Techniques for cattle detection, duplicate removal and counting in large pasture areas using multiple aerial images

• Ano de defesa: 2023
• Autor(a) principal: Soares, Victor Hugo Andrade
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Cattle counting; Contagem de gado; Detecção de objetos; Duplicate removal; Fazenda de precisão; Graph-based method; Livestock; Método baseado em grafos; Object detection; Pecuária; Precision farming; Remoção de duplicados; Unmanned aerial vehicles (UAVs); Veículos aéreos não tripulados
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-13052024-091728/

Among the production areas with largest impact on global economy, agriculture and livestock play a prominent role. Technologies have been developed in order to automate and increase the efficiency of these fields. The use of Unmanned Aerial Vehicles (UAVs) has been extensively investigated to improve the efficiency of agricultural production and in the monitoring of animals. One of the most important and challenging tasks in animal monitoring is cattle counting. Traditional manual counting methods are laborious and error-prone, while existing automated approaches struggle with duplicate animal detection. This work presents a method for detecting and counting cattle in aerial images acquired via UAVs. This method leverages Convolutional Neural Networks (CNNs) and employs a graph-based optimization technique to eliminate duplicate animal detection in overlapping images. Our results emphasize the importance of maximizing animal matching to mitigate duplicate counts. Additionally, we integrate multi-attributes, encompassing velocity, direction, state (lying down or standing), color, and distance, to enhance duplicate removal and counting precision. We conducted extensive experiments and training to seamlessly incorporate these attributes into our methodology. Furthermore, we provide a dataset comprising authentic images captured in extensive pasture areas, suitable for both training and testing/benchmarking cattle counting techniques. When evaluating detection and counting, our outcomes underscore the competitiveness of the proposed method while significantly reducing the computational cost of the overall counting process. When focusing solely on duplicate removal, our method surpasses state-of-the-art techniques, achieving an average percentage error of 2.34%. In summary, the proposed method marks a substantial stride towards more efficient cattle counting practices and enhanced livestock management in agriculture.