Fenotipagem de variedades crioulas de feijão-caupi em diferentes estádios fenológicos com uso de inteligência artificial.

Abstract

The objective of this work was to perform the phenotyping of cowpea landraces at different phenological stages using artificial intelligence techniques for digital image processing and adjustment of machine learning models. Anthropogenic actions are related to global climate change, which reinforces the role of family farmers as guardians of seeds of landraces of cowpea. Because there are many varieties, there is a risk of confusion in identifying these materials. For that, six landrace varieties were cultivated to obtain digital images at different phenological stages, which were processed using the vectors InceptionV3, SqueezeNet, VGG16 and VGG19. Subsequently, the k-nearest neighbors (KNN - number of nearest neighbors), decision tree (Tree), random forest (RF - Random Forest), gradient boost (GB - Gradient Boosting), vector support machine (SVM - Support Vector Machines) and artificial neural network (MLP - Multi-Layer Perceptron). The performance of the models was tested using the cross-validation method. Machine learning algorithms such as Artificial Neural Network and Vector Support Machine have high performance for phenotyping of cowpea landrace varieties at different phenological stages from digital image processing.