Developing a stacked ensemble model for predicting the mass of fresh carrot

Accurate estimation of carrot mass plays a vital role in carrot grading and packaging. In this study, a robust prediction method for carrot mass based on machine vision and machine learning (ML) algorithms was proposed. Six morphological features of each carrot were extracted by image processing techniques to build a new dataset with 1254 carrots whose mass ranged from 36 g to 366 g. Several popular ML algorithms were employed to build a stacked ensemble model (EM) for predicting the carrot mass accurately. The individual model and EM were trained using 5-fold cross-validation with features from image processing. Moreover, the effects of different features on the EM were explored and found that the EM with the features except middle diameter (d2) performed best with MAPE, RMSE, and R2 of 1.28 %, 3.02 g, and 0.997, respectively. The proposed EM can be applied as an accurate, objective, and efficient method for on-line grading and packaging of carrots based on mass.