CN114882213A

CN114882213A - Animal weight prediction estimation system based on image recognition

Info

Publication number: CN114882213A
Application number: CN202210311372.0A
Authority: CN
Inventors: 张剑
Original assignee: Chengdu Aiji Technology Co ltd
Current assignee: Chengdu Aiji Technology Co ltd
Priority date: 2022-03-28
Filing date: 2022-03-28
Publication date: 2022-08-09

Abstract

The invention discloses an animal weight prediction and estimation system based on picture recognition, which comprises a semantic segmentation module, wherein the semantic segmentation module is in data connection with a key point conversion module, the key point conversion module is in data connection with an intelligent body ruler module, and the intelligent body ruler module is in data connection with a dynamic fusion module; through the intelligent identification technology, a large amount of cost caused by field exploration and pig carrying and weighing is saved for insurance companies, and the processes of exploration, damage assessment, damage checking and settlement and claim settlement of live pig insurance are accelerated.

Description

Animal weight prediction estimation system based on image recognition

Technical Field

The invention relates to the technical field of picture recognition, in particular to an animal weight prediction and estimation system based on picture recognition.

Background

China is a big pig breeding country, the number of lost pigs reaches 7 hundred million in 2018, farmers can pay the live pigs in the live pig breeding industry, when the live pigs die abnormally due to epidemic diseases and accidents in the breeding process, insurance companies can pay the farmers according to the weight of the dead live pigs, so that when the settlement is carried out, the weight of the live pigs is determined to become a necessary step and task before the settlement and payment of the insurance companies, if the dead live pigs can be weighed, the insurance companies can obtain accurate weight, but in the practical operation, a plurality of pigs cannot be weighed, for example, the epidemic pigs need to be subjected to quick harmless treatment to check the damage by the insurance companies, or the pigs cannot be manually weighed due to sanitary reasons or excessive weight, the weight of the pigs can only be estimated, in the traditional method, some insurance companies can estimate the weight of dead pigs by investigators or workers of a harmless treatment station to obtain a rough result, but the estimation cannot guarantee the precision and is easily influenced by the experience of the estimators; secondly, the manual treatment brings an operation space for estimating the benefits between people and farmers, and is not beneficial to the prevention and control risks of insurance companies.

Disclosure of Invention

The present invention is directed to a system for estimating and predicting animal weight based on image recognition, so as to solve the problems mentioned in the background art.

In order to achieve the purpose, the invention provides the following technical scheme: the animal weight prediction and estimation system based on image recognition comprises a semantic segmentation module, wherein the semantic segmentation module is in data connection with a key point conversion module, the key point conversion module is in data connection with an intelligent body scale module, the intelligent body scale module is in data connection with a dynamic fusion module, the dynamic fusion module is in data connection with a data storage module and a data interaction module respectively, and an automatic updating module is arranged on one side of the data interaction module.

Preferably, the semantic segmentation module comprises a picture reading module, a picture classification module, a picture processing module and a detection unit modeling module.

Preferably, the key point conversion module includes a key point labeling module, a key point modeling module, a key point extracting module, and a key point integrating module.

Preferably, the intelligent body scale module comprises a body scale conversion module, a body scale adjusting module, a body scale stabilizing module and a body scale feature combination module.

Preferably, the dynamic fusion module comprises a heterogeneous fusion module, a weight display module, a confidence index display module and a weight prediction module.

Preferably, the weight prediction module comprises a first estimation module, a second estimation module, a third estimation module and a fourth estimation module.

Preferably, the body size information processed by the intelligent body size module includes body length, chest circumference, waist circumference, leg interval, core trunk area, and the like.

Preferably, the image processing module processes the image by using methods such as image enhancement and environmental noise reduction.

