CN114170292A - Pig weight estimation method and device based on depth image - Google Patents
Pig weight estimation method and device based on depth image Download PDFInfo
- Publication number
- CN114170292A CN114170292A CN202111258590.4A CN202111258590A CN114170292A CN 114170292 A CN114170292 A CN 114170292A CN 202111258590 A CN202111258590 A CN 202111258590A CN 114170292 A CN114170292 A CN 114170292A
- Authority
- CN
- China
- Prior art keywords
- pig
- point
- depth image
- weight
- weight estimation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
- G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
- G01B11/0608—Height gauges
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a pig weight estimation method and device based on a depth image, and relates to the technical field of livestock informatization. According to the method, the depth image is shot for the pig, the Keypoint-RCNN network is adopted to extract the key points of the pig, the body size characteristic data of the pig is calculated according to the key point information, and then the weight data of the pig is obtained.
Description
Technical Field
The invention relates to the technical field of livestock information, in particular to a pig weight estimation method and device based on a depth image.
Background
The monitoring of the change of the weight indexes of the pigs can reflect the development speed, the body state, the feed-meat ratio and the diet condition of the pigs, the evaluation of the indexes is beneficial to the quality identification, the marketing and the breeding and the accurate feeding of the pigs, meanwhile, the meat quantity and the corresponding quality after slaughtering can be estimated, and whether the weight of the pigs reaches the hurdle mark or not can be judged quickly. The weight of the pigs is quickly detected, so that the weight of the pigs is of great significance.
Currently, the weight of a pig is estimated by a manual body size measurement method (pig weight (kg) ═ chest circumference (cm) × body length (cm)/coefficient), or a group weight is estimated by driving the pig to a weighbridge in a specific area. None of these methods achieves accurate measurement of individual pigs and is inefficient.
Publication No.: CN110426112A, published as 2019-11-08, discloses a live pig weight measurement method and device, and the method is used for extracting key points and measuring volumes based on rgb images of videos, although the stability of recognition is enhanced by using a neural network. But this scheme is based on ordinary rgb image measurement pig's buttock width and buttock height, and the accuracy is lower, and this scheme is compared and is extracted the key point slowly through the depth map.
Disclosure of Invention
The invention aims to overcome the technical problems and provides the pig body weight estimation method based on the depth image, which has high speed of extracting key points and accurate body size characteristic data.
The technical scheme of the invention is as follows:
a pig weight estimation method based on a depth image comprises the following steps:
s1, obtaining the depth image of the pig taken down and the breed of the pig;
s2, extracting key points of the pigs in the depth image by using a Keypoint-RCNN network;
s3, calculating the body size characteristic data and the overlooking projection area of the pig according to the key points of the pig;
s4, inputting the pig variety, the body size characteristic data and the overlook projection area into an weight estimation model;
s5, obtaining the weight of the pig;
wherein, the weight estimation model in step S4 is established based on a regression model.
According to the technical scheme, weight estimation is carried out on the basis of the depth image, the height and width data of the pigs can be obtained at one time, the extraction speed of the key points is high, the positions of the key points are accurate, so that the method is high in robustness, and the finally obtained weight data is small in error.
Further, the method for obtaining the depth image in step S1 is as follows: firstly, a pig is subjected to downward shooting through a depth camera to obtain point cloud data, and then a depth image is generated according to the point cloud data;
when the pig is shot in a depression mode, the depth camera detects the real-time posture of the depth camera through the gyroscope, the depth camera meets the preset shooting visual angle and automatically shoots images, and otherwise, a prompt for adjusting the posture of the depth camera is sent.
Further, in step S2, before extracting the key points of the pig in the depth image using the Keypoint-RCNN network, the depth image needs to be segmented to obtain the pig contour in the image, and the segmentation step includes:
s21, carrying out noise reduction processing on the depth image according to a point cloud density filtering noise reduction algorithm to remove noise;
s22, calculating the height difference between the ground and the contour of the back of the pig by a sorting method, identifying the ground part by using the height information and deleting the ground part;
and S23, after the height information converted from the depth information is sequenced, extracting the maximum connected region to obtain the image including the contour of the pig.
According to the technical scheme, the contour of the pig in the image is segmented before the key point is extracted, so that whether the pig in the image completely finishes picture screening is judged, and the fact that the depth image subjected to key point segmentation comprises the complete pig is guaranteed.
