Detecting Apple Trees Based on Improved YOLOv3-Tiny

Abstract. With a shortage of skilled labor and rising labor costs, many researchers are studying robotics for apple production. Apple tree detection in orchard environment is the first key step of robot work. In this study, an object detection method based on deep learning were developed to accomplish this task. According to the characteristics of orchard environment, Inception v2 structure was introduced to improve the algorithm of YOLOv3-Tiny. The performance of the algorithm was verified by field tests. The precision, F1-score, recall and average precision are used as indicators to evaluate the algorithm, and compared with other object detection networks. This paper provides technical support for apple picking robots.