weighing sensor at the bottom of each hole 2, and obtain the weight data after analog-to-digital conversion. For each weighing sensor, data is collected 10 times consecutively, and the median value or the steady-state weight is obtained after Kalman filtering. .

[0066] Trigger image acquisition: Turn on the supplementary light to control camera 6 to capture a global image of the upper surface of the shovel body 1 and store it as digital image data. During filming, the limiting stop 5 ensures that the relative position of the camera 6 and the shovel 1 is fixed, so the area coordinates of the hole 2 in the image are pre-calibrated.

[0067] Fusion analysis algorithm:

[0068] The control module has built-in image processing and data analysis programs, which perform the following steps:

[0069] Step S1: Image Preprocessing

[0070] For images Distortion correction, white balance correction, and histogram equalization are performed to eliminate the effects of lens distortion and uneven lighting. Using pre-calibrated hole location templates, the region of interest (ROI) corresponding to each hole is segmented from the image. .

[0071] Step S2: Instance segmentation and feature extraction

[0072] For each ROI Lightweight instance segmentation algorithms (such as YOLACT-Mobile) are used to identify individual samples exposed after being pushed up; since the samples have detached from the sidewalls of the holes and protrude from the upper surface of the shovel, their complete outlines can be clearly segmented; the following image features are calculated for the segmented sample regions:

[0073] Projected area (Pixel area converted to actual area, mm²);

[0074] equivalent diameter ;

[0075] Roundness ,in The perimeter of the outline;

[0076] Color feature vector Extract the mean, variance, and histogram peaks in the HSV or Lab color space;

[0077] Texture feature vector Contrast, correlation, energy, and homogeneity are extracted using Local Binary Mode (LBP) or Gray-Level Co-occurrence Matrix (GLCM).

[0078] Step S3: Weighing data alignment

[0079] According to the hole number The image features extracted in step S2 are compared with the weight obtained in step 1. One-to-one correspondence.

[0080] Step S4: Weight-Area Consistency Analysis

[0081] Calculate the weight-projected area ratio for each sample. For all holes Calculate the mean and standard deviation If a sample satisfies If so, it is marked as "weight-area mismatch anomaly" (possibly due to internal voids, foreign objects, or abnormal water content in the sample).

[0082] Step S5: Multimodal feature fusion classification

[0083] The feature vectors of each sample are concatenated as follows: .Will Input a pre-trained multilayer perceptron (MLP) classifier and output the category label of the sample (such as "normal soybean", "moldy soybean", "broken soybean", "foreign object" etc.) and the corresponding confidence score.

[0084] The classifier was obtained through offline training: a large amount of single-particle sample weighing and image data were collected, the categories were manually labeled, and a 3-layer fully connected network was trained using the cross-entropy loss function (the input layer dimension depends on the total length of the features, the hidden layer has 128 neurons, and the output layer has 4 categories).

[0085] Step S6: Comprehensive Judgment and Report Generation

[0086] For each sample:

[0087] If there are no anomalies in step S4 and the confidence score output by the classifier is ≥0.85, it is judged as "qualified" and the category label and weight are recorded.

[0088] If step S4 is abnormal, or the classifier confidence score is <0.85, or the output category is "moldy / broken / foreign object", then it is determined to be an "abnormal product" and the abnormality type is recorded.

[0089] Count all holes:

[0090] Total number of samples (equal to the number of holes, unless a hole has no sample, in which case the void needs to be detected by image detection);

[0091] Pass rate = (Number of qualified samples / Total number of samples) × 100%

[0092] Weight distribution: mean, standard deviation, maximum value, minimum value;

[0093] Abnormal item classification and counting.

[0094] Generate JSON format or graphical sampling result reports, and display key results on the display screen on handle 4, or send them to a mobile APP or Laboratory Information Management System (LIMS) via a wireless communication module (such as Bluetooth / Wi-Fi).

[0095] Feedback adjustment function (optional expansion):

[0096] If the system detects a "porosity" exceeding 20% ​​for three consecutive times (i.e., the proportion of pores in the sample that were not detected by image segmentation is too high), the control module will remind the user via an indicator light or buzzer to check if the sample particles are too small or if the scooping method is improper, and suggest increasing the scooping depth or repeating the sampling.

[0097] If the system detects the sample's "roundness" multiple times "If the value is below the threshold (indicating that the sample is deformed by the pressure of the hole sidewall), the control module can prompt the user to reduce the spring return speed (if an electric drive is used) or suggest using a pore size plate that matches the sample."