Further, the key points of the pig in step S2 include: ear root point, anterior elbow point, posterior spine point, and caudal root point; wherein the ear root point, the anterior elbow point, the posterior elbow point and the tail root point are respectively arranged on the left and the right;
step S3, the body size feature data includes: shoulder width, hip width, body height, body length;
the shoulder width is obtained by calculating the distance between two front elbow points;
the hip width is obtained by calculating the distance between two back elbow points;
the height of the body is calculated by the distance of the difference between the height of the back point of the ridge and the height of the ground;
the body length is obtained by the connecting line distance of the middle point of the connecting line of the two ear root points and the middle point of the connecting line of the two tail root points.
Further, in step S2, the Keypoint-RCNN network includes: the system comprises a backbone network unit, a region suggestion unit, a region of interest matching unit, a key point detection branch unit, a mask detection branch unit and a frame regression and classification unit;
the method comprises the steps that a Keypoint-RCNN network obtains a depth image of a pig, and a backbone network unit extracts image features from the depth image to obtain a feature map of the depth image; the region suggesting unit extracts the region-of-interest feature maps with different sizes from the feature map; the interesting region matching unit converts the interesting region feature maps with different sizes into an interesting region feature map with a fixed size; the key point monitoring branch unit obtains an ear root point, an anterior elbow point, a posterior spine point and a tail root point on a region-of-interest characteristic diagram with a fixed size; the mask monitoring branch unit divides the characteristic diagram of the region of interest with fixed size to obtain a division diagram of the contour line of the pig; and the frame regression and classification unit identifies the pig from the region-of-interest feature map with the fixed size and provides other objects.
Further, the method for calculating the overhead projection area in step S3 is: the head and the tail of the pig are divided through key points, the head and the tail of the pig contour line segmentation image obtained by the mask branching unit are removed, and then the area of the rest part in the pig contour line segmentation image is calculated to be used as the overlook projection area.
In the technical scheme, the purpose of removing the head and the tail of the pig image is to avoid the change of the projection area of the pig caused by the change of the postures of the head and the tail of the pig in the actual application scene, so that the condition of weight estimation precision is influenced.
Further, the method for establishing the weight estimation model comprises the following steps:
the method comprises the steps of collecting weight data, body size characteristic data, overlooking projection area and pig varieties of pigs to establish a sample library of the pigs, corresponding the characteristics to actually measured weight data to establish a regression model of the pigs, and determining regression coefficients of the regression model of the pigs by using a least square method to obtain an estimated weight model.
Further, the regression model includes: a linear regression model, a polynomial regression model, a ridge regression model, a lasso regression model; the regression model corresponding to the different breeds of pigs is determined based on the weight estimation precision of sample data of the different breeds of pigs under each model.
A pig weight estimation device comprising: the device comprises a depth camera, a key point extraction module, a body size calculation module and an weight estimation module;
the method comprises the steps of inputting the variety of a pig to be detected, using a depth camera to perform downward shooting on the pig to obtain a depth image, extracting key points of the pig in the depth image by a key point extraction module, calculating body ruler feature data and an overlooking projection area of the pig by the body ruler calculation module by using the key points, and obtaining the weight of the pig by a weight estimation module according to the variety of the pig, the body ruler feature data and the overlooking projection area;
the key point extracting module adopts a Keypoint-RCNN network to extract key points, and the weight estimating module adopts a regression model to estimate weight.
This technical scheme has provided a pig weight estimation equipment, and equipment is equipped with the degree of depth camera for only shoot the depth map to the pig, then the key point draws the module and draws the key point to the depth map through the Keypoint-RCNN network, and this technical scheme's pig weight estimation equipment adopts the degree of depth camera, can once only obtain the height and body width data of pig, and the key point extraction rate is fast, and the key point position is accurate, the pig weight data of the low error that can be fast convenient acquisition.
Further, still include controller and gyroscope, the real-time gesture of gyroscope detection depth camera, the controller passes through the gyroscope and judges whether the shooting angle of depth camera accords with the preset range, if not accord with then the controller sends the correction suggestion, if accord with then the controller control depth camera shoots the image and shows
The technical scheme provides a pig weight estimation method and device based on a depth image, and compared with the prior art, the technical scheme has the beneficial effects that: the method comprises the steps of shooting a depth image of a pig, extracting key points of the pig by adopting a Keypoint-RCNN network, calculating body size characteristic data of the pig according to key point information, and further obtaining weight data of the pig.
