Non-contact estimation device for body weight of cattle

Abstract

The invention provides a non-contact estimation device for cattle weight, and belongs to the field of machine vision. The system is characterized in that the system is composed of an image data acquisition module, an image information processing module and a neural network algorithm model platform, image acquisition equipment is erected in a farm, and normal operation of equipment power supply and correct transmission of equipment acquisition data are ensured; a farm environment, illumination conditions of different time periods and target detection effects of different distances are considered. Compared with the prior art, the set of equipment is higher in image acquisition efficiency, lower in cost and applicable to different farm environments, including night environment monitoring and image data acquisition.

Description

Technical Field

[0001] The present invention proposes a contactless estimation device for cattle weight, specifically a device that simultaneously collects depth image information and color RGB image information of cattle through a depth camera in an actual breeding environment, utilizes the information provided by these two types of image information data, establishes a corresponding model through machine vision, and realizes contactless estimation of cattle weight, which belongs to the field of machine vision. Background Art

[0002] In cattle breeding, accurate measurement of cattle weight is very important. Cattle weight is not only an important basis for determining market prices, but also has a certain correlation with body size and weight for various physiological indicators of cattle growth cycle, which can effectively evaluate growth status and adjust feeding strategies in time. The traditional method of measuring cattle weight is to set up a separate scale in the corral for measurement, because the instability of cattle will lead to inaccurate measurement results. Frightened cattle will not cooperate or even riot and hurt people during measurement. Therefore, estimating cattle weight through safer machine vision methods is also an important reference indicator. This non-contact weight estimation method reduces interference to cattle, improves work efficiency, and can achieve continuous monitoring, which is convenient for long-term tracking of individual weight changes. It has certain practical significance in the field of combining computer vision and agricultural animal husbandry. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of traditional cattle weighing methods such as low manual weighing efficiency, high risk factor and inability to update data in real time. At the same time, previous machine vision methods have deficiencies such as complex background interfering with detection results and inability to work normally at night. The present invention designs a contactless cattle weight estimation device suitable for complex environments and multiple target detection.

[0004] The present invention proposes a non-contact estimation method for cattle weight, which consists of three parts: image data acquisition module, image information processing module, and neural network algorithm model platform. The image acquisition equipment is set up in the farm to ensure that the equipment power supply can operate normally and the data collected by the equipment is correctly transmitted, taking into account the farm environment, lighting conditions in different time periods, and target detection effects at different distances. Compared with the prior art, this set of equipment has higher image acquisition efficiency and lower cost, and can be applied to different farm environments, including night environments and large farm monitoring and image data acquisition. The image data acquisition module uses a depth camera to simultaneously collect the depth image information and color RGB image information of cattle in the cattle shed of the farm, and passes the initially collected image information to the image information processing module; the image information processing module selects qualified image data according to the standard, first uses the edge algorithm to extract the contour line of the image theme for the color RGB image data, marks the important feature points of the cattle in the image, and then aligns it with the corresponding depth three-dimensional image information to obtain accurate feature point coordinates in three-dimensional coordinates, and then passes the data to the neural network algorithm model platform; the neural network algorithm model platform calculates the body size information based on the coordinate points in the three-dimensional image, and establishes a suitable model based on multiple body size parameters to complete the estimation of the cattle weight. During development, multiple devices use data cables to connect to the host to realize power supply and data transmission of the equipment. In subsequent applications, the entire system and model can be deployed on the end side to achieve fully automated processing.

[0005] The camera used is oak-d-pro-w, which provides the function of collecting color images and depth images, and secondly has night vision and wide angle, and can handle complex application environments and multi-target recognition problems.

[0006] The image information acquisition equipment consists of a tripod and a depth camera, which are connected using a 1 / 4 standard interface. The two cameras are set up on both sides of the cowshed to achieve full coverage of image information of the large cowshed.

[0007] Ultimately, all functions are integrated into an entire software system platform, which can upload video or image information, and then use integrated algorithms to extract key feature points and body size parameters, and estimate weight. The results can be displayed in different types of visualizations to support subsequent work using weight data, such as dynamically adjusting feed feeding according to weight changes.

[0008] The positive effects of the present invention are as follows: (1) The development and use cost of the device and software system based on the present invention is low, the operation is simple, and it is applicable to a variety of complex environments; (2) The present invention innovatively integrates depth information and color RGB information to effectively improve the detection accuracy; (3) The use of this device for contactless estimation of cattle weight can reduce manual risks and manual workload, and automated operation can also improve detection efficiency. The application of this technology can not only improve the traditional weight measurement method, but also promote the development of smart agriculture and bring long-term benefits to animal husbandry. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The present invention is a flow chart of the overall system of a contactless cattle weight estimation device designed by the present invention.

[0010] Figure 2 It is a schematic diagram of the design of the device of the present invention.

[0011] Figure 3 It is a schematic diagram of the structure of the acquisition device of the device of the present invention.

[0012] Figure 4 These are the key body size measurement points for image processing in the present invention and the various body size parameters calculated using them.

[0013] The reference numerals in the above drawings are as follows: Figure 2 Mark in: 1. High-precision depth camera; 2. Tripod; 3. Data transmission line; 4. Data processing platform; 5. Web-side software display DETAILED DESCRIPTION

[0014] The specific implementation of the present invention is described below in conjunction with the accompanying drawings.

