ABSTRACT

Body weight measurement of animals is often labor-intensive for farmers and stressful for animals. To this end, several methods have been researched and implemented to automate this process. In this study, we performed a Systematic Literature Review to identify and synthesise the published studies on the body weight estimation approaches for livestock (i.e. cattle and pigs). Information about features of models, underlying methods, performance evaluation parameters, challenges, and solutions using computer vision-based weight estimation, and characteristics of the future vision-based weight estimation models were presented based on the identified scientific papers. We found 151 papers, of which 26 papers were selected as primary studies that we analyzed in detail. We identified that: (1) seven features, namely top view body area, withers height, hip height, body length, hip-width, body volume, and chest girth are widely used in approaches; (2) 3D Time of Flight camera is the most preferred one; (3) the linear regression is the most used algorithm; (4) the application of Deep Learning algorithms is still very limited; and (5) coefficient of determination is the most used evaluation parameter for weight estimation. In addition to these observations, 13 challenges, 22 solutions, and guidelines for future research direction were presented.

摘要

对动物的体重测量对于农民而言通常是劳动密集型的，并且对动物而言压力很大。为此，已经研究并实现了几种方法来使该过程自动化。在这项研究中，我们进行了系统文献综述，以鉴定和综合已发表的有关牲畜（即牛和猪）体重估算方法的研究。基于已识别的科学论文，介绍了有关基于模型的特征，基础方法，性能评估参数，挑战和解决方案（使用基于计算机视觉的权重估计）的信息，以及未来基于视觉的权重估计模型的特征。我们发现151篇论文，其中26篇被选为我们详细分析的基础研究。我们确定：（1）七个特征，即顶视图身体区域，枯萎高度，臀部高度，身长，臀部宽度，身体体积和胸围被广泛采用。（2）3D飞行时间相机是最喜欢的一种；（3）线性回归是最常用的算法；（4）深度学习算法的应用仍然非常有限；（5）确定系数是权重估计最常用的评估参数。除了这些观察，还提出了13个挑战，22个解决方案和未来研究方向的指南。（5）确定系数是权重估计最常用的评估参数。除了这些观察，还提出了13个挑战，22个解决方案和未来研究方向的指南。（5）确定系数是权重估计最常用的评估参数。除了这些观察，还提出了13个挑战，22个解决方案和未来研究方向的指南。