Drawings
FIG. 1 is a schematic diagram of the steps of a pig weight estimation method based on depth images;
FIG. 2 is a schematic view of an operator using the apparatus to photograph pigs;
FIG. 3 is a diagram of a Keypoint-RCNN network structure;
FIG. 4 is a schematic view of key points on the back of a pig;
wherein, 1, ear root points; 2. the anterior elbow point; 3. posterior elbow point; 4. a tail root point; 5. posterior point of spine.
Detailed Description
For clearly illustrating the pig weight estimation method based on depth images of the present invention, the present invention will be further described with reference to the following examples and drawings, but the scope of the present invention should not be limited thereby.
Example 1
A pig weight estimation method based on depth images is shown in figure 1 and comprises the following steps:
s1, obtaining the depth image of the pig taken down and the breed of the pig;
s2, extracting key points of the pigs in the depth image by using a Keypoint-RCNN network;
s3, calculating the body size characteristic data and the overlooking projection area of the pig according to the key points of the pig;
s4, inputting the pig variety, the body size characteristic data and the overlook projection area into a weight estimation model to obtain the weight data of the pig;
wherein, the weight estimation model in step S4 is established based on a regression model.
The technical scheme carries out weight estimation based on the depth image, can obtain height and width data of the pigs at one time, is high in key point extraction speed and accurate in key point positions, and therefore the method is high in robustness and small in error of finally obtained weight data.
Example 2
A pig weight estimation method based on depth images is shown in figure 1 and comprises the following steps:
s1, obtaining the depth image of the pig taken down and the breed of the pig;
s2, extracting key points of the pigs in the depth image by using a Keypoint-RCNN network;
the key points of the pigs comprise: ear root point 1, anterior elbow point 2, posterior elbow point 3, posterior spine point 5, and caudal root point 4; wherein, the ear root point 1, the anterior elbow point 2, the posterior elbow point 3 and the tail root point 4 are respectively arranged on the left and the right;
step S3, the body size feature data includes: shoulder width, hip width, body height, body length;
the shoulder width is calculated by the distance between two anterior elbow points 2;
the hip width is calculated by the distance between two back elbow points 3;
the height is calculated by the distance of the difference between the height of the posterior ridge point 5 and the height of the ground;
the body length is obtained by the connecting line distance of the connecting line middle points of the two ear root points 1 and the connecting line middle points of the two tail root points 4.
The Keypoint-RCNN network includes: the system comprises a backbone network unit, a region suggestion unit, a region of interest matching unit, a key point detection branch unit, a mask detection branch unit and a frame regression and classification unit;
the method comprises the steps that a Keypoint-RCNN network obtains a depth image of a pig, and a backbone network unit extracts image features from the depth image to obtain a feature map of the depth image; the region suggesting unit extracts the region-of-interest feature maps with different sizes from the feature map; the interesting region matching unit converts the interesting region feature maps with different sizes into an interesting region feature map with a fixed size; the key point monitoring branch unit obtains an ear root point 1, an anterior elbow point 2, a posterior elbow point 3, a posterior spine point 5 and a tail root point 4 on a feature map of an interested area with a fixed size; the mask monitoring branch unit divides the characteristic diagram of the region of interest with fixed size to obtain a division diagram of the contour line of the pig; and the frame regression and classification unit identifies the pig from the region-of-interest feature map with the fixed size and provides other objects.
S3, calculating the body size characteristic data and the overlooking projection area of the pig according to the key points of the pig;
the method for calculating the overlook projection area comprises the following steps: the head and the tail of the pig are divided through key points, the head and the tail of the pig contour line segmentation image obtained by the mask branching unit are removed, and then the area of the rest part in the pig contour line segmentation image is calculated to be used as the overlook projection area.
S4, inputting the pig variety, the body size characteristic data and the overlook projection area into a weight estimation model to obtain the weight data of the pig;
wherein, the weight estimation model in step S4 is established based on a regression model.
