Implementation of Convolutional Neural Network (CNN) Method in Determining the Level of Ripeness of Mango Fruit Based on Image

This study aims to classify the ripeness level of mango fruit using a Convolutional Neural Network (CNN) model based on digital images. This classification is important to help the automatic sorting process in the agricultural industry that relies on accuracy in determining fruit quality. Based on the literature review, CNN has been widely used in image-based object recognition because of its ability to extract visual features automatically. Previous studies have shown that CNN is effective in image classification, but the results are highly dependent on the quality of the data and the model parameters used. This research method involves collecting mango fruit images at three levels of ripeness (raw, half-ripe, ripe), which are then processed and analyzed using the Orange application with CNN architecture. Model evaluation was carried out using accuracy metrics, AUC, confusion matrix, and visualization through box plots and scatter plots to see the distribution and differences in data between classes. The results showed that the CNN model obtained an accuracy of 53.3% and an AUC value of 0.717, which indicates the model's initial ability to distinguish ripeness categories but with a fairly high level of misclassification. There is still overlapping data between classes, especially between the raw and half-ripe classes, which indicates the need for additional features and parameter refinement. In conclusion, CNN has the potential to be used in classifying the ripeness level of mango fruit, but its performance can be improved through feature development and deeper model tuning.