Deep learning opens new horizons in agriculture, offering opportunities for increased efficiency and economic growth. A particularly promising direction is the application of deep learning in accurate fruit detection and segmentation. However, there are certain challenges in complex orchard environments such as vineyards. The aim of this study is to provide fast and accurate detection and segmentation of different types of grapes in complex orchard environments. To provide a clearer understanding of the model structure, we propose an improved Mask R-CNN architecture. The model outline highlights the dual-path input approach, feature extraction enhancements, and specialized branches, which together contribute to the superior performance of the model in object detection and segmentation tasks. The results largely proved that the proposed model provides high accuracy. It was possible to achieve an accuracy above 98% using the proposed network.