Strawberry Fruit Quality Assessment for Harvesting Robot using SSD Convolutional Neural Network

Abstract

Strawberry has a tremendous economic value as well as being visually appealing. Therefore, strawberry farmers need to ensure that they only harvest good quality strawberries. However, assessing the quality of strawberries is not an easy problem, especially for local plantations which do not have enough human resources. As robotics becomes accessible and widely used for agriculture work such as harvesting fruit, the real-time embedded system computation power becomes much more powerful nowadays. This paper discusses the harvesting robot's ability to distinguish the quality of strawberries in realtime detection using computer vision technology in the form of object detection by utilizing a deep neural network in a single board computer (SBC). The robot software is built on Robot Operating System (ROS) framework. The proposed method is tested on a robot equipped with a monocular camera. The learning process shows that the robot can detect and differentiate between good and bad quality strawberries with 90% accuracy and maintain a high frame rate.