The estimation of crop phenological distribution is of great importance for controlling time of thinning flowers. In order to improve the efficiency of flower thinning in modern orchard, a detection method of apple flower phenological distribution based on YOLO-CG network model is proposed to detect, which aims at improving incomprehensive and low-efficient manual traditional detection method of apple flower phenological distribution. First of all, the YOLO-CG network model is to integrate the CA mechanism into the YOLOv5 network, which could obtain more shallow features to improve network performance; Secondly, in order to improve the training speed to reduce the calculation amount of the network model, the Ghost-Bottleneck module is proposed to replace the Bottleneck module; Finally, the CIOU is used as the bounding box regression loss function to improve the stability of the target box regression. The model is fine-tuned and trained with manually-marked apple flower images in 4 phenological stages. The proposed method was compared with the detection models of YOLOv3, YOLO v4, YOLO v5 and Faster R-CNN, and the detection performance of apple flower under different shooting conditions are discussed, which proves the effectiveness of this method. Experimental results show that the mAP value of apple flower detection at different stages was 94.90%, an increase of 1.98%, 7.1%, 5.42% and 2.53% respectively compared with Faster R-CNN, YOLO v3, YOLO v4 and YOLO v5.