INTELLIGENT SYSTEM FOR AUTOMATIC CLASSIFICATION OF FRUIT DEFECT USING FASTER REGION-BASED CONVOLUTIONAL NEURAL NETWORK (FASTER R-CNN)

Abstract

In 2018, the Indonesian fruit exports increased by 24% from the previous year. The surge in demand for tropical fruits from non-tropical countries is one of the contributing factors for this trend. Some of these countries have strict quality requirements – the poor level quality control of fruit is an obstacle in achieving greater export yield. This is because some exporters still use manual sorting processes performed by workers, hence the quality standard varies depending on the individual perception of the workers. Therefore, we need an intelligent system that is capable of automatic sorting according to the standard set. In this research, we propose a system that can classify fruit defects automatically. Faster R-CNN (FRCNN) architecture proposed as a solution to detect the level of defect on the surface of the fruit. There are three types of fruit that we research, its mangoes (sweet fragrant), lime, and pitaya fruit. Each fruit divided into three categories (i) Super, (ii) middle, (iii) and fruit defects. We exploit join detection and video tracking to calculate and determine the quality fruit in real-time. The datasets are taken in the field, then trained using the FRCNN Framework using the Tensorflow platform. We demonstrated that this system can classify fruit with an accuracy level of 88% (mango), 83% (lime), and 99% (pitaya), with an average computation cost of 0.0131 m/s. We can track and calculate fruit sequentially without using additional sensors and check the defect rate on fruit using the video streaming camera more accurately and with greater ease.

Keywords

• Deep Learning
• Faster R-CNN
• Object Detection
• TensorFlow
• Tracking Detection