Method Applied To Animal MonitoringThrough VANT Images

• Published in: Revista IEEE América Latina 2020-07, Vol.18 (7), p.1280-1287
• Main Authors: Campos de Vasconcellos, Bruno, Pereira Trindade, Jose Pedro, Bochi da Silva Volk, Leandro, Bidese de Pinho, Leonardo
• Format: Article
• Language: English
• Subjects: Aerial surveying; Aerial surveys; Agriculture; Animals; Artificial neural networks; cattle herding; Collection; convolutional neural network; Convolutional neural networks; Counting; deep learning; Graphics processing units; Image resolution; Livestock; Machine learning; Monitoring; Remote sensing; Unmanned aerial vehicles
• ISSN: 1548-0992; 1548-0992
• DOI: 10.1109/TLA.2020.9099770

One of the necessary demands in extensive livestock systems is the counting of animals in areas of tens of hectares, costly when carried out manually and locally. In this context, this work proposes and discusses the efficacy of a semi-autonomous, non-invasive method for remote identification of animals in the field, applicable to precision livestock systems. The method was conceived from an exploratory research methodology based on remote sensing techniques that include image collection processes by aerial surveying with RGB camera embedded in unmanned aerial vehicle, persistence of images obtained by means of storage in space-time databases and processing of stored images for the construction of a rural property orthomosaic succeeded by the application of patterns discovery processes, making use of deep learning, especially convolutional neural networks. According to the experiments carried out, the method was effective, being able to identify and count animals from the collection of images made at 100 m height, with an accuracy of up to 95%, including the approximate geographical position of the animals to field.