<![CDATA[This research aims to develop an automated sorting system for post-harvest oranges to address subjectivity and the risk of errors caused by human fatigue. The method employed involves developing an automated sorting conveyor prototype based on an ESP32 microcontroller integrated with deep learning technology via a webcam. The continuous ripeness classification process utilizes a Convolutional Neural Network (CNN) with the MobileNetV2 architecture to categorize the fruit into ripe and half-ripe stages. The detection results are then transmitted serially to the ESP32 to drive the sorting servo in real-time. Based on testing with 50 orange samples, the MobileNetV2 CNN model achieved an accuracy of 90% with an F1-Score of 0.90. The implementation of this system is proven to significantly increase post-harvest productivity by reducing processing time by 63.3% compared to manual methods. Additionally, the data counter feature for the Ripe and Half-Ripe categories operates simultaneously to standardize production output monitoring. This study demonstrates that the integration of MobileNetV2 and ESP32 is effectively applied to a prototype-scale sorting system.]]>
Copyrights © 2026
