A cost-effective image-based machine learning framework for automating active iron estimation in Peach (Prunus persica (L.) Batsch) Leaves

Iron (Fe) deficiency significantly affects the quality of peaches in temperate zones. Conventional iron assessment methods, such as laboratory analysis, are time- and resource- consuming. This study presents an image-based diagnostic tool for rapid estimation of the iron status in peach (Prunus persica (L.) Batsch) orchards. Therefore, a dataset of 1000 peach leaves from 200 trees across 65 orchards was collected and photographed. The leaves were then labeled based on the active iron (Fe2+) concentration, as determined by the orthophenanthroline extraction and atomic absorption spectrometry in the laboratory. Thirty-six different features were extracted from the images and analyzed using linear regression and neural network models. Subsequently, the most relevant features were identified by stepwise linear regression and sensitivity analysis to refine the models. The optimal linear model, incorporating blue difference chroma (Cb), lightness (L), green-minus-red (GMR), and normalized red index (NRI), achieved an [Formula: see text], [Formula: see text], and [Formula: see text]. The neural network model, with selected features showed slightly better accuracy ([Formula: see text], [Formula: see text], and [Formula: see text]) compared to the linear model. The results indicate the potential of this method as an in-situ alternative to laboratory analysis for estimating active iron in peach leaves.