[0015] like Figure 1 A contactless estimation method for cattle weight is shown, characterized in that it includes three parts: an image data acquisition module, an image information processing module, and a neural network algorithm model platform. The image data acquisition module uses a depth camera to simultaneously collect depth image information and color RGB image information of cattle in a cattle shed in a breeding farm, and transmits the initially collected image information to the image information processing module; the image information processing module selects qualified image data according to the standard, first uses an edge algorithm to extract the contour line of the image subject for the color RGB image data, marks the important feature points of the cattle in the image, and then aligns it with the corresponding deep three-dimensional image information to obtain accurate feature point coordinates in three-dimensional coordinates, and then transmits the data to the neural network algorithm model platform; the neural network algorithm model platform calculates body size information according to the coordinate points in the three-dimensional image, and establishes a suitable model based on multiple body size parameters to complete the estimation of cattle weight.

[0016] like Figure 2 As shown in the figure, a schematic diagram of a non-contact estimation method for cattle weight in actual use is shown. The entire detection device consists of a depth camera, a tripod, and a data receiving and processing platform. The depth camera and the tripod are integrated using a 1 / 4 standard interface and installed on both sides of the cowshed to collect the depth and color RGB information of the cattle in the cowshed. After the image information is collected, it is transmitted to the central control platform using a data transmission line, and the data processing and model establishment are realized on the platform. This set of equipment can not only realize the deployment of the gold system in the cowshed, but also has convenient mobility and can achieve good results even outdoors.

[0017] like Figure 3 As shown, the camera used to collect image data has a front camera and a back interface. The front camera consists of a 12-megapixel RGB camera and two 1-megapixel depth cameras to form a binocular depth camera; a type-c interface is used on the back to realize all camera power supply and data transmission functions.

[0018] like Figure 4 As shown in the figure, after collecting the image, the edge algorithm is used to obtain the key body measurement point information in the cattle image information, including the cow shoulder point, withers point, ischial node, high point of the cow back, and low point of the cow belly. Then the coordinate information in the depth image is used to calculate the distance between the relevant points to obtain different body size parameters, including cow height, body length, oblique body length, abdominal width and other parameters. These parameters are passed into the appropriate machine learning model to complete the estimation of cattle weight after training.

[0019] The above description is merely an explanation of the present invention, but not a limitation of the present invention. Any simple deformation or modification made without departing from the core of the device described in the present invention shall fall within the protection scope of the present invention.

Claims

1. A contactless estimating device for cattle weight, characterized in that It consists of three parts: image data acquisition module, image information processing module, and neural network algorithm model platform, among which: Image data acquisition module: Use the depth camera to simultaneously collect the depth image information and color RGB image information of the cattle in the cattle shed of the farm, ensure that the image size, acquisition distance and other parameters are the same, and pass the initially collected image information to the image information processing module; Image information processing module: Select qualified image data according to the standards. First, for the color RGB image data, use the edge algorithm to extract the contour line of the image subject, mark the important feature points of the cow in the image, and then align it with the corresponding deep 3D image information to obtain accurate feature point coordinates in 3D coordinates, and then pass the data to the neural network algorithm model platform; Neural network algorithm model platform: calculate body size information based on the coordinate points in the three-dimensional image, and establish a suitable model based on multiple body size parameters to estimate the weight of cattle.

2. The system device according to claim 1, characterized in that In order to be applicable to data collection in a large cattle shed, the acquisition equipment is composed of two high-precision depth cameras placed on both sides of the cattle shed to cover the image information of the entire cattle shed. The acquisition equipment can simultaneously acquire depth images and color RGB images.

3. The system device according to claim 1, characterized in that The image information processing module follows the following steps to build a high-quality experimental data set: S1-1, selecting appropriate key frames from the captured video information, taking into account different breeds, different age groups, different postures, and different background environments of cattle; S1-2, simultaneously obtain the depth information and color RGB information at the same time; S1-3, realize the alignment and registration of depth information and color RGB information; S1-4, edge algorithm is used to extract the outline of the main object (cow); S1-5, extract and mark key feature points, including the cow shoulder point, withers point, ischial node, the lowest point of the cow abdomen and the highest point of the cow back; S1-6. Repeat steps S1-1 to S1-5 to process enough data sets and establish a complete high-quality data set.

4. The method according to claim 3, wherein different body size parameters of cattle are calculated based on the obtained different feature points, comprising: Body oblique length: refers to the distance from the end point of the cow's shoulder to the ischial tuberosity on the same side; body straight length: refers to the horizontal distance from the end point of the cow's shoulder to the vertical line of the ischial tuberosity; Shoulder width: refers to the maximum width of the shoulders on both sides of the cow; abdominal width: refers to the maximum width of the cow's abdomen; body height: refers to the vertical distance from the highest point of the toe to the ground. These body size parameters are calculated using Euclidean distance. After knowing their coordinate information using RGBD image information, the calculation formula is:

5. The system device according to claim 1, characterized in that In order to be applicable to data collection in cowsheds in larger areas, the acquisition equipment is composed of two high-precision depth cameras placed on both sides of the cowshed to cover the image information of the entire cowshed.

6. The system device according to claim 1, characterized in that It includes an application software system that allows users to upload photos, automatically annotate key point information, and then use the developed machine learning model to estimate the weight of the cattle in the photos and visualize it in the software.

Images