Experimental animal model feed detection and separation equipment based on Yolo v5 and color features

Abstract

The present invention discloses a feed detection and separation device for an experimental animal model based on Yolov5 and color features. The device comprises a shock-absorbing seat, a pillar is installed on the top of the shock-absorbing seat, a roller is installed on the top of the pillar, a belt is sleeved on the outside of the roller, and an unqualified material collection port is placed at one end of the belt. The present invention introduces the Yolov5 model to extract the color features of feed particles for identification, and uses filling feed to pre-treat the residual material in a pelletizer. Before producing the target feed, the filling feed is first used to remove the residual feed, which can achieve a better cleaning effect. The color of the filling feed is relatively easy to identify and can be easily separated from the target feed. Using Yolov5 to detect the feed particle area can eliminate the interference of invalid information in the background, and can ensure that accurate color information is extracted and accurately identified. In addition, using a quadratic curve to fit the purity change curve can predict the feed purity in advance, making the operation more convenient and efficient.

Description

Technical Field

[0001] The present invention relates to the field of feed production technology, and in particular to a feed detection and separation device for experimental animal models based on Yolo v5 and color features. Background Art

[0002] When pelleting equipment for producing experimental animal model feed produces pellets, some residual raw materials will be difficult to clean inside the machine. Therefore, the first batch of feed produced by the machine will contain some impurities from the tailings of the previous batch of feed. To ensure the purity of the model feed, the first and tailings of the machine need to be manually sorted out to separate the contaminated feed. The existing technology requires workers to visually identify whether the experimental animal model feed is cross-contaminated and separate it. The deep learning neural network used by Yolo v5 can effectively learn and extract high-level semantic information from images, and can accurately extract the feed pellet targets in the image.

[0003] However, manual separation takes a long time and there are differences in personal selection. Yolo v5 is not efficient in processing the underlying color. Different samples need to be re-collected and trained. In addition, Yolo v5 requires the feed particles to be formed before the image is obtained, which increases time and cost. Summary of the Invention

[0004] The present invention provides a feed detection and separation device for experimental animal models based on Yolo v5 and color features, which can effectively solve the problems raised in the above background technology, such as long manual separation time, personal selection differences, inefficient Yolo v5 processing of underlying colors, the need for re-collection and training of different samples, and the need to obtain images after the feed particles are formed using Yolo v5, which increases time and cost.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an experimental animal model feed detection and separation device based on Yolo v5 and color features, comprising a shock-absorbing seat, a pillar installed on the top of the shock-absorbing seat, a roller installed on the top of the pillar, a belt sleeved on the outside of the roller, an unqualified material collection port placed at one end of the belt, and a qualified material collection port placed at the other end of the belt, the roller at one end is connected to a motor, the motor signal input end is connected to a motor control module, the top of the belt is aligned with the discharge port of a pellet mill, a camera is installed at the discharge port of the pellet mill, and the camera is connected to an image acquisition and processing module via a data cable;

[0006] The image acquisition and processing module is controlled by a background server, which also includes a detection system;

[0007] The detection system includes a detection module, an open database, and a Yolo v5 module;

[0008] The steps for testing and separating the experimental animal model feed detection and separation equipment are as follows:

[0009] S1. Use the Yolo v5 module to train the Yolo v5 model. The detection module detects the feed particle area in the image, extracts the normalized color feature vector of the feed area, and calculates the correlation with the standard vector.

[0010] S2. Clear the remaining material in the feed pellet mill;

[0011] S3, clearing out the feed for the experimental animal model;

[0012] S4, predict feed purity;

[0013] The specific steps of S1 are as follows:

[0014] S11. Select low-cost raw materials with marked colors and recyclable materials as feed for experimental animal models;

[0015] S12. Use a camera to collect images of different feed pellet samples, and use images from an open database to classify the images into feed pellets and non-feed pellets, label the images as feed pellets and non-feed pellets, and divide all images into a training set, a test set, and a validation set;

[0016] S13. Use the Yolo v5 network to train the training set images. The Yolo v5 network consists of an input part, a backbone part, a neck part, and a prediction part. In the backbone, a 1x1 convolution kernel is added before the first 3x3 convolution layer, and an attention mechanism module is added before the third CSP to achieve the goal of accurate recognition.

