A pig live backfat thickness intelligent evaluation method

Abstract

The present application relates to a kind of pig live backfat thickness intelligent evaluation method, comprising the following steps: collecting the multi-angle contour image of test pig;Based on simplified multi-task cascade convolutional neural network SMTCNN model, the image of each target site of test pig is quickly detected and positioned;Based on multi-objective optimization deep learning model MOCNN, the image of target site of test pig is estimated;The index value of target site of test pig is further input into random forest model, and the live backfat thickness estimation value of test pig is obtained.The present application can realize the automatic estimation of pig live backfat thickness and other indicators, reduce manual intervention, save manpower, realize the intelligent analysis and rapid determination of pig phenotype, and provide effective data support for genetic evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of in vivo measurement, in particular to a pig in vivo back fat thickness intelligent evaluation method. BACKGROUND

[0002] Boar performance measurement is an important link of breeding work, and is also the cornerstone of all breeding work.

[0003] Most of the existing performance measurement techniques are based on sensory evaluation and traditional instrument equipment, such as when the boar is 100-115 kg and the growth and fattening is completed, the electronic cage scale is used to weigh and fix the pig, and then the back fat instrument is used to measure the in vivo back fat and eye muscle area. This method has low digitalization and intelligence, and has the disadvantages of low objectivity and precision, low measurement throughput, and high dependence on manual operation.

[0004] Therefore, there is an urgent need for a pig in vivo back fat thickness intelligent evaluation method. SUMMARY

[0005] The present application provides a pig in vivo back fat thickness intelligent evaluation method, which can realize high-throughput and intelligent acquisition of main performance indicators of boars such as in vivo back fat.

[0006] To achieve the above purpose, the present application adopts the following technical scheme:

[0007] The present application provides a pig in vivo back fat thickness intelligent evaluation method, which includes the following steps:

[0008] Collecting multi-angle contour images of the test pig, including the front, side, back and top of the test pig;

[0009] Based on a simplified multi-task cascade convolutional neural network model, the multi-angle contour images of the test pig are quickly detected and positioned to obtain the image of the target part of the test pig;

[0010] Based on a multi-objective optimization deep learning model, the image of the target part of the test pig is estimated to obtain the body size index value of the target part of the test pig;

[0011] The body size index value of the target part of the test pig is further input into a random forest model to obtain the in vivo back fat thickness estimation value of the test pig.

[0012] The pig in vivo back fat thickness intelligent evaluation method, preferably, the step of "based on a simplified multi-task cascade convolutional neural network SMTCNN model to quickly detect and position the image of each target part of the test pig" specifically includes the following steps:

[0013] The P-Net is used to extract a preliminary candidate region, a candidate window is calibrated, and non-maximum suppression is used to sort the candidate frame, and finally a preliminary screened target part candidate frame of the test pig is obtained.

[0014] The R-Net is used for fine screening of the preliminary screened target part candidate frame of the test pig, and most of the non-target part regions are removed, and finally the image of the target part of the test pig is obtained.

[0015] The pig live backfat thickness intelligent evaluation method, preferably, the implementation method of the step of collecting multi-dimensional image information of the test pig and the basic information data of the test pig is to use an intelligent breeding pig electronic cage scale, the intelligent breeding pig electronic cage scale comprises a cage body, a camera device, an electronic display screen, an intelligent tablet and a weighing weighbridge, the four side walls and the top of the cage body are provided with the camera device, the bottom wall of the cage body is provided with the weighing weighbridge, and the cage body is provided with the electronic display screen and the intelligent tablet.

[0016] The pig live backfat thickness intelligent evaluation method, preferably, the camera device comprises a folding support and an industrial-grade camera, and the industrial-grade camera is arranged on the side wall and the top of the cage body through the folding support.

[0017] The pig live backfat thickness intelligent evaluation method, preferably, the bottom of the cage body is further provided with universal rollers.

