Branch Identification and Junction Points Location for Apple Trees Based on Deep Learning

Branch identification is key to the robotic pruning system for apple trees. High accuracy and positioning of junction points between branch trunk are important prerequisites with a arm. Recently, development deep learning, Transformer has been gradually applied field computer vision achieved good results. However, effect based on not verified so far. Taking Swin-T Resnet50 as backbone, this study detected segmented trunk, primary support trees basis Mask R-CNN Cascade R-CNN. The results show that, when Intersection over Union (IoU) 0.5, bbox mAP segm highest, which 0.943 0.940; each category identification, shows no significant difference other three algorithms in branch; identified object support, AP significantly higher than that algorithms, 0.879 0.893. Next, SW-T combined Zhang & Suen obtain point. Compared direct application algorithm, skeleton obtained by method advantaged diameter information, its shape position closer actual This model can be follow-up research offer new solution