Classification of Lemon Quality Using the Residual Convolutional Neural Network Deep Learning Model

Abstract

Determining lemon quality is an important post-harvest process. Lemon producers want the lemon quality classification process to be done correctly. In this way, the monetary return they will receive will be right for them. In this study, lemons with good quality and bad quality images were classified. 951 images for bad lemons and 1125 good quality images were used in the dataset. Convolutional Neural Networks (ResCNN) deep learning model and 5-fold cross validation process were performed in lemon classification process. As a result of the statistical measurements, the highest accuracy value was obtained as 0.8848 for both lemon varieties as the average value. While the highest sensitivity value was obtained from good quality lemon as 0.9013, the highest specificity value was obtained as 0.9013 from bad quality lemon variety.