[0018] The second aspect of the present application provides a pig live backfat thickness intelligent evaluation system, comprising:

[0019] A first processing unit collects multi-angle contour images of a test pig;

[0020] A second processing unit performs rapid detection and positioning on the multi-angle contour images of the test pig based on a simplified multi-task cascaded convolutional neural network model to obtain an image of a target part of the test pig;

[0021] A third processing unit performs body size index estimation on the image of the target part of the test pig based on a multi-target optimization deep learning model to obtain a body size index value of the target part of the test pig;

[0022] A fourth processing unit further inputs the body size index value of the target part of the test pig into a random forest model to obtain a live backfat thickness estimation value of the test pig.

[0023] The third aspect of the present application provides a computer storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the pig live backfat thickness intelligent evaluation method.

[0024] The fourth aspect of the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for intelligently evaluating the backfat thickness of a boar in vivo when executing the computer program.

[0025] The present application has the following advantages due to the above technical solutions:

[0026] (1) The present application realizes the collection of image information of the measured pig by setting a new electronic scale with an industrial-grade zoom camera on the basis of a common boar scale, and further realizes the high-throughput and accurate determination of key performance indicators such as the backfat thickness of the pig in vivo by using artificial intelligence technology.

[0027] (2) The folding support ensures the normal functioning of the operator and the saving of physical space during transportation.

[0028] (3) The present application realizes the image extraction of the target part of the pig based on the pig image based on a simplified multi-task cascaded convolutional neural network (SMTCNN), further estimates the key body size indicator values by using the above-mentioned image based on a multi-objective optimization deep learning model (MOCNN), and finally inputs the key body size indicator values into the random forest model to accurately estimate the backfat thickness of the test pig in vivo, and embeds an APP on the intelligent tablet of the boar determination scale to realize the manual input of the boar breeding phenotype data and the automatic display of the estimation result sheet.

[0029] (4) The simplified multi-task cascaded convolutional network SMTCNN involved in the present application only uses P-net and R-net, and optimizes the non-maximum suppression NMS method, so that the target number in one picture can also be accurately detected. Compared with the traditional multi-task cascaded convolutional network MTCNN, the problem of poor detection effect when the number of targets is too large can be solved.

[0030] In addition, the present application designs a multi-objective function and the corresponding network structure for the estimation of the backfat thickness related characteristic indicators in vivo, carefully balances all tasks during the learning process, and can place a priori matrix on the fully connected layer in addition to the shared structure and specific task layer, so that the model can learn the relationship between tasks. Compared with the traditional deep learning method which only optimizes a single objective function, the present application has great advantages in complex practical production problems.

[0031] The present application can realize the fine and automatic determination of the backfat thickness and other indicators of pigs in vivo, reduce manual intervention, save manpower, realize the rapid determination and intelligent analysis of pig data, and provide effective data support for pig breeding value estimation. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1The front view of the intelligent pig breeding electronic cage scale provided by the application;

[0033] Figure 2 The left view of the intelligent pig breeding electronic cage scale provided by the application;

[0034] Figure 3 The top view of the intelligent pig breeding electronic cage scale provided by the application;

[0035] Figure 4 The state diagram of the folding support when it is unfolded;

[0036] Figure 5 The state diagram of the folding support when it is folded;

[0037] In the figure:

[0038] 1, cage body; 2, industrial-grade camera; 3, folding support; 4, electronic display screen;

[0039] 5, intelligent tablet; 6, weighing weight; 7, universal roller. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0041] The pig live body back fat thickness intelligent evaluation method provided by the application comprises the following steps: collecting multi-dimensional image information of a test pig; rapidly detecting and positioning images of each target part of the test pig based on a simplified multi-task cascade convolutional neural network SMTCNN model; estimating the body size index of the target part of the test pig based on a multi-objective optimization deep learning model MOCNN; further inputting the index value of the target part of the test pig into a random forest model to obtain the live body back fat thickness estimation value of the test pig. The present application can realize fine and automatic determination of pig live body back fat thickness and other indexes, reduce manual intervention, save manpower, realize rapid determination and intelligent analysis of pig data, and provide effective data support for pig breeding value estimation.