Compared with the prior art, the invention has the beneficial effects that: the method creatively provides that by shooting the dead pigs, modeling is carried out on a large number of historical pig photos and pig weights based on deep learning and machine learning technologies, and learning how to capture, identify and calculate the key body size information related to the pig weights from the images, the weights of the pigs are predicted and calculated, a solid foundation is laid for settlement and calculation claims, case quality inspection and risk control of insurance companies, calculation accuracy is improved, and errors are reduced; the intelligent identification technology saves a large amount of cost for insurance companies caused by field exploration and pig carrying and weighing, improves the risk prevention and control level of the insurance process, greatly accelerates the processes of exploration, loss assessment, loss checking and claim settlement of live pig insurance, improves the working efficiency, is beneficial to enabling farmers to take claim money of the insurance companies more quickly, and improves the insurance acceptance and claim settlement experience and satisfaction of the insurance companies.

Drawings

FIG. 1 is a block diagram of the system architecture of the present invention;

FIG. 2 is a block flow diagram of the present invention;

FIG. 3 is a system flow diagram of the present invention;

in the figure: 1. a semantic segmentation module; 2. a key point conversion module; 3. an intelligent body ruler module; 4. a dynamic fusion module; 5. a data storage module; 6. a data interaction module; 7. an automatic update module; 101. a picture reading module; 102. a picture classification module; 103. a picture processing module; 104. a detection unit modeling module; 201. a key point labeling module; 202. a key point modeling module; 203. a key point extraction module; 204. a key point integration module; 301. a body ruler conversion module; 302. a body ruler adjusting module; 303. a body ruler stabilizing module; 304. a body size feature set matching module; 401. a heterogeneous fusion module; 402. a weight display module; 403. a confidence index display module; 404. a weight prediction module; 4041. a first estimation module; 4042. a second pre-estimation module; 4043. a third estimation module; 4044. and a fourth estimation module.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1-3, an embodiment of the present invention is shown: an animal weight prediction estimation system based on picture recognition comprises a semantic segmentation module 1, wherein the semantic segmentation module 1 is in data connection with a key point conversion module 2, the key point conversion module 2 is in data connection with an intelligent body scale module 3, the intelligent body scale module 3 is in data connection with a dynamic fusion module 4, the dynamic fusion module 4 is in data connection with a data storage module 5 and a data interaction module 6 respectively, an automatic updating module 7 is arranged on one side of the data interaction module 6, the semantic segmentation module 1 comprises a picture reading module 101, a picture classification module 102, a picture processing module 103 and a detection unit modeling module 104, the key point conversion module 2 comprises a key point marking module 201, a key point modeling module 202, a key point extracting module 203 and a key point collecting module 204, and the intelligent body scale module 3 comprises a body scale conversion module 301, a body scale conversion module 204, a dynamic body scale module 3 and a dynamic fusion module 4, The body size adjusting module 302, the body size stabilizing module 303 and the body size feature set matching module 304, the dynamic fusion module 4 includes a heterogeneous fusion module 401, a weight display module 402, a confidence index display module 403 and a weight prediction module 404, the weight prediction module 404 includes a first prediction module 4041, a second prediction module 4042, a third prediction module 4043 and a fourth prediction module 4044, the body size information processed by the intelligent body size module 3 includes body length, chest circumference, waist circumference, leg distance, core trunk area and the like, and the picture processing module 103 processes pictures by adopting methods of image enhancement, environmental noise reduction and the like.

