No abstract

<span>Sorting and grading are qualitative operational tasks performed in food processing industries. Realization of higher accuracy in mass estimation is the key inclination of this work. In this work, an automated technique for mass estimation of citrus limetta is devised based on the geometrical features derived from pre-processed images. Dataset includes 250 data samples of citrus limetta, whose images are acquired in different orientations. Two novel handcrafted distance-based geometrical features along with four conventional geometrical features were employed for regression analysis. Predictive modeling is conducted with configuration of 150 training and 100 testing data samples and subject to regression analysis for mass estimation. Multiple linear and support vector regression models with linear, polynomial and radial basis function (RBF) kernels were employed for mass estimation with two different model configurations, conventional and conventional with handcrafted features, for which an R2 score of 0.9815, root mean squared error (RMSE) of 10.94 grams, relative averages of accuracy and error of 96.61% and 3.39% respectively is achieved for the proposed model and configuration which was validated using k-fold cross-validation. Through comparison with performance of model with conventional and conventional with handcrafted features configurations, it was established that inclusion of handcrafted features was able to increase the performance of the models.</span>