Intelligent scoring system for dairy cow body condition based on deep learning and long-distance video

Abstract

The invention particularly relates to an intelligent scoring system for dairy cow body condition based on depth learning and remote video, comprises a cow walking channel, an ear tag reader, a 3D camera and a processing module, wherein the ear tag reader is used for reading cow ear tag information, the 3D camera is used for shooting a 3D image of the cow, and the processing module adopts a depth learning algorithm Faster R-CNN processes the received cow 3D image to obtain cow scoring data, associates the cow scoring data with cow ear tag information, and outputs and/or stores the cow ear tag information. The original 3D images of cow body were obtained by 3D camera and the feature database of cow BCS images was constructed. The sample set was trained by depth learning algorithm and the classification accuracy of BCS was 0.1 point. Then, 3D infrared camera and electronic ear tag are used to construct a model that can continuously track the healthy growth and production of dairy cows inthe life cycle, so that the system can identify and early warn the abnormal condition of dairy cows in advance.

Description

technical field [0001] The invention relates to the technical field of intelligent monitoring equipment, in particular to an intelligent scoring system for cow body condition based on deep learning and remote video. Background technique [0002] Dairy Cow Body Condition Score, English: Dairy Cow Body Condition Score, referred to as BCS, is the best evaluation system summarized by the international animal husbandry in the past 30 years. It can objectively and reasonably evaluate the energy reserve of individual dairy cows and reflect the animal diet more scientifically status, reproductive capacity, etc. There are many forms of BCS scoring system. In addition to the 8-point (1-8) BCS system proposed by Australia Earle et al. in the early days, the 6-point (1-6) system proposed by British scholars Mulvaney et al., and the New Zealand Macdonald et al. 10 points (1 to 10) to make the BCS system. Of course, the 5-point (1-5) BCS system proposed by Wildman in the United States i...

AI Technical Summary

Problems solved by technology

Solution 1: Zhang Hailiang and He Dongjian used image processing technology to analyze and detect images of beef cattle, and indirectly measured the functional indicators of body length, chest width, and chest depth, and then used the measured values ​​to score body conditions. However, due to indirect measurement errors Larger, resulting in lower accuracy of body condition scoring; scheme two, Wu Fuzheng et al. focused on the study of dairy cow tail images, collected a large number of dairy cow tail images to establish an image library, and proposed a method of scoring cow body condition based on Kernel-PCA, but this scheme The error range is too large, reaching ±0.5; scheme 3, Fischer et al. use a 3D camera to obtain a depth map of the cow’s back, and manually mark four key points, combined with the principal component analysis method, use the multiple current regression model for modeling, and conduct body condition analysis. Scoring, but the system cannot automatically score the body condition, and the cost of the 3D camera is relatively high; scheme 4, Spoliansky et al. use the 3D Kinect camera to automatically score the body condition, but the model needs additional information, such as cow weight, age, etc., and the accuracy It can only reach 0.75, which is far from the artificial score; scheme five, the method of using 3D cameras by Fischer et al. and the Kinect camera used by Spoliansky et al. have binocular structures inside, but these cameras are relatively fragile and cannot be applied to complex farms. Extreme environment of light, wind and sun

Images