Citrus fruits are crucial in the agricultural sector, and accurate yield estimation is essential for orchard management. However, traditional methods often struggle with severe fruit occlusion due to dense foliage or overlapping fruits. This study introduces AG-YOLO, an attention-based network designed to address these challenges. AG-YOLO leverages NextViT as its primary architecture, which captures holistic contextual information within nearby scenes. Additionally, it incorporates a Global Context Fusion Module (GCFM) to facilitate the interaction and fusion of local and global features through self-attention mechanisms, significantly improving the model's ability to detect occluded targets. An independent dataset of over 8000 outdoor images was collected, including instances of occlusion, severe occlusion, overlap, and severe overlap. AG-YOLO demonstrated exceptional performance, achieving a precision (P) of 90.6%, a mean average precision (mAP)@50 of 83.2%, and an mAP@50:95 of 60.3%. These metrics surpass existing mainstream object detection methods. AG-YOLO also achieved a detection speed of 34.22 frames per second (FPS), maintaining a high level of detection accuracy. The study highlights the effectiveness of AG-YOLO in handling severe occlusions, making it an efficient and reliable solution for object detection in agricultural settings.Citrus fruits are crucial in the agricultural sector, and accurate yield estimation is essential for orchard management. However, traditional methods often struggle with severe fruit occlusion due to dense foliage or overlapping fruits. This study introduces AG-YOLO, an attention-based network designed to address these challenges. AG-YOLO leverages NextViT as its primary architecture, which captures holistic contextual information within nearby scenes. Additionally, it incorporates a Global Context Fusion Module (GCFM) to facilitate the interaction and fusion of local and global features through self-attention mechanisms, significantly improving the model's ability to detect occluded targets. An independent dataset of over 8000 outdoor images was collected, including instances of occlusion, severe occlusion, overlap, and severe overlap. AG-YOLO demonstrated exceptional performance, achieving a precision (P) of 90.6%, a mean average precision (mAP)@50 of 83.2%, and an mAP@50:95 of 60.3%. These metrics surpass existing mainstream object detection methods. AG-YOLO also achieved a detection speed of 34.22 frames per second (FPS), maintaining a high level of detection accuracy. The study highlights the effectiveness of AG-YOLO in handling severe occlusions, making it an efficient and reliable solution for object detection in agricultural settings.