[0017] S14, achieve the required classification accuracy in the test set;

[0018] S15. Divide the experimental animal model feed into qualified target feed and qualified filler feed, collect images of different feeds, and use Yolo v5 to detect the area [x1, y1, x2, y2] of the feed particles in the image, where x1 and y1 are the coordinates of the upper left corner and x2 and y2 are the coordinates of the lower right corner;

[0019] S16. Perform HSV transformation on the extracted feed particle image area, process the H channel and S channel separately to extract 32-dimensional color histograms, and after experimental comparison, extract the data of the S channel to form a 32-dimensional feature vector, and further obtain the normalized color feature vectors VS1 (qualified filler feed) and VS2 (qualified target feed).

[0020] According to the above technical solution, the specific steps of S2 are as follows:

[0021] S21. First, remove the remaining tailings that are easy to remove. At this time, there is still tailings in the gaps inside the pellet mill. Add the filling feed to the pellet mill feed port;

[0022] S22. Install a camera at the feed outlet of the experimental animal model, and transmit the camera acquisition results to the detection system;

[0023] S23, using the detection module to detect the sample image and obtain the target area [x1, y1, x2, y2];

[0024] S24, extracting color features from the target area to obtain the sample's feature vector V1, comparing it with the standard sample's feature vector VS1, calculating the correlation coefficient according to formula (1), setting the threshold to 0.9, and considering a value greater than 0.9 as qualified, otherwise the filling feed is unqualified;

[0025]

[0026] S25. The equipment continues to operate until the detection module gives a signal that the filled feed is qualified. At this time, it can be considered that the remaining raw materials have been removed.

[0027] According to the above technical solution, the specific steps of S3 are as follows:

[0028] S31, add the target feed into the pellet mill inlet, and the pellet mill operates normally;

[0029] S32, transmitting the acquisition results of the discharge port camera (1) to the detection system;

[0030] S33, using the Yolo v5 model in the detection module to detect the sample image, and extracting features of the detected target area [x3, y3, x4, y4];

[0031] S34, comparing the extracted characteristic vector of the sample with the characteristic vector of the designated target feed standard sample, calculating the correlation coefficient according to formula (1), setting the threshold to 0.95, and considering the sample above 0.95 as qualified, otherwise giving a signal that the target feed is unqualified;

[0032] S35, if it is unqualified, the motor is controlled to reverse, and the conveyor belt of the discharge port runs in the unqualified area;

[0033] S36, continue to operate the machine until the detection system gives a signal that the target feed is qualified, at which point it can be considered that all impurities have been removed;

[0034] S37. By controlling the motor to rotate forward after a delay of Δt, the conveyor belt at the discharge port is changed to a qualified zone, where Δt is an empirical time, to ensure that there is no residual tailings on the conveyor belt.

[0035] According to the above technical solution, the specific steps of S4 are as follows:

[0036] S41. If the feed purity is unqualified, continue to run the machine and record the correlation coefficient every certain period of time (e.g. 10s) to obtain (t i , r i ), record the r values at different times (t1, r1), (t2, r2), ... (t n , r n );

[0037] S42. Use r=a*t 2 +b*t+c to fit the correlation (purity) change curve;

[0038] S43, calculating the values of a, b, and c using the least squares method;

[0039]

[0040] For Ax=b, use the generalized inverse to calculate the values of a, b, and c;

[0041] Ax=b (3)

[0042] x=(A T A) -1 A T b (4)

[0043] S44, prediction t (n+1) Moment purity, if the purity is qualified, the system will prompt that qualified finished products will appear after 10 seconds;

[0044] S45, prepare for the event, reduce the motor speed, and replace the aggregate opening;

[0045] S46, until the detection system gives a signal that the target experimental animal model feed is qualified, then it can be considered that all the filler tailings have been separated;

[0046] S47. The conveyor belt at the discharge port is changed to a qualified zone by controlling the motor to rotate forward after a delay of Δt, where Δt is an empirical time, to ensure that there is no residual tailings on the conveyor belt.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The Yolo v5 model was introduced to extract the color features of feed pellets for identification, and filler feed was used to pre-process the remaining feed in the pellet mill. Before producing the target feed, filler feed was used to remove the remaining feed, which can achieve a better removal effect. The color of filler feed is relatively easy to identify and can be easily separated from the target feed. Using Yolo v5 to detect the feed particle area can eliminate the interference of invalid information in the background, and extract color features based on this, which can ensure the extraction of accurate color information and accurate identification;

[0049] 2. When a new feed sample is used, there's no need to retrain the model. Simply mixing the raw materials evenly and extracting the color characteristics of the feed once can distinguish between different types of feed. Formed feed pellets are not required. Using Yolo v5 to distinguish different feeds requires using formed feed pellets, which makes acquiring training samples more time-consuming and costly. Using a quadratic curve to fit the purity change curve allows for early prediction of feed purity, making operations more convenient and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0051] In the attached figure:

[0052] Figure 1 It is a diagram of the detection and separation equipment of the present invention;

[0053] Figure 2 This is a physical picture of the feed pellets of the present invention;

[0054] Figure 3 This is a comparison diagram of the feature vectors of samples of the same type according to the present invention;

[0055] Figure 4 This is a comparison diagram of the characteristic vectors of different types of samples of the present invention;

[0056] Figure 5 It is the correlation detection and fitting curve diagram of the present invention;

[0057] Figure 6 It is a flow chart of the detection and separation scheme of the present invention;

[0058] Figure 7 It is the Yolo v5 model structure diagram of the present invention;

[0059] Figure 8 It is a schematic diagram of the detection and separation principle of the present invention;

[0060] Numbers in the figure: 1. Camera; 2. Image acquisition and processing module; 3. Motor control module; 4. Motor; 5. Roller; 6. Belt; 7. Unqualified material collection port; 8. Qualified material collection port; 9. Shock absorber; 10. Pillar. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] Example: Figure 1-2 、 Figure 6-8 As shown, the present invention provides a technical solution, an experimental animal model feed detection and separation device based on Yolo v5 and color characteristics, including a shock-absorbing seat 9, a pillar 10 is installed on the top of the shock-absorbing seat 9, a roller 5 is installed on the top of the pillar 10, a belt 6 is sleeved on the outside of the roller 5, an unqualified material collection port 7 is placed at one end of the belt 6, and a qualified material collection port 8 is placed at the other end of the belt 6, the roller 5 at one end is connected to a motor 4, the signal input end of the motor 4 is connected to a motor control module 3, the top of the belt 6 is aligned with the discharge port of the pelletizer, a camera 1 is installed at the discharge port of the pelletizer, and the camera 1 is connected to an image acquisition and processing module 2 via a data cable;

[0063] The image acquisition and processing module 2 is controlled by a background server, which also includes a detection system;

[0064] The detection system includes a detection module, an open database, and a Yolo v5 module;

[0065] The steps for testing and separating the experimental animal model feed detection and separation equipment are as follows:

[0066] The first step is to train the Yolo v5 model using the Yolo v5 module. The detection module detects the feed particle area in the image, extracts the normalized color feature vector of the feed area, and calculates the correlation with the standard vector.

[0067] 1. Select low-cost raw materials with marked colors and recyclable materials as feed for experimental animal models;

[0068] 2. Use a camera to collect images of different feed pellet samples and use images from an open database to classify the images into feed pellet images and non-feed pellet images;

[0069] 3. Use labelImg software to label the images as feed particles and non-feed particles, output the images in .xml format, and divide all images into training set, test set, and validation set;

[0070] 4. The images in the training set are standardized and preprocessed, and then subjected to Mosaic data enhancement, adaptive anchor box calculation, and adaptive image scaling before being input into Backbone.

[0071] 5. Backbone includes the Focus structure and the CSP structure. A 1*1 convolution kernel is added before the first 3*3 convolution layer, and a visual attention model is added before the third CSP. After processing the data, feature maps of various scales are obtained and then input into the Neck part.

[0072] 6. Neck includes FPN and PAN structures, which upsamples the data and then performs downsampling feature fusion to obtain multi-scale tensor data;

[0073] 7. In the Prediction part, CIOU_Loss is used as the loss function of the Bounding Box in the training phase. In the inference phase, the weighted NMS mechanism is used to find the optimal target bounding box. The gradient is calculated based on the loss function and back propagation. The gradient is updated in real time and verified using the validation set to obtain the Yolo v5 detection model.

[0074] 8. Achieve the required classification accuracy in the test set;

[0075] 9. Divide the experimental animal model feed into qualified feed and qualified filling feed, collect images of different feeds, and use Yolo v5 to detect the area [x1, y1, x2, y2] of the feed particles in the image, where x1 and y1 are the coordinates of the upper left corner and x2 and y2 are the coordinates of the lower right corner;

[0076] 10. Perform HSV transformation on the extracted feed particle image area, process the H channel and S channel separately to extract 32-dimensional color histograms. After experimental comparison, extract the data of the S channel to form a 32-dimensional feature vector, and further obtain the normalized color feature vectors VS1 (qualified filling feed) and VS2 (qualified target feed).

