Creation of Aerial Arecanut Dataset and Aerial Arecanut Ripeness Level Detection using Deep Neural Networks

Background: Arecanut, a commercially significant crop grown across various regions of the country and contributes significantly to India’s global ranking as the second-largest producer. Harvesting arecanut at precise stages is paramount to achieve optimal yields. Typically, this task requires minimum of two skilled individuals: a professional tree climber proficient in nut plucking and another adept at assessing ripeness. Before proceeding to harvest all the arecanuts, the climber picks one or two nuts and indicates the other person to check it for ripeness assessment. The assessor verifies ripeness by peeling the nut with teeth and uses roti/dhoti (an equipment used to harvest arecanut) to harvest arecanut if it is ripened. Inaccurate predictions can lead to significant crop losses. Manual harvesting, while effective, presents challenges such as labour shortages, time consumption, and potential for life-threatening circumstances. Methods: To address these challenges this paper presents the creation of aerial arecanut dataset and accurate ripeness detection of arecanut using Deep Neural Networks (DNN) such as AlexNet and VGG-16. Images of Arecanut bunch are fed as input to the DNN. The features like colour, shape, texture, etc., will be extracted and given as input features to the classifier. The classifier will categorize the arecanut into three classes: unripened, intermediate and ripened. Result: The accuracy of classification of arecanut ripeness level is 94.58% and 96.87% using AlexNet and VGG-16 respectively. If this model is integrated in roti/dhoti, the life of the human being can be saved. Or if this model is integrated with the climber unit and cutter unit, it will transform the arecanut harvesting system into a fully automated harvesting system and further it reduces the man power required to harvest the arecanut, overcomes the life threats during harvesting and arecanut harvesting process becomes faster compared to the conventional method.