[0042] Next, a pig live body back fat thickness intelligent evaluation method provided by an embodiment of the present application will be described in detail in conjunction with the drawings.

[0043] Embodiment 1

[0044] A pig live body back fat thickness intelligent evaluation method comprises the following steps:

[0045] S1, collecting multi-angle test pig contour images of the test pigs, including front view, side view, back view and top view of the test pigs.

[0046] As shown in Figures 1 to 5 The implementation method of step S1 is to use an intelligent breeding pig electronic cage scale, which includes a cage 1, a camera device, an electronic display screen 4, a smart tablet 5 and a weighing weighbridge 6. The camera device is arranged on the four side walls and directly above the cage 1, which is used for individual identification of pigs and collection of all-around image information. The cage 1 is provided with front and back doors to facilitate the entry and exit of test pigs. The weighing weighbridge 6 is arranged on the bottom wall of the cage 1. The electronic display screen 4 and the smart tablet 5 are arranged on the side wall of the cage 1. The electronic display screen 4 is convenient for real-time reading of the accurate weight of the test pigs. The smart tablet 5 has an APP built-in, which can upload the on-site measurement data to the management system in real time. The cage 1 is also provided with universal wheels 7 at the bottom to facilitate the movement of the cage 1.

[0047] The camera device includes a folding support 3 and an industrial-grade camera 2. The industrial-grade camera 2 is arranged on the side wall and directly above the cage 1 through the folding support 3. The folding support 3 is specially arranged for the information collection industrial-grade camera 2. When the equipment is working, the folding support 3 can be unfolded to ensure the normal function of the industrial-grade camera 2. When the equipment is idle, the folding support 3 can be folded to save physical space and facilitate storage and transportation.

[0048] S2, rapid detection and positioning of the multi-angle test pig contour images based on a simplified multi-task cascade convolutional neural network SMTCNN model to obtain images of target parts of the test pigs.

[0049] The step S2 specifically includes the following steps:

[0050] S21, using a P-Net (proposal Network) in the multi-task cascade convolutional neural network to extract a preliminary candidate region, calibrating the candidate window, and using non-maximum suppression to sort the candidate frame to finally obtain a test pig target part candidate frame after preliminary screening.

[0051] The candidate window refers to a fixed size or multi-scale rectangular frame defined in the sliding window search process in the image. The preliminary candidate region refers to a region set that may contain the target generated by the algorithm through the candidate window. The candidate frame refers to a rectangular frame that further circumscribes the candidate target in the image.

[0052] S22. The R-Net (refine network) in the multi-task cascade convolutional neural network is used to finely screen the candidate frames of the target parts of the test pigs after the preliminary screening, remove most of the non-target areas, and finally realize the image extraction of the target parts of the test pigs.

[0053] S3. The deep learning model MOCNN based on multi-objective optimization estimates the body size index of the image of the target part of the test pig to obtain the body size index value of the target part of the test pig.

[0054] S4. The index value of the target part of the test pig is further input into the random forest model to obtain the estimated value of the backfat thickness of the test pig in vivo.

[0055] Before performing an intelligent assessment of pig backfat thickness, the multi-task cascade convolutional neural network model (SMTCNN) in step S2, the deep learning model (MOCNN) in step S3, and the iterative optimization of the random forest model in step S4 are iteratively optimized. The iterative optimization specifically includes the following steps:

[0056] S31. Divide the data set into a training set, a validation set, and a test set, wherein the data set includes multi-angle contour images of the test pigs obtained through the "smart breeding pig electronic cage", body size index values ​​of target parts of the test pigs, and predicted back fat thickness estimation values ​​of the test pigs in vivo.