[0077] Step 2: Clear the remaining material in the feed pellet mill;

[0078] 1. First, remove the remaining tailings that are easy to remove. At this time, there are still tailings in the gaps inside the pellet mill. Add the filling feed to the pellet mill feed port;

[0079] 2. Install a camera 1 at the feed outlet of the experimental animal model, and transmit the collected data of the camera 1 to the detection system;

[0080] 3. Use the detection module to detect the sample image, obtain the target area [x1, y1, x2, y2], and extract the feature vector V1. V1 is a 32-dimensional feature vector, such as Figure 3The following is a comparison of the feature vectors of samples of the same type. Figure 4 Shown is a comparison of feature vectors of different types of samples;

[0081] 4. Extract color features from the target area to obtain the sample’s feature vector V1, compare it with the standard sample’s feature vector VS1, and calculate the correlation coefficient according to formula (1). For example, the correlation coefficient of the test results of the same sample is 0.97, and the correlation coefficient of different samples is 0.71. The threshold is set to 0.9. If it is greater than 0.9, it is considered qualified, otherwise the filling feed is unqualified.

[0082]

[0083] 5. The equipment continues to operate until the detection module gives a signal that the filling feed is qualified. At this time, it can be considered that the remaining raw materials have been removed.

[0084] Step 3: Remove the feed that fills the experimental animal model;

[0085] 1. Add the target feed into the pellet mill inlet and the pellet mill operates normally;

[0086] 2. Transmit the acquisition results of the discharge port camera 1 to the detection system;

[0087] 3. Use the Yolo v5 model in the detection module to detect the sample image and extract the feature vector V2 of the detected target area [x3, y3, x4, y4]. V2 is a 32-dimensional feature vector.

[0088] 4. Compare the extracted sample feature vector with the feature vector of the specified target feed standard sample, calculate the correlation coefficient R2 between the sample feature vector V2 and the standard target feed feature vector VS2 according to formula (1), and set a threshold of 0.95 for comparison. If it is greater than 0.95, it is considered qualified, otherwise a signal of unqualified target feed is given;

[0089] 5. If it is unqualified, the motor 4 is controlled to reverse so that the conveyor belt 6 at the discharge port runs in the unqualified area.

[0090] Step 4: Predict feed purity;

[0091] 1. If the feed purity is unqualified, continue to run the machine and record the correlation coefficient every 10 seconds to obtain (t i , r i );

[0092] 2. Record the r values at different times (t1, r1), (t2, r2), ... (t n , r n );

[0093] 3. Use r = a * t 2 +b*t+c to fit the correlation (purity) change curve, the fitting curve is as follows Figure 5 As shown;

[0094] 4. Calculate the values of a, b, and c using the least squares method;

[0095]

[0096] 5. For Ax = b, use the generalized inverse to calculate the values of a, b, and c;

[0097] Ax=b (3)

[0098] x=(A T A) -1 A T b (4)

[0099] 6. Predict t (n+1) Moment purity, if the purity is qualified, the system will prompt that qualified finished products will appear after 10 seconds;

[0100] 7. Be prepared, reduce the speed of motor 4, and replace the aggregate port;

[0101] 8. Until the detection system gives a signal that the target experimental animal model feed is qualified, it can be considered that all the filling machine tailings have been separated;

[0102] 9. By controlling the motor 4 to rotate forward after a delay of Δt, the conveyor belt 6 at the discharge port is changed to the qualified zone, where Δt is the empirical time to ensure that there is no residual tailings on the conveyor belt;

[0103] The delay time is to ensure that the belt stops before turning, and secondly to transport the remaining tailings on the belt to the unqualified area.

