A Two-Step Deep Semantic Segmentation and Object Detection Approach for Runner Recognition in Strawberry Plants

Abstract

Strawberry production in California, the leading producer in the United States, relies on effective field treatments such as cutting runners (stolon) to enhance crop productivity. Manual runner cutting is laborious and costly, motivating the exploration of robotic solutions. This paper presents a deep learning-based approach for runner detection in strawberry plants using RGB images. A two-step framework is proposed, involving a segmentation network to create runner masks and an object detection network to recognize and localize runners. The methodology was evaluated on RGB images of three strawberry cultivars (Monterey, Cabrillo, and Albion), and the results demonstrated the effectiveness of the deep-learning models. The Albion cultivar showed the highest accuracy, emphasizing the influence of cultivar and imaging settings on model performance. Further research is recommended to investigate the relationship between imaging settings and model performance and explore avenues for improvement. The findings contribute to the development of an autonomous runner-cutting robot, offering potential labor-saving technology for strawberry growers.

Keywords

• Strawberry
• runner detection
• deep learning
• autonomous runner-cutting robot
• crop productivity