Machine vision algorithm for detection and maturity prediction of Brinjal

Abstract

One of the biggest issues faced by farmers is the lack of labor for harvesting vegetables. Harvesting brinjals requires a deep understanding and intuitive expertise. Transferring the knowledge of human involved methods to visual system of the harvesting robots is difficult due to the complex background and the nature of the plant bearing multiple brinjals on a single branch. And it is observed that all the brinjals belong to one cluster may not have the same maturity level, this add complexity to identify the moderately matured brinjals through a computer vision. This study proposes an algorithm to detect and identify the moderately matured KKM-1 brinjals. The images are taken from the actual fields through the mobile camera and are considered as the basic data for this study. The algorithm involves two stages, (i) detecting the brinjal and (ii) predicting the moderately matured brinjal. The size of the brinjal determines its maturity and the size prediction is found to be influenced by the shades that is caused due to many reasons. In this study, the k-means clustering algorithm is employed to (i) identify the shaded region of the brinjal and (ii) segment the area of interest. The brinjals are detected from the segmented image using contour method and the non-brinjals are removed based on the shape. A threshold value is calculated using median function that takes the sample area from brinjal dataset. It is observed that the suitability to harvest is more when the area of the brinjal exceeds the threshold. An extensive experimental study is conducted on the collected dataset. During the detection phase, the proposed technique achieved a precision of 79% and an F1-Score of 85%. During the detection phase and in the moderately maturity prediction process, the precision is 96% and the F1-Score is 91%.