[0104] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. Experimental animal model feed detection and separation equipment based on Yolo v5 and color features, characterized by: It comprises a shock-absorbing seat (9), a pillar (10) is installed on the top of the shock-absorbing seat (9), a roller (5) is installed on the top of the pillar (10), a belt (6) is sleeved on the outside of the roller (5), an unqualified material collection port (7) is placed at one end of the belt (6), and a qualified material collection port (8) is placed at the other end of the belt (6), the roller (5) at one end is connected to a motor (4), a signal input end of the motor (4) is connected to a motor control module (3), the top of the belt (6) is aligned with the discharge port of the granulator, a camera (1) is installed at the discharge port of the granulator, and the camera (1) is connected to an image acquisition and processing module (2) via a data line; The image acquisition and processing module (2) is controlled by a background server, which also includes a detection system; The detection system includes a detection module, an open database, and a Yolo v5 module; The steps for testing and separating the experimental animal model feed detection and separation equipment are as follows: S1. Use the Yolo v5 module to train the Yolo v5 model. The detection module detects the feed particle area in the image, extracts the normalized color feature vector of the feed area, and calculates the correlation with the standard vector. S2. Clear the remaining material in the feed pellet mill; S3, clearing out the feed for the experimental animal model; S4, predict feed purity; The specific steps of S1 are as follows: S11. Select low-cost raw materials with marked colors and recyclable materials as feed for experimental animal models; S12. Use a camera to collect images of different feed pellet samples, and use images from an open database to classify the images into feed pellets and non-feed pellets, label the images as feed pellets and non-feed pellets, and divide all images into a training set, a test set, and a validation set; S13. Use the Yolo v5 network to train the training set images. The Yolo v5 network consists of an input part, a backbone part, a neck part, and a prediction part. In the backbone, a 1x1 convolution kernel is added before the first 3x3 convolution layer, and an attention mechanism module is added before the third CSP to achieve the goal of accurate recognition. S14, achieve the required classification accuracy in the test set; S15. Divide the experimental animal model feed into qualified target feed and qualified filler feed, collect images of different feeds, and use Yolo v5 to detect the area [x1, y1, x2, y2] of the feed particles in the image, where x1 and y1 are the coordinates of the upper left corner and x2 and y2 are the coordinates of the lower right corner; S16. Perform HSV transformation on the extracted feed particle image area, process the H channel and S channel separately to extract 32-dimensional color histograms, and after experimental comparison, extract the data of the S channel to form a 32-dimensional feature vector, and further obtain the normalized color feature vectors VS1. VS1 is a qualified filling feed, and VS2, VS2 is a qualified target feed.

2. The experimental animal model feed detection and separation device based on Yolo v5 and color features according to claim 1, characterized in that: The specific steps of S2 are as follows: S21. First, remove the remaining tailings that are easy to remove. At this time, there is still tailings in the gaps inside the pellet mill. Add the filling feed to the pellet mill feed port; S22, installing a camera (1) at the feed outlet of the experimental animal model, and transmitting the collected data of the camera (1) to the detection system; S23, using the detection module to detect the sample image and obtain the target area [x1, y1, x2, y2]; S24, extract the color features of the target area, obtain the sample's feature vector V1, compare it with the standard sample's feature vector VS1, calculate the correlation coefficient according to formula (1), set the threshold to 0.9, and consider it qualified if it is greater than 0.9, otherwise the filling feed is unqualified; （1） S25. The equipment continues to operate until the detection module gives a signal that the filled feed is qualified. At this time, it can be considered that the remaining raw materials have been removed.

3. The experimental animal model feed detection and separation device based on Yolo v5 and color features according to claim 2, characterized in that, The specific steps of S3 are as follows: S31, add the target feed into the pellet mill inlet, and the pellet mill operates normally; S32, transmitting the acquisition results of the discharge port camera (1) to the detection system; S33, using the Yolo v5 model in the detection module to detect the sample image, and extracting features of the detected target area [x3, y3, x4, y4]; S34, comparing the extracted characteristic vector of the sample with the characteristic vector of the designated target feed standard sample, calculating the correlation coefficient according to formula (1), setting the threshold to 0.95, and considering the sample above 0.95 as qualified, otherwise giving a signal that the target feed is unqualified; S35, if it is unqualified, the motor (4) is controlled to reverse, and the conveyor belt (6) at the discharge port runs in the unqualified area; S36, continue to operate the machine until the detection system gives a signal that the target feed is qualified, at which point it can be considered that all impurities have been removed; S37, by controlling the motor (4) in the delay After t time, the conveyor belt (6) at the discharge port is rotated forward and the running direction is changed to the qualified area, where t is the empirical time to ensure that there is no remaining tailings on the conveyor belt.

4. The experimental animal model feed detection and separation device based on Yolo v5 and color features according to claim 3, characterized in that: The specific steps of S4 are as follows: S41. If the feed purity is unqualified, continue to run the machine and record the correlation coefficient every once in a while to obtain (t i , r i ), record the r values at different times (t1, r1), (t2, r2), ... (t n , r n ); S42. Use r=a*t2+b*t+c to fit the correlation (purity) change curve; S43, calculating the values of a, b, and c using the least squares method; （2） For Ax=b, use the generalized inverse to calculate the values of a, b, and c; （3） （4） S44, prediction t (n+1) Moment purity: if the purity is qualified, the system will prompt that a qualified finished product will appear after 10 seconds; S45, prepare, reduce the speed of motor (4), and replace the aggregate opening; S46, until the detection system gives a signal that the target experimental animal model feed is qualified, then it can be considered that all the filler tailings have been separated; S47, by controlling the motor (4) in the delay After t time, the conveyor belt (6) at the discharge port is turned forward and the running direction is changed to the qualified area, where t is the empirical time to ensure that there is no remaining tailings on the conveyor belt.

Images