Segmentation and Classification of Arecanut Bunches before harvesting

Abstract

In the agriculture sector, arecanuts are an extremely valuable crop. The price of an arecanut depends on its stage of ripeness. As a result of a lack of expertise in judging the maturity level of arecanut bunches before harvest, farmers often lose profit. Precision agricultural techniques based on image processing and computer vision have recently assisted farmers in determining crop maturity quality.  Precision agricultural techniques based on image processing and computer vision have recently assisted farmers in determining crop maturity quality. Therefore, accuracy in segmenting arecanut bunches is vital for automated maturity level identification. In proposed work S-channel, Cr-channel and Pr-channel of HSV, YCbCr and YPbPr respectively color models are used to segment arecanut bunches. Three color features (i.e., mean of an arecanut bunch image on red, green, and blue bands), and two texture features (i.e, correlation, and entropy) were used in classification procedure. A random forest classifier was employed to classify maturity levels of arecanut bunch. This experiment uses a dataset of 1017 images of arecanut bunches to assess the segmentation performance of each color model. As a result of the experiment, it has been concluded that the S-channel of the HSV color model was effective in segmenting arecanut bunches from input images. The proposed methodology effectively classifies arecanut bunch maturity levels with an accuracy of 87.80%.