Automatic detection of Ground Control Point (GCPs) is important in UAV surveying applications for accurate measurements and building reliable information. However, this is a challenging problem, and typically existing methods require fine tuning, lots of heuristics, and manual annotation. Cutting-edge approaches based on Deep Learning (DL) models have achieved promising results on GCPs detection. This research proposes a YOLOv8-based DL model for multi-scale GCPs detection and an adequate performance assessment to validate the reliability of the model. The proposed method starts by considering images with different size targets (The GCPs for the analysis) acquired by means of a DJI Phantom 3 Standard Drone at multiple scales or flight altitudes. Care was taken to ensure that at least 7 of the 12 targets were visible in each image. Then, a dataset of 60 annotated images was created using the Labelling Annotation Tool for training purposes. The model was trained over 50 epochs with a batch size of 4 and a resize image of 1920 x 1920 pixels based on the available GPU resources. Leveraging the lightweight YOLOv8s architecture, transfer learning from pretrained weights was applied, achieving a high precision of 0.997 and a recall of 0.993 on the validation set. Additionally, the model demonstrated a mean average precision (mAP50-95) of 0.887. The proposed model proved to be robust to the mis-detection of certain GCPs, since it is able to work with at least 7 Targets.