[0098] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Any changes, modifications, substitutions and variations made by those skilled in the art to the above embodiments within the scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A food testing and sampling device, characterized in that, include: The shovel body (1) has a handle (4) connected to its rear end, and the upper surface of the shovel body (1) is the sample receiving surface. Multiple holes (2) are formed on the upper surface of the shovel body (1); The depth adjustment mechanism (3) is located inside the shovel body (1) and includes: a liftable support platform (31), the number of which is the same as the number of holes (2) and they are arranged in a one-to-one correspondence. The upper surface of each support platform (31) is provided with an arc-shaped groove (32) for supporting a single sample; a bracket (33) for fixing all the support platforms (31); a traction rope (34) for fixing to the bottom of the bracket (33), the traction rope (34) extending to the outside of the shovel body (1) and close to the handle (4); and a spring (35) fixed to the bottom of the bracket (33). The weighing sensor is fixed on the arc-shaped groove (32) and moves up and down together with the support platform (31); The depth adjustment mechanism (3) is configured to have a first depth position and a second depth position: The first depth position is: when the traction rope (34) is pulled, the traction rope (34) overcomes the elastic force of the spring (35) and drives the bracket (33) and all the support platforms (31) to move down synchronously, so that the support platform (31) is located in the lower position in the hole (2), and the bottom of the arc-shaped groove (32) is lower than the upper surface of the shovel body (1), forming a deep groove that can accommodate multiple samples; The second depth position is: when the traction rope (34) is released, under the action of the restoring force of the spring (35), the bracket (33) and all the support platforms (31) rise synchronously, the bottom of the arc groove (32) is higher than the upper surface of the shovel body (1), so that only one sample embedded in the arc groove (32) is lifted during the rising process, and the excess samples automatically slide off, and only one sample is ultimately retained in each hole (2) and the sample protrudes from the upper surface of the shovel body (1).

2. The food testing and sampling device according to claim 1, characterized in that, The holes (2) are arranged in an array along the length and width directions on the upper surface of the shovel body (1).

3. The food testing and sampling device according to claim 1, characterized in that, The opening shape of the hole (2) is circular, square or regular hexagonal.

4. The food testing and sampling device according to claim 1, characterized in that, The radius of curvature of the arc-shaped groove (32) matches the average radius of the sample to be tested, and its cross-section in the depth direction is circular or elliptical.

5. The food testing and sampling device according to claim 1, characterized in that, At the first depth position, the bottom of the arc-shaped groove (32) is 3-5 mm lower than the upper surface of the shovel body (1); at the second depth position, the bottom of the arc-shaped groove (32) is 0.5-2.0 mm higher than the upper surface of the shovel body (1).

6. The food testing and sampling device according to claim 1, characterized in that, A limit stop bar (5) is also provided on the outer side of the shovel body (1) near the handle (4). A camera (6) is provided on the limit stop bar (5), and the lens of the camera (6) faces the hole (2) area on the upper surface of the shovel body (1).

7. The food testing and sampling device according to claim 6, characterized in that, The limiting stop bar (5) is also integrated with a fill light, which is a ring-shaped LED light group arranged around the lens of the camera (6).

8. A food testing and sampling system, characterized in that, include: The food testing and sampling device as described in any one of claims 1 to 7; The image acquisition module includes a camera (6) and its supplementary light installed on the shovel body (1); The control module is electrically connected to the stroke detection component of the depth adjustment mechanism (3), the weighing sensor and the image acquisition module, respectively; The control module is configured as follows: When the depth adjustment mechanism (3) is detected to be in the second depth position, the weighing sensor is triggered to collect the weight data of the sample in each hole (2). And trigger the camera (6) to collect image data of the upper surface of the shovel (1); The sampling results are output by performing a fusion analysis based on the weight data and the image data. The fusion analysis includes the following steps: Step S1: Preprocess the image data to segment the region of interest corresponding to each hole (2); Step S2: Perform instance segmentation algorithm on each region of interest to identify the individual samples exposed after being pushed up, and calculate the image feature parameters of each sample. The image feature parameters include at least the projected area. equivalent diameter Roundness Color feature vector and texture feature vector ; Step S3: Record the weight data for each sample. Align with the corresponding image feature parameters; Step S4: Calculate the weight-projected area ratio for each sample. Calculate the mean of all samples. and standard deviation ,like ( If the threshold coefficient is set to 0, it is marked as a weight-area mismatch anomaly. Step S5: Convert the feature vector of each sample Input a pre-defined classifier model, and output the class label and confidence score of the sample; Step S6: Based on the results of steps S4 and S5, determine whether each sample is qualified or abnormal, and generate a sampling result report.

9. The food detection and sampling system according to claim 8, characterized in that, The classifier model is any one of support vector machine, random forest, gradient boosting tree or multilayer perceptron neural network, and is pre-trained using a training sample set labeled with weight, area, roundness, color, texture and category labels.

Images