[1N5-GS-13] AI application: Machine learning and application (2)

[1N5-GS-13-05] Using Crowdsourcing to Construct Image Dataset of Beef Cattle's Mounting Action for Estrus Detection

〇Yuriko Kawano¹, Susumu Saito^1,2, Teppei Nakano^1,2, Makoto Akabane^1,2, Ikumi Kondo³, Ryota Yamazaki³, Hiromi Kusaka³, Minoru Sakaguti³, Tetsuji Ogawa¹ (1. Waseda University, 2. Intelligent Framework Lab, 3. Kitazato University)

Keywords:Crowdsourcing, Neural network , Estrus detection, Precision livestock farming

This paper presents an image dataset construction technique using crowdsourcing and object detection for detecting mounting actions of beef cattle, which are known to express estrus signs; and a system for automatic detection of the mounting actions developed based on the constructed dataset.
Each interaction between a pair of cattle was obtained by extracting two cattle regions from the pair and integrating them, thereby using the result as an input of the mounting action detection system.
For achieving high-performance estrus detection, the presented study aims at extracting as many cattle regions as possible, despite a large variety in the cropped images, and for efficiently annotating them by using crowdsourcing.
For reliable crowdsourced annotation, majority voting was employed to determine final labels, considering differences in crowd workers' abilities.
Experimental comparisons using the constructed dataset demonstrated that the developed system was capable of detecting estrus with a precision rate of 80% and a recall rate of 76%.