Semi-Automated Image Annotation for Cannabis Seed Gender Detection Model

Abstract

Many areas such as industrial, academic and medical have turned attention to the value and market opportunities that use cannabis as the new mixture. Due to the potential of its compounds, cannabis has become an economically viable plant. Many growers in Thailand are turning to cannabis as one of the yield crops and probably an alternative economically crop. The growers generally buy regular seeds for planting outdoor, resulting in the mix of growing up to male and female plants. The fact that only female cannabis plants are preferred as they provide valuable cannabinoid-rich flowers to harvest. Therefore, automated screening the gender of cannabis seeds upfront prior to plant could help lower waste and increase profitability. Agri-technology such as computer vision has played an important role in agriculture. Object detection models would facilitate sorting out female from male seeds. However, dataset annotation is resource consumption. This paper thus presents an approach of semi-automated image annotation as part of the YOLOv5 object detection model for gender screening of cannabis seeds. The preliminary result is promising. Compared to the model using manual annotation, the model using the proposed semi-automated annotation achieved comparable performance, and reduced half of the time spent on dataset annotation.