Real time weed detection using computer vision and deep learning

• Autor(a) principal: Luiz Carlos, M. [UNESP]
• Data de Publicação: 2021
• Outros Autores: Ulson, José Alfredo C. [UNESP]
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1109/INDUSCON51756.2021.9529761; http://hdl.handle.net/11449/222501
• topic: Agroindustry; Deep learning; Real time; Weed detection; YoloV5

Maintain a high crop yield and yet manage with efficiency and sustainability the resources use is one of the biggest challenges that the agroindustry sector faces. Among these challenges highlights the control of weeds and pests in the field, since many weed species present resistance for the most used commercial herbicides. Detect these weed species through computer vision and deep learning is a possible solution, once with local detection weeds can be removed by mechanical, chemical or electrical systems, significantly reducing environmental impacts due to excessive use of herbicides and economic losses caused by weeds. Therefore, in this work, it is proposed and explored a real time weed detection system, based on the YoloV5 architectures. The architectures performance was evaluated without and with transfer learning on a custom dataset based on 5 weed species resistant to Glyphosate. Results indicate that the system is functional, being able to correctly detect the resistant weeds at 62 FPS.