Título: Machine Learning for Monitoring the Spread of Carmine Cochineal on Different Accessions of Forage Palm Plants
Autores: Josenalde Barbosa de Oliveira, João Vitor Gomes Vieira, Marcone Chagas
Resumo: Entomologists and experts on the mitigation and control of agricultural pests need to evaluate the resistance of different species and accessions within the same plant species, which is usually carried out manually with visual inspection. This paper presents an automatic approach for carmine cochineal identification and counting based on machine learning algorithms. A dataset is created from actual experiments and through a pipeline for feature engineering, getting both morphological and color features. The models Gradient Boosting, XGBoost, Stochastic Gradient Descent (SGD), Multilayer Perceptron Neural Network (MLP) and Logistic Regression are evaluated in a binary classification task. XGBoost and Gradient Boosting outperform the other models, with an accuracy of 93% on the test set, with a good generalization property and without the need of deep learning or more complex models. Moreover, the solution can be embedded into a software to aid experts on these agronomic comparisons for proper pest management.
Palavras-chave: supervised learning; image processing; agricultural pests; emsemble models; artificial intelligence; entomology.
Páginas: 6
Código DOI: 10.21528/CBIC2025-1176020
Artigo em PDF: CBIC_2025_paper1176020.pdf
Arquivo BibTeX:
CBIC_2025_1176020.bib