<![CDATA[Apple variety classification based on digital images is a challenging problem because some varieties have similar visual characteristics, such as color, shape, and surface texture of the fruit. This similarity can lead to misidentification when performed manually. This study aims to develop an apple variety classification model using the MobileNetV2 architecture with a transfer learning and fine-tuning approach. The dataset used consists of 1,237 apple images representing four varieties: Anna, Manalagi, Rome Beauty, and Granny Smith. After data cleaning to remove duplicate images, 1,050 images were obtained for use in the study. The pre-processing stage includes image resizing to 224 × 224 pixels, normalization, and texture enhancement using a combination of Unsharp Masking and the Sobel operator. To improve the model's generalization capability, data augmentation and class weighting were applied during the training process. The model was then evaluated using a confusion matrix, accuracy, precision, recall, and F1-score. The test results showed that the MobileNetV2 model achieved an accuracy of 99.37%, a precision of 99.48%, a recall of 99.07%, and an F1-score of 99.27%. Confusion matrix analysis showed that out of 158 test data, there was only one misclassification, namely the Rome Beauty image predicted as Anna. In addition, the model was successfully implemented into a web-based application using Streamlit so that it can be used to directly identify apple varieties. The results showed that MobileNetV2 is effective for classifying apple varieties with a high level of accuracy and good computational efficiency, so it has the potential to be applied to image-based fruit identification systems in agriculture.]]>
