Development of Self-Training Algorithm for Predicting Mango Maturity

The quality and maturity of mangoes are inhomogeneous, even when mangoes are harvested from the same tree at the same time, however, the maturity of mangoes greatly affects the storage and transport time. Therefore, the determination of mango maturity is very important. This study aims to determine the mango maturity by using the internal and external features of mangoes (length, width, defect, weight, density, and color) based on a hybrid model of a multilayer Feed-Forward Neural Network (FFNN). In detail, the mango is segmented based on analyzing color space then algorithms in image processing are applied. After determining the architecture, the FFNN model is trained with the dataset in which each data point has 14 features. Another self-training algorithm is applied to increase the accuracy of FFNN. The proposed system has a mean-square error of 0.259 in maturity prediction which is shown in the results and experiments section.