Paper Title
Real Time Fruit Recognition With CNN For Vision-Based Robot Arm

Abstract
Automatic recognition of fruits via computer vision is still a complicated and difficult task due to the various properties of numerous types of fruits. Robotic systems using a monocular camera as their input device have much more challenges as the result must be real-timed and accurate. The paper proposes the recognition of fruits (banana, carrot, cucumber and pear) among 16 objects using a mono vision camera. The real-time input image from a camera is firstly segmented with color-based segmentation and then the morphological processing is used to aid the segmentation process. The segmented objects are then classified with CNN (Convolutional Neural Network) based trained network to determine whether the desire objects are detected or not. This method is particularly suited for the environment where the background is relatively clear. The main purpose of the system is to give the simultaneous and accurate data to the monocular voice-controlled domestic robot arm. Index Terms - Monocular Camera, Color-Based Segmentation, Morphological Processing, Convolutional Neural Network