The method for establishing the weight estimation model comprises the following steps:
the method comprises the steps of collecting weight data, body size characteristic data, overlooking projection area and pig varieties of pigs to establish a sample library of the pigs, corresponding the characteristics to actually measured weight data to establish a regression model of the pigs, and determining regression coefficients of the regression model of the pigs by using a least square method to obtain an estimated weight model. The regression model includes: a linear regression model, a polynomial regression model, a ridge regression model, a lasso regression model; and (3) testing the weight estimation precision of various regression models in the weight estimation model by using the test set in the sample library to different breeds of pigs, and determining the regression model with the highest precision for each breed of pig.
When the weight estimation model is actually applied, the regression model with the highest precision is determined according to the variety of the pig to be estimated, and the weight estimation model inputs the body size characteristic data and the overlooking projection area into the regression model to obtain the weight data of the pig.
Example 3
A pig weight estimation method based on depth images is shown in figure 1 and comprises the following steps:
s1, obtaining the depth image of the pig taken down and the breed of the pig;
the method for acquiring the depth image comprises the following steps: firstly, a pig is subjected to downward shooting through a depth camera to obtain point cloud data, and then a depth image is generated according to the point cloud data;
when the pig is shot in a depression mode, the depth camera detects the real-time posture of the depth camera through the gyroscope, the depth camera meets the preset shooting visual angle and automatically shoots images, and otherwise, a prompt for adjusting the posture of the depth camera is sent.
S2, extracting key points of the pigs in the depth image by using a Keypoint-RCNN network;
in this embodiment, in step S2, before extracting the key points of the pig in the depth image using the Keypoint-RCNN network, the depth image needs to be segmented to obtain the contour of the pig in the image, and whether the pig in the image is complete is determined, where the segmenting step includes:
s21, carrying out noise reduction processing on the depth image according to a point cloud density filtering noise reduction algorithm to remove noise;
s22, calculating the height difference between the ground and the contour of the back of the pig by a sorting method, identifying the ground part by using the height information and deleting the ground part;
and S23, after the height information converted from the depth information is sequenced, extracting the maximum connected region to obtain the image including the contour of the pig.
In this embodiment, before the key points are extracted, the contour of the pig in the image is segmented, so that whether the pig in the image completely completes picture screening is judged, and it is ensured that the depth image subjected to key point segmentation comprises the complete pig.
The key points of the pigs comprise: ear root point 1, anterior elbow point 2, posterior elbow point 3, posterior spine point 5, and caudal root point 4; wherein, the ear root point 1, the anterior elbow point 2, the posterior elbow point 3 and the tail root point 4 are respectively arranged on the left and the right;
step S3, the body size feature data includes: shoulder width, hip width, body height, body length;
the shoulder width is calculated by the distance between two anterior elbow points 2;
the hip width is calculated by the distance between two back elbow points 3;
the height is calculated by the distance of the difference between the height of the posterior ridge point 5 and the height of the ground;
the body length is obtained by the connecting line distance of the connecting line middle points of the two ear root points 1 and the connecting line middle points of the two tail root points 4.
The Keypoint-RCNN network includes: the system comprises a backbone network unit, a region suggestion unit, a region of interest matching unit, a key point detection branch unit, a mask detection branch unit and a frame regression and classification unit;
the Keypoint-RCNN network structure is shown in FIG. 3, the Keypoint-RCNN network obtains a depth image of a pig, and the backbone network unit extracts image features from the depth image to obtain a feature map of the depth image; the region suggesting unit extracts the region-of-interest feature maps with different sizes from the feature map; the interesting region matching unit converts the interesting region feature maps with different sizes into an interesting region feature map with a fixed size; the key point monitoring branch unit obtains an ear root point 1, an anterior elbow point 2, a posterior elbow point 3, a posterior spine point 5 and a tail root point 4 on a feature map of an interested area with a fixed size; the mask monitoring branch unit divides the characteristic diagram of the region of interest with fixed size to obtain a division diagram of the contour line of the pig; and the frame regression and classification unit identifies the pig from the region-of-interest feature map with the fixed size and provides other objects.
The backbone network unit adopts ResNext.
ResNext is a combination of ResNet and inclusion, on the basis of the inclusion V4, the ResNext simplifies the detail part of the inclusion structure, optimizes the topological structure of each branch, and realizes the characteristics that the parameter quantity is less than the inclusion V4 and the running speed is superior to the inclusion V4. The backbone network unit is used for extracting image characteristics from input original RGBD picture data after network processing.
The region-of-interest matching unit (ROIAlign) uses a bilinear interpolation method to obtain an image numerical value on a pixel point with coordinates of a floating point number.
