Development of a Robotic Arm Using a Convolutional Neural Network for Areca Nut Sorting

Abstract

Manual post-harvest sorting of areca nuts is time-consuming and may result in inconsistent classification. To address this limitation, this study developed an automated sorting system that integrates a Convolutional Neural Network (CNN) with a robotic arm to classify and transfer areca nuts into the juice and split categories. The system comprises webcam-based image acquisition, image processing and classification using Python and the MobileNetV3 architecture, and a microcontroller-based control system for a 4 degree of freedom (4 DOF) robotic arm. The CNN model was trained to classify three categories: juice areca nut, split areca nut, and surrounding objects. Based on the reported confusion matrix, the model correctly classified 188 of 189 validation images, achieving an accuracy of 99.47%. The image processing evaluation also reported accuracy values of 99.07% for color, 91.65% for shape, and 94.62% for dimensional measurement. Integrated system testing was conducted on 30 areca nuts, comprising 15 juice type and 15 split type nuts. All 30 objects were successfully transferred to their designated bins, resulting in a sorting success rate of 100% under the tested conditions. The reported CNN inference time was 0.032 s, while the average robotic arm movement time was 15 s per cycle, yielding an estimated total sorting time of 15.032 s per object. These findings demonstrate the feasibility of integrating CNN based image classification with robotic manipulation for automated areca nut sorting. Further testing involving more diverse samples and operating conditions is required to assess the system's generalizability and operational reliability.