Arecanut Bunch Segmentation Using Deep Learning Techniques

Abstract

Agriculture and farming as a backbone of many developing countries provides food safety and security. Arecanut being a major plantation in India, take part an important role in the life of the farmers. Arecanut growth monitoring and harvesting needs skilled labors and it is very risky since the arecanut trees are very thin and tall. A vision-based system for agriculture and farming gains popularity in the recent years. Segmentation is a fundamental task in any vision-based system. A very few attempts been made for the segmentation of arecanut bunch and are based on hand-crafted features with limited performance. The aim of our research is to propose and develop an efficient and accurate technique for the segmentation of arecanut bunches by eliminating unwanted background information. This paper presents two deep-learning approaches: Mask Region-Based Convolutional Neural Network (Mask R-CNN) and U-Net for the segmentation of arecanut bunches from the tree images without any pre-processing. Experiments were done to estimate and evaluate the performances of both the methods and shows that Mask R-CNN performs better compared to U-Net and methods that apply segmentation on other commodities as there were no bench marks for the arecanut.