S3, calculating the body size characteristic data and the overlooking projection area of the pig according to the key points of the pig;
the method for calculating the overlook projection area comprises the following steps: the head and the tail of the pig are divided through key points, the head and the tail of the pig contour line segmentation image obtained by the mask branching unit are removed, and then the area of the rest part in the pig contour line segmentation image is calculated to be used as the overlook projection area.
The head part is the area in front of the connecting line of the left and right ear root points 1 in the pig contour, and the tail part is the area in back of the connecting line of the left and right tail root points 4 in the pig contour.
S4, inputting the pig variety, the body size characteristic data and the overlook projection area into a weight estimation model to obtain the weight data of the pig;
wherein, the weight estimation model in step S4 is established based on a regression model.
The method for establishing the weight estimation model comprises the following steps:
the method comprises the steps of collecting weight data, body size characteristic data, overlooking projection area and pig varieties of pigs to establish a sample library of the pigs, corresponding the characteristics to actually measured weight data to establish a regression model of the pigs, and determining regression coefficients of the regression model of the pigs by using a least square method to obtain an estimated weight model. The regression model includes: a linear regression model, a polynomial regression model, a ridge regression model, a lasso regression model; and (3) testing the weight estimation precision of various regression models in the weight estimation model by using the test set in the sample library to different breeds of pigs, and determining the regression model with the highest precision for each breed of pig.
When the weight estimation model is actually applied, the regression model with the highest precision is determined according to the variety of the pig to be estimated, and the weight estimation model inputs the body size characteristic data and the overlooking projection area into the regression model to obtain the weight data of the pig.
In the embodiment, after the weight data is calculated, the weight data is matched with the RFID of the pig, and is uploaded to the database.
Example 4
A pig weight estimation device comprising: the device comprises a depth camera, a key point extraction module, a body size calculation module and an weight estimation module;
the method comprises the steps of inputting the variety of a pig to be detected, using a depth camera to perform downward shooting on the pig to obtain a depth image, extracting key points of the pig in the depth image by a key point extraction module, calculating body ruler feature data and an overlooking projection area of the pig by the body ruler calculation module by using the key points, and obtaining the weight of the pig by a weight estimation module according to the variety of the pig, the body ruler feature data and the overlooking projection area;
the key point extracting module adopts a Keypoint-RCNN network to extract key points, and the weight estimating module adopts a regression model to estimate weight.
In this embodiment, the pig weight estimation device further comprises a controller and a gyroscope, the gyroscope detects a real-time posture of the depth camera, the controller judges whether a shooting angle of the depth camera meets a preset range through the gyroscope, if not, the controller sends out a correction prompt, and if so, the controller controls the depth camera to shoot images.
An operating personnel utilizes equipment to shoot a pig schematic diagram as shown in figure 2, operating personnel handheld device is directly over the pig, and adjust the shooting angle according to the equipment suggestion, make the shooting angle accord with the preset range, then the automatic image of shooing of equipment, estimate equipment through the pig weight of this embodiment, operating personnel can be at the weight of the direct real-time measurement pig of current aquaculture environment, need not to reform transform aquaculture environment, application scope is wide, and this embodiment adopts the degree of depth camera, can once only obtain the height and width data of the body of pig, the key point extraction rate is fast, the key point position is accurate, the pig weight data of the low error of acquisition that can be fast convenient.
Claims (10)
1. A pig weight estimation method based on a depth image is characterized by comprising the following steps:
s1, obtaining the depth image of the pig taken down and the breed of the pig;
s2, extracting key points of the pigs in the depth image by using a Keypoint-RCNN network;
s3, calculating the body size characteristic data and the overlooking projection area of the pig according to the key points of the pig;
s4, inputting the pig variety, the body size characteristic data and the overlook projection area into a weight estimation model to obtain the weight data of the pig;
wherein, the weight estimation model in step S4 is established based on a regression model.
2. The method for estimating pig body weight based on depth image according to claim 1, wherein the method for obtaining the depth image in step S1 is: firstly, a pig is subjected to downward shooting through a depth camera to obtain point cloud data, and then a depth image is generated according to the point cloud data;
when the pig is shot in a depression mode, the depth camera detects the real-time posture of the depth camera through the gyroscope, the depth camera meets the preset shooting visual angle and automatically shoots images, and otherwise, a prompt for adjusting the posture of the depth camera is sent.
