Potato Visual Navigation Line Detection Based on Deep Learning and Feature Midpoint Adaptation

Potato machinery has become more intelligent thanks to advancements in autonomous navigation technology. The effect of crop row segmentation directly affects the subsequent extraction work, which is an important part line detection. However, shape differences crops different growth periods often lead poor image segmentation. In addition, noise such as field weeds and light also affect it, these problems are difficult address using traditional threshold methods. To this end, paper proposes end-to-end potato detection method. first step replace original U-Net’s backbone feature structure with VGG16 segment rows. Secondly, a fitting method midpoint adaptation proposed, can realize adaptive adjustment vision position according potato. results show that used strong robustness accurately detect lines periods. Furthermore, compared U-Net model, accuracy improved by 3%, average deviation fitted 2.16°, superior visual guidance