The global blueberry market has been expanding vastly driven by consumer demand for healthier food. As a top blueberry producer, United States generated a revenue of $932 million in 2020. A profitable blueberry industry relies on continued cultivar improvement. One challenge faced by blueberry breeders, researchers, and growers, is yield data collection. Measuring blueberry yield by manual sampling is labor-intensive and time-