Computer Vision and Machine Learning Based Grape Fruit Cluster Detection and Yield Estimation Robot

Abstract

Estimation and detection of fruits plays a crucial role in harvesting. Traditionally, fruit growers rely on manual methods but nowadays they are facing problems of rapidly increasing labor costs and labour shortage. Earlier various techniques were developed using hyper spectral cameras, 3D images, clour based segmentation where it was difficult to find and distinguish grape bunches. In this research computer vision based novel approach is implemented using Open Source Computer Vision Library (OpenCV) and Random Forest machine learning algorithm for counting, detecting and segmentation of blue grape bunches. Here, fruit object segmentation is based on a binary threshold and Otsu method. For training and testing, classification based on pixel intensities were taken by a single image related to grape and non-grape fruit. The validation of developed technique represented by random forest algorithm achieved a good result with an accuracy score of 97.5% and F1-Score of 90.7% as compared to Support Vector Machine (SVM). The presented research pipeline for grape fruit bunch detection with noise removal, training, segmentation and classification techniques exhibit improved accuracy.