The improved YOLO v8-Pose model was established to identify red ripe strawberries and detect the key points of the stem in greenhouse strawberries under elevated cultivation mode. By comparing the YOLO v5-Pose, YOLO v7-Pose and YOLO v8-Pose models, the YOLO v8-Pose model was determined to be used as the model to identify and predict the key points of red ripe strawberries. Based on YOLO v8-Pose, Slim-neck module and CBAM attention mechanism module were added to its network structure to improve the feature extraction ability of the model for small target objects, so as to adapt to the characteristics of strawberry data set. The P, R and mAP-kp of the improved YOLO v8-Pose were 98.14%, 94.54% and 97.91%, respectively, which can effectively detect red ripe strawberries and accurately mark the key points of the fruit stalk, which was 5.41, 5.31 and 8.29 percentage points higher than that of YOLO v8-Pose. The model memory footprint was 22MB, which was 6MB less than that of the YOLO v8-Pose footprint.In addition, according to the unstructured characteristics of the orchard, the influence of light, occlusion and shooting angle on the model prediction was explored. Compared with the recognition and stem prediction of the improved YOLO v8-Pose model in the complex environment, the mAP-kp of the improved YOLO v8-Pose under the influence of occlusion, light and angle was 94.52%, 95.48% and 94.63%, respectively. Compared with YOLO v8-Pose, it was 8.9, 10.75 and 5.17 percentage points higher, respectively. The improved YOLO v8-Pose can ensure the accuracy of the network model, and at the same time, it had good robustness to the effects of occlusion, light and shooting angle, etc., which can realize the identification of red ripe strawberries in complex environments and the prediction of key points of fruit stalk.