Browsing by Author "Corceiro, Ana Catarina Antunes"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Artificial Intelligence-Powered Classification of Flora in VineyardsPublication . Corceiro, Ana Catarina Antunes; Gaspar, Pedro Miguel de Figueiredo Dinis Oliveira; Pereira, Nuno José MatosRapid global population growth has put considerable pressure on the agricultural sector, which is forced to grow by 70 % in order to meet the growing demand for food. The biggest challenge facing agricultural producers is the existence of plants that become weeds. These not only compete with crops for vital resources, but also pose significant economic and environmental threats. In the context of comprehensive vineyard management, vegetation control becomes especially critical, influencing the prevalence of various pests and emphasising the need for environmentally sustainable plant control methods. Although herbicides were initially used to control the growth of unwanted plants, the excessive and indiscriminate use of these chemicals leads to environmental pollution and the development of resistance on the part of weed vegetation. In order to deal with these challenges, the agricultural sector has been relying on ML models. Rapid advances in Artificial Intelligence (AI), particularly Convolutional Neural Networks (CNN), have emerged as a predominant solution for solving problems related to this type of plant. Precision Agriculture (PA) has taken advantage of ML algorithm technology as a resource for classifying images to distinguish cultivated and undesirable plants. This work aims to contribute to the promotion of sustainable agriculture and to the advancement of image classification in the field of plant classification. The main objectives involve the development of algorithms using CNN for classification using the PyTorch framework, with a focus on hyperparameter optimisation. A comprehensive review of the area emphasises plant diversity and uniqueness, as well as data acquisition methods, vegetation indices used, evaluation metrics, CNN as well as its PyTorch models. It also analyses recent advances in ML models and methods for detecting and classifying weeds to improve the sustainability of agricultural crops. The results highlight the excellent performance of the MaxVit, ShuffleNet and EfficientNet models, especially when faced with an expanded dataset. The choice of hyperparameters, including learning rate, layer configuration and weight reduction, significantly influenced model accuracy. When testing the models in a web application, EfficientNet_B1 and EfficientNet_B5 obtained an exceptional accuracy of 96.15 per cent, standing out among all the models. These models have the potential to revolutionise agriculture, increasing productivity and environmental sustainability. By integrating these models into technology solutions, farmers can monitor crop health, identify pests and undesirable plants, and optimise the use of natural resources such as water and fertiliser. Automating the detection of crop problems reduces crop waste and minimises the use of herbicides, promoting organic farming. In addition, through the prediction of weather conditions and plant growth patterns, farmers can optimise their planting and harvesting schedules to maximise productivity. Technology is also crucial to precision farming, enabling individualised treatment of crops and a reduction in wasted resources. By creating real-time monitoring systems, farmers can make informed decisions and improve crop adaptability in the face of climatic challenges. Ultimately, by incorporating PyTorch models, farming becomes more efficient, minimises waste and reduces environmental impact, contributing to a greener and more resilient future for agriculture.