Fruit Grading based on Deep Learning and Active Vision System

Abstract

This paper presents a low-cost computer vision-based solution to obtain the size of fruits without contact. It consists of a low-cost webcam and a cross-shaped laser beam rigidly assembled. The proposed approach acquires and processes the images in real-time. Due to the low computational cost of the proposed algorithm, a robust solution is obtained using a frame redundancy approach, which consists in processing several frames of the same scene and hence computing a robust estimation of the fruit size. The proposed solution is evaluated with different tropical fruits (e.g., banana, avocado, dragon fruit, mamey, papaya, and taxo). Obtained results show on mean average percentage error (MAPE) below 1.50% in the computed sizes.