3. The method as claimed in claim 1, wherein the step S2 is to segment the depth image to obtain the pig contour before extracting the key points of the pig in the depth image using the Keypoint-RCNN network, and the step of segmenting includes:
s21, carrying out noise reduction processing on the depth image according to a point cloud density filtering noise reduction algorithm to remove noise;
s22, calculating the height difference between the ground and the contour of the back of the pig by a sorting method, identifying the ground part by using the height information and deleting the ground part;
and S23, after the height information converted from the depth information is sequenced, extracting the maximum connected region to obtain the image including the contour of the pig.
4. The method for pig weight estimation based on depth image according to claim 1, wherein the key points of the pig at step S2 include: ear root point (1), anterior elbow point (2), posterior elbow point (3), posterior spine point (5), and caudal root point (4); wherein the ear root point (1), the front elbow point (2), the back elbow point (3) and the tail root point (4) are respectively arranged on the left and the right;
step S3, the body size feature data includes: shoulder width, hip width, body height, body length;
the shoulder width is calculated by the distance between two front elbow points (2);
the hip width is calculated by the distance between two back elbow points (3);
the height of the body is calculated by the distance of the difference between the height of the back point (5) of the ridge and the height of the ground;
the body length is obtained through the connecting line distance of the connecting line middle points of the two ear root points (1) and the connecting line middle points of the two tail root points (4).
5. The depth image-based pig weight estimation method according to claim 4, wherein the Keypoint-RCNN network of step S2 comprises: the system comprises a backbone network unit, a region suggestion unit, a region of interest matching unit, a key point detection branch unit, a mask detection branch unit and a frame regression and classification unit;
the method comprises the steps that a Keypoint-RCNN network obtains a depth image of a pig, and a backbone network unit extracts image features from the depth image to obtain a feature map of the depth image; the region suggesting unit extracts the region-of-interest feature maps with different sizes from the feature map; the interesting region matching unit converts the interesting region feature maps with different sizes into an interesting region feature map with a fixed size; the key point monitoring branch unit acquires an ear root point (1), an anterior elbow point (2), a posterior elbow point (3), a posterior spine point (5) and a caudal root point (4) on a region-of-interest characteristic diagram with a fixed size; the mask monitoring branch unit divides the characteristic diagram of the region of interest with fixed size to obtain a division diagram of the contour line of the pig; and the frame regression and classification unit identifies the pig from the region-of-interest feature map with the fixed size and provides other objects.
6. The pig weight estimation method based on the depth image as claimed in claim 5, wherein the method for calculating the overlook projection area in step S3 is as follows: the head and the tail of the pig are divided through key points, the head and the tail of the pig contour line segmentation image obtained by the mask branching unit are removed, and then the area of the rest part in the pig contour line segmentation image is calculated to be used as the overlook projection area.
7. The pig weight estimation method based on the depth image according to claim 1, wherein the weight estimation model is established by the following method:
the method comprises the steps of collecting weight data, body size characteristic data, overlooking projection area and pig varieties of pigs to establish a sample library of the pigs, corresponding the characteristics to actually measured weight data to establish a regression model of the pigs, and determining regression coefficients of the regression model of the pigs by using a least square method to obtain an estimated weight model.
8. The method of claim 7, wherein the regression model comprises: a linear regression model, a polynomial regression model, a ridge regression model, a lasso regression model; the regression model corresponding to the different breeds of pigs is determined based on the weight estimation precision of sample data of the different breeds of pigs under each model.
9. The pig weight estimation device for executing the pig weight estimation method based on the depth image according to any one of claims 1 to 8, comprising: the device comprises a depth camera, a key point extraction module, a body size calculation module and an weight estimation module;
the method comprises the steps of inputting the variety of a pig to be detected, using a depth camera to perform downward shooting on the pig to obtain a depth image, extracting key points of the pig in the depth image by a key point extraction module, calculating body ruler feature data and an overlooking projection area of the pig by the body ruler calculation module by using the key points, and obtaining the weight of the pig by a weight estimation module according to the variety of the pig, the body ruler feature data and the overlooking projection area;
the key point extracting module adopts a Keypoint-RCNN network to extract key points, and the weight estimating module adopts a regression model to estimate weight.
