High-precision prototype for garlic apex reorientation based on artificial intelligence models

Sowing and harvesting are the most expensive operations in garlic cultivation (Allium sativum L.). For mechanized sowing to be feasible, the garlic clove must be placed in the soil with the apex pointing upwards, otherwise, yield can be reduced by up to 23%. In this context, artificial intelligence (AI) emerges as a viable solution to address these issues, particularly artificial neural networks (ANN). This research presents the development and evaluation of a garlic apex orientation device, which utilizes AI models adapted to all types of garlic clove shapes. The evaluated models are support vector machine (SVM), random forest (RF), ANN, convolutional neural network (CNN), and transfer learning (TL). To increase the number of available images for training, a generative adversarial network (GAN) was used. Three different databases were used to train models to determine achieved the best performance in terms of model accuracy. The databases used are the original database, an augmentation version of the original database incorporating images generated by the GAN model, and only images generated by the GAN model. The results show that the best model (ANN) achieves a validation accuracy of 99.74% when using an augmentation of the original database with artificial images generated by the GAN model.