Desenvolvimento de manejo de alface em hidroponia por rede neural artificial convolucional

• Ano de defesa: 2024
• Autor(a) principal: Leal, Eduardo Ribeiro Pereira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Medianeira; Brasil; Programa de Pós-Graduação em Tecnologias Computacionais para o Agronegócio; UTFPR
• Palavras-chave em Português: Inteligência artificial; Aprendizado do computador; Algorítmos computacionais; Artificial intelligence; Machine learning; Computer algorithms; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/33800

It is a major challenge obtaining higher yields in production or even reducing waste, especially when combining productivity and quality. Environmental preservation and lack of automation technology to assist the producer in monitoring their crops are some issues encountered. Hydroponic cultivation in a controlled environment offers a sustainable option for fast and healthy production, with a reduction in the negative impacts of soil-based planting. Sensors and cameras use allows for non-invasive plants data collection, without harming them. This study aimed to detect and classify image patterns of hydroponic green leaf lettuce (Lactuca sativa L.) subjected to irrigation system failures using Convolutional Neural Network (CNNs). Due to this paper, the experimental design used was completely randomized, evaluating four treatments - two with water stress and one without stress but with the same sizes, with five replicates of each cultivation, totalizing twenty samples with controlled interruption of water and nutrient flow for thirty minutes to simulate hydro stress, analyzing wet mass (WM), number of leaves larger than 10 cm (NL10), and dry mass (DM). TensorFlow and Keras were used for detection and classification, employing Central Processing Unit (CPU), and Yolo with Darknet using Graphics Processing Unit (GPU). Controlled water stress simulation results showed statistical differences for the three observed variables, as per the Tukey test with 5% probability, indicating a 45% mass loss, 29% decrease in leaf number, and 43% decrease in wet mass. The average accuracy using TensorFlow was 83%, lower than Yolo which achieved 98% Mean Average Precision (mAP). These results suggest that digital image processing with artificial convolutional neural networks is a promising approach for pattern detection in hydroponic cultivation. It is expected that the contributions of this study will be applied in future computer vision developments, aiding in the fast crisis identification, preventing losses and achieve higher production yields and waste reduction.