Pomegranates from Mengzi City, Yunnan Province, are renowned for their nutritional value and flavor, boosting the local economy. Traditional yield prediction relies on manual sampling, which is costly and inefficient. This study proposes an intelligent prediction method using the YOLOv5 deep learning model.Mobile phones captured images of pomegranate trees from four directions to construct a dataset. These images were cropped and labeled for training the YOLOv5 model. The model achieved an 89% recognition accuracy and a 0.94 effectiveness score, accurately identifying and counting pomegranate fruits.Data visualization and model inference estimated the fruit quantity, with predicted values being 91.7% of the actual yield at a confidence level above 0.25. The average yield per tree was estimated at 103.1 kg, closely matching the actual 105.9 kg. This method offers robust data support for rapid and accurate yield estimation, outperforming traditional methods in efficiency and accuracy.