[0057] S32. After data preprocessing of the training set, the multi-dimensional image data of the test pigs are input into the multi-task cascade convolutional neural network model for training and solidification, and the images of the target parts of the test pigs are obtained, which are input into the deep learning model for training and solidification.

[0058] S33. After data preprocessing of the test set, the multidimensional image data of the test pigs were input into the solidified multi-task cascade convolutional neural network model, the target parts were detected and the target part data were input into the solidified MOCNN model, the values ​​of each feature index were estimated, the error analysis of the results was performed, and the SMTCNN and MOCNN models were iteratively optimized.

[0059] S34. Input the characteristic index value obtained in step S33 into the solidified random forest model to estimate the backfat value, and perform error analysis on the result by establishing a regression relationship between the estimated value and the measured value.

[0060] Example 2

[0061] The embodiment 1 provides the pig live backfat thickness intelligent evaluation method, and correspondingly, the embodiment provides a pig live backfat thickness intelligent evaluation system. The pig live backfat thickness intelligent evaluation system provided by the embodiment can implement the pig live backfat thickness intelligent evaluation method of the embodiment 1. The pig live backfat thickness intelligent evaluation system can be realized by software, hardware or a combination of software and hardware. For example, the pig live backfat thickness intelligent evaluation system can include integrated or separated functional modules or functional units to execute corresponding steps in the methods of the embodiment 1. Since the pig live backfat thickness intelligent evaluation system of the embodiment is basically similar to the method embodiment, the process of the embodiment is relatively simple, and the related parts can be referred to the part of the embodiment 1. The pig live backfat thickness intelligent evaluation system of the embodiment is only illustrative.

[0062] The pig live backfat thickness intelligent evaluation system provided by the embodiment includes:

[0063] The first processing unit collects multi-angle contour images of the subject pig;

[0064] The second processing unit performs rapid detection and positioning on the multi-angle contour images of the subject pig based on a simplified multi-task cascade convolutional neural network model to obtain images of target parts of the subject pig;

[0065] The third processing unit performs body size index estimation on the images of the target parts of the subject pig based on a multi-target optimization deep learning model to obtain body size index values of the target parts of the subject pig;

[0066] The fourth processing unit further inputs the body size index values of the target parts of the subject pig into a random forest model to obtain live backfat thickness estimation values of the subject pig.

[0067] Embodiment 3

[0068] The embodiment provides a processing device for implementing the pig live backfat thickness intelligent evaluation method provided by the embodiment 1. The processing device can be a processing device for a client, such as a mobile phone, a notebook computer, a tablet computer, a desktop computer, etc., to execute the pig live backfat thickness intelligent evaluation method of the embodiment 1.

[0069] The processing device includes a processor, a memory, a communication interface and a bus. The processor, the memory and the communication interface are connected through the bus to complete the communication among each other. The memory stores a computer program that can run on the processor. When the processor runs the computer program, the pig live backfat thickness intelligent evaluation method provided by the embodiment 1 is executed.

[0070] Preferably, the memory can be a high-speed random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory.

[0071] Preferably, the processor can be a central processing unit (CPU), a digital signal processor (DSP), or various types of general-purpose processors, without limitation.

[0072] Embodiment 4

[0073] The pig live backfat thickness intelligent evaluation method of the present embodiment 1 can be specifically implemented as a computer program product, which can include a computer readable storage medium having computer readable program instructions loaded thereon for executing the pig live backfat thickness intelligent evaluation method described in the present embodiment 1.

