Phenological stages analysis in peach trees using electronic nose

Abstract

Abstract Thinning is an expensive and time-consuming management practice used in peach orchards to improve resource distribution among plants and improve production quality. Determining the right time and intensity for thinning is challenging and involves expertise. Furthermore, it generally consumes many hours of work, which makes, in some cases, unfeasible to analyze an entire orchard. For this reason, information that can assist in making making decisions about thinning can improve the cost–benefit ratio of the technique. To mitigate these problems, an electronic nose system, the e-nose, that explores the relationship between the smell of peach trees outdoors and the different growth phases was developed. Twenty-two composed volatile samples were collected from around peach trees (open environment) during its reproductive period (around 39 days) and five supervised machine learning classification algorithms (k-nearest neighbors (KNN), multilayer perceptron (MLP), random forest (RF), logistic regression (LR), and support vector machine (SMV)) were used to analyze the data, to evaluate the possibility of estimating phenological stages from odor environment. The result showed that all models achieved a balanced accuracy greater than 97.5%. As a secondary contribution, the importance of sensors was also analyzed for this application, and a combination of three sensors achieved a classification rate of 100% with the KNN classifier. The e-nose system was successful in distinguishing between petal drop, initial fruit formation, advanced fruit formation, and formed fruit. These results demonstrate the potential of using an electronic nose in a remote system to assist in decision-making in orchard practices as thinning.