The working principle is as follows: in the using process of the invention, firstly, a picture of a pig is shot by using a mobile terminal or a camera, a reference object is required to be placed in front of, behind, on the body or in the place of the pig before shooting, A4 paper is currently selected as the reference object, the picture of the pig is shot and then processed by using a semantic segmentation module 1, firstly, a picture reading module 101 reads the shot picture, then, a picture classification module 102 is used for classifying the picture, after the classification is finished, a picture processing module 103 is used for carrying out image enhancement and environmental noise reduction on the picture, interference noise is removed, the image quality is improved, then, a semantic segmentation model is established by using a detection unit modeling module 104, wherein the semantic segmentation classification model of the pig and the semantic segmentation model of the reference object are respectively two models, after the model is established, the model is transmitted to a key point conversion module 2 for processing, a key point labeling module 201 labels key points on the processed image, then, a key point modeling module 202 is used for modeling the probability density distribution of the key point positions through data generation algorithms such as a GMM Gaussian mixture model and the like, so that the best estimation of the positions of key points required by body size calculation is obtained, then, a key point extraction module 203 is used for dividing the image into a plurality of regions, feature points in each region are extracted, a key point aggregation module 204 generates a key point aggregation after the extraction is completed, then, the data are transmitted to an intelligent body size module 3, the pig body size data including information such as body length, chest circumference, waist, leg spacing and core trunk area are constructed through the intelligent body size module 3 by aggregating the key points, then, the real size of the predicted object is calculated on the premise that the pixel size and the real size of a reference object are obtained through a body size conversion module 301, and then, a shooting distance, the method comprises the steps of correcting the real size of a predicted object to standard side-view measurement based on information such as a shooting angle, a relative position of a reference object and the predicted object, ensuring the accuracy of the real size of the predicted pig, stabilizing body scale data by a body scale stabilizing module 303, collecting the real sizes of all parts of the predicted pig by a body scale feature set matching module 304, transmitting the body scale feature set data to a dynamic fusion module 4, fitting a group of machine learning machine regressors including linear regression, K neighbor regression, LASSO regression, a support vector machine, a multi-layer sensing machine, a Gaussian mixture model and the like in a body weight predicting module 404 according to the corresponding relation between historical body scales and body weights by a first pre-estimating module 4041, a second pre-estimating module 4042, a third pre-estimating module 4043 and a fourth pre-estimating module 4044, and then adopting Adaboost, GBRT and the like based on the machine learning machine regressors, The method comprises the steps of performing integration and reinforcement on integrated learning algorithms such as Bagging, RandomForest, Stacking and the like to obtain predicted weight data of pigs, transmitting the weight data to a heterogeneous fusion module 401 for fusion judgment, displaying predicted weight information of the pigs by a weight display module 402, displaying a predicted confidence index by a confidence index display module 403, transmitting the data to a data storage module 5 for storage, transmitting the data to a data interaction module 6 for transmission to a cloud server, and updating the system by using an automatic updating module 7 when the system needs to be updated.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims

1. An animal weight prediction estimation system based on image recognition comprises a semantic segmentation module (1), and is characterized in that: the semantic segmentation module (1) is in data connection with the key point conversion module (2), the key point conversion module (2) is in data connection with the intelligent body ruler module (3), the intelligent body ruler module (3) is in data connection with the dynamic fusion module (4), the dynamic fusion module (4) is in data connection with the data storage module (5) and the data interaction module (6) respectively, and the automatic updating module (7) is arranged on one side of the data interaction module (6).

2. The system of claim 1, wherein the system comprises: the semantic segmentation module (1) comprises a picture reading module (101), a picture classification module (102), a picture processing module (103) and a detection unit modeling module (104).

3. The system of claim 1, wherein the system comprises: the key point conversion module (2) comprises a key point labeling module (201), a key point modeling module (202), a key point extracting module (203) and a key point collecting module (204).

4. The system of claim 1, wherein the system comprises: the intelligent body scale module (3) comprises a body scale conversion module (301), a body scale adjusting module (302), a body scale stabilizing module (303) and a body scale feature set matching module (304).

5. The system of claim 1, wherein the system comprises: the dynamic fusion module (4) comprises a heterogeneous fusion module (401), a weight display module (402), a confidence index display module (403) and a weight prediction module (404).

6. The system of claim 5, wherein the system comprises: the weight prediction module (404) comprises a first estimation module (4041), a second estimation module (4042), a third estimation module (4043) and a fourth estimation module (4044).

7. The system of claim 1, wherein the system comprises: the body size information processed by the intelligent body size module (3) comprises body length, chest circumference, waist circumference, leg distance, core trunk area and the like.

8. The system of claim 2, wherein the system comprises: the picture processing module (103) processes pictures by adopting methods such as image enhancement, environmental noise reduction and the like.