[0074] The computer readable storage medium can be a tangible device that maintains and stores instructions for use by an instruction execution device. The computer readable storage medium can be, for example but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0075] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent method for evaluating backfat thickness of live pigs, characterized in that: The steps include: Collect the outline images of the test pigs from multiple angles; The multi-angle test pig contour images are quickly detected and located based on a simplified multi-task cascade convolutional neural network model, specifically comprising the following steps: The candidate network in the multi-task cascade convolutional neural network model was used to extract preliminary candidate regions, the candidate windows were calibrated, and the candidate frames were sorted using non-maximum suppression. Finally, the candidate frames of the target parts of the test pigs were obtained after preliminary screening. Using the fine-tuning network in the multi-task cascade convolutional neural network model to finely screen the candidate frames of the target part of the test pig after the preliminary screening, and finally obtaining an image of the target part of the test pig; A deep learning model based on multi-objective optimization estimates body size indices of target parts of the test pigs, thereby obtaining body size index values ​​of the target parts of the test pigs. The body size index estimation includes designing multiple objective functions and corresponding network structures, carefully balancing all tasks during the learning process, and placing a priori matrices on fully connected layers in addition to shared structures and task-specific layers, so that the model can learn relationships between tasks. The body size index values ​​of the target parts of the test pigs are further input into the random forest model to obtain the estimated backfat thickness values ​​of the test pigs in vivo.

2. The intelligent evaluation method for backfat thickness of living pigs according to claim 1, characterized in that: The multi-angle outline images of the test pig include outline images of the front, side, back and top views of the test pig.

3. The intelligent evaluation method for backfat thickness of living pigs according to claim 1, characterized in that: The preliminary candidate area refers to a set of areas that may contain targets generated by the algorithm through the candidate window. The candidate window refers to a fixed-size or multi-scale rectangular box defined during the sliding window search process in the image. The candidate box refers to a rectangular box that further encircles the candidate target in the image.

4. The intelligent evaluation method for backfat thickness of living pigs according to claim 1 or 2, characterized in that: The method for collecting multi-angle contour images of the test pigs is to use an intelligent breeding pig electronic cage scale for measurement. The intelligent breeding pig electronic cage scale includes a cage body, a camera device, an electronic display screen, an intelligent tablet and a weighing scale. The camera device is provided on the four side walls and the top of the cage body, the weighing scale is provided on the bottom wall of the cage body, and the cage body is provided with the electronic display screen and the intelligent tablet.

5. The intelligent evaluation method for backfat thickness of living pigs according to claim 4, characterized in that: The camera device comprises a folding bracket and an industrial-grade camera. The industrial-grade camera is arranged on the side wall and directly above the cage body through the folding bracket.

6. The intelligent evaluation method for backfat thickness of living pigs according to claim 4, characterized in that: The bottom of the cage is also provided with a universal roller.

7. An intelligent assessment system for pig backfat thickness, characterized in that: include: The first processing unit collects the test pig's outline images from multiple angles; The second processing unit performs rapid detection and positioning of the multi-angle test pig contour images based on a simplified multi-task cascade convolutional neural network model, specifically comprising the following steps: The candidate network in the multi-task cascade convolutional neural network model was used to extract preliminary candidate regions, the candidate windows were calibrated, and the candidate frames were sorted using non-maximum suppression. Finally, the candidate frames of the target parts of the test pigs were obtained after preliminary screening. Using the fine-tuning network in the multi-task cascade convolutional neural network model to finely screen the candidate frames of the target part of the test pig after the preliminary screening, and finally obtaining an image of the target part of the test pig; A third processing unit estimates body size indices of the target part images of the test pigs based on a multi-objective optimized deep learning model to obtain body size index values ​​of the target parts of the test pigs. The body size index estimation includes designing multiple objective functions and corresponding network structures, carefully balancing all tasks during the learning process, and placing a priori matrices on fully connected layers in addition to shared structures and task-specific layers to enable the model to learn relationships between tasks. The fourth processing unit further inputs the body size index value of the target part of the test pig into the random forest model to obtain the estimated value of the backfat thickness of the test pig in vivo.

8. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent method for evaluating the backfat thickness of living breeding pigs according to any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the intelligent evaluation method for backfat thickness of living breeding pigs according to any one of claims 1 to 6 are implemented.

Images