Implementation of Convolutional Neural Network Based on InceptionV3 to Classify Guava Quality

Guava, a significant commodity in the market, has traditionally relied on farmers to sort it manually. The advancement of technology, one of which is machine learning, has offered a promising solution to automate this task through image classification. This study, evaluating the effectiveness and performance of the Convolutional Neural Network method employing the InceptionV3 transfer learning architecture to classify guava quality, was conducted with meticulous attention to detail. The datasets, comprising 166 images categorized as Super Class, A Class, B Class, and Reject Class, were used to conduct experiments on the dataset split ratios and hyperparameter tuning to achieve optimal performance. In this study, using a 0.00005 learning rate, 200 epochs, and an 8:2 train-test split ratio, 15% of the training data was allocated for validation, resulting in machine-learning models with a testing accuracy of 94.11%. We believe that this study's results can help farmers automate guava fruit sorting.