Orchard target detection is one of the key prerequisites for realizing agricultural automation. With the continuous development of agricultural science and technology, agricultural production is gradually moving toward intelligence and automation. As an important part of the agricultural field, the management and production of orchards must also use advanced technology to improve efficiency and quality. Deep learning models utilizing instance segmentation, such as Mask R-CNN and YOLACT, can not only detect the trees in the orchard but also segment the safe range for vehicles to operate, realizing accurate identification and localization of the orchard environment, as well as guaranteeing the safety of vehicles in the process of operation. This provides an important database and technical support for agricultural automation. By automatically detecting trees and roads in the orchard, operations such as automatic inspection of the orchard, monitoring of pests and diseases, and fruit picking can be realized, thus reducing the burden of manual labor and improving production efficiency.