从人工到智能:牛个体识别技术研究进展

From Artificial to Intelligent:Research Progress of Individual Idendification Technology for Cattle

牛个体识别技术是实现牛的智能测重、体况评分、体型鉴定、行为监测等自动化技术的先决条件,各种用于牛个体识别的设备和方法在不同的时间段被提出。作者首先对不同牛个体识别方法进行分类阐述,介绍了传统识别方法、生物特征识别方法和深度学习方法在牛个体识别中的研究进展,特别是当前深度学习方法在实际应用中的难点,然后详细分析了不同识别方法的优缺点。传统识别方法如耳切、耳纹、热烙等方法,依靠人工对牛进行识别,识别准确率和识别效率低,忽视了动物福利和标记持久性;无线射频识别技术将牛个体识别由人工识别转向了自动识别,提高了识别效率,但数据安全无法得到保障。随着图像识别技术的发展和深度学习方法的崛起,基于生物特征识别方法和深度学习方法的牛个体识别技术实现了非接触、安全高效的牛智能识别,但基于鼻纹印、视网膜血管和虹膜的生物特征识别方法因理想图像获取难度较大,实用性较差。深度学习方法通过深度神经网络学习图像特征,更适用于复杂条件下的图像应用,在真实的牛场养殖环境中具有广泛的应用前景和潜在价值。作者还对比了不同识别方法在各方面的差异,并在此基础上对牛个体识别技术的研究前景进行了展望。

Individual identification technology was the primary requirement of automatic production techniques,including intelligence weighting,body condition scoring,body shape identification,and behavior monitoring.Individual cattle identification methods were categorized and elaborated in current review,that the research progress of different individual cattle identification methods including traditional,biometric,and deep learning methods were introduced,especially the difficulties of applying deep learning method in practice were discussed.Moreover,the advantages and disadvantages of different identification methods were analyzed.Traditional identification methods such as ear notching,ear tattooing,and hot branding identified cattle relies on manual labor,which had low identification accuracy and efficiency,and ignored animal welfare and mark persistence.Radio frequency identification techniques provided an automatic identifying protocol that elevated the efficiency but the data security could not be guaranteed.With the development of image recognition technology and deep learning methods,non-contact,safe and efficient cattle intelligent recognition had been realized with the technology based on biometric recognition and deep learning methods.However,the biometric identification method based on nose print,retinal blood vessels and iris had poor practicability,because the ideal image was difficult to obtain.Based on deep neural networks learning of image features,the deep learning method was more practical and promising in complex dairy farm circumstances.What’s more,different individual identification methods were compared,and the research on individual cattle identification technology was prospected in current review.