10. The pig weight estimation device according to claim 8, further comprising a controller and a gyroscope, wherein the gyroscope detects a real-time posture of the depth camera, the controller judges whether a shooting angle of the depth camera meets a preset range through the gyroscope, if not, the controller sends out a correction prompt, and if so, the controller controls the depth camera to shoot an image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111258590.4A CN114170292A (en) | 2021-10-27 | 2021-10-27 | Pig weight estimation method and device based on depth image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111258590.4A CN114170292A (en) | 2021-10-27 | 2021-10-27 | Pig weight estimation method and device based on depth image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114170292A true CN114170292A (en) | 2022-03-11 |
Family
ID=80477691
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111258590.4A Pending CN114170292A (en) | 2021-10-27 | 2021-10-27 | Pig weight estimation method and device based on depth image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114170292A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115331266A (en) * | 2022-10-17 | 2022-11-11 | 天津大学四川创新研究院 | Pig unique identification duplicate removal alarm method |
CN116452597A (en) * | 2023-06-20 | 2023-07-18 | 厦门农芯数字科技有限公司 | Sow backfat high-precision determination method, system, equipment and storage medium |
CN118171784A (en) * | 2024-05-14 | 2024-06-11 | 正大农业科学研究有限公司 | Feed conversion ratio prediction method and device |
-
2021
- 2021-10-27 CN CN202111258590.4A patent/CN114170292A/en active Pending
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115331266A (en) * | 2022-10-17 | 2022-11-11 | 天津大学四川创新研究院 | Pig unique identification duplicate removal alarm method |
CN115331266B (en) * | 2022-10-17 | 2023-02-10 | 天津大学四川创新研究院 | Pig unique identification duplicate removal alarm method |
CN116452597A (en) * | 2023-06-20 | 2023-07-18 | 厦门农芯数字科技有限公司 | Sow backfat high-precision determination method, system, equipment and storage medium |
CN116452597B (en) * | 2023-06-20 | 2023-08-29 | 厦门农芯数字科技有限公司 | Sow backfat high-precision determination method, system, equipment and storage medium |
CN118171784A (en) * | 2024-05-14 | 2024-06-11 | 正大农业科学研究有限公司 | Feed conversion ratio prediction method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114170292A (en) | Pig weight estimation method and device based on depth image | |
CN110426112B (en) | Live pig weight measuring method and device | |
US20230281265A1 (en) | Method for estimating body size and weight of pig based on deep learning | |
CN112950667B (en) | Video labeling method, device, equipment and computer readable storage medium | |
CN109459119B (en) | Weight measurement method, device and computer readable storage medium | |
CN110609037A (en) | Product defect detection system and method | |
US20100224140A1 (en) | Method and a System for Measuring an Animal's Height | |
CN109636779B (en) | Method, apparatus and storage medium for recognizing integrated ruler of poultry body | |
CN110689519B (en) | Fog drop deposition image detection system and method based on yolo network | |
CN111666855B (en) | Animal three-dimensional parameter extraction method and system based on unmanned aerial vehicle and electronic equipment | |
CN105260750B (en) | A kind of milk cow recognition methods and system | |
Banhazi et al. | Improved image analysis based system to reliably predict the live weight of pigs on farm: Preliminary results | |
CN108491807B (en) | Real-time monitoring method and system for oestrus of dairy cows | |
Wang et al. | Automated calculation of heart girth measurement in pigs using body surface point clouds | |
CN112288793B (en) | Method and device for detecting backfat of livestock individuals, electronic equipment and storage medium | |
CN109740659A (en) | A kind of image matching method and device, electronic equipment, storage medium | |
Ruchay et al. | Accurate 3d shape recovery of live cattle with three depth cameras | |
CN113989353A (en) | Pig backfat thickness measuring method and system | |
CN111145205B (en) | Pig body temperature detection method based on infrared image under multiple pig scenes | |
CN115752683A (en) | Weight estimation method, system and terminal based on depth camera | |
CN111814698A (en) | Method for detecting calf-protecting behavior of cows in pasturing area based on artificial intelligence and aerial images | |
CN113180640B (en) | Size measuring method and system for livestock breeding | |
Liu et al. | Estimation of weight and body measurement model for pigs based on back point cloud data | |
CN113706512A (en) | Live pig weight measurement method based on deep learning and depth camera | |
CN110415246B (en) | Analysis method of abdomen fat component |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |