Classification of lime defects with convolutional neural network (CNN)

The defection of lime are been easily visible to classify by humans. But it required time-consuming and costly labor for the industry. Using computer vision is one of the approaches for determining the defection of lime since it’s less time-consuming and less expensive. The main purpose is to determine the defections of fresh market lime by utilizing color feature value, computer vision, and machine learning. In this project, using a convolutional neural network (CNN) classify lime into 2 classes; defection (1) and healthy (0) since the CNN model is deep learning and the performance of the result is very high accuracy. The 5 hyperparameters for tuning the CNN model— Epoch, Learning Rate, Decaying Rate, and Momentum—are taken into account and designed using factorial methods to improve the model's performance. For the training, validation, and test models, respectively, the average performance of results by f1-score is 96.30%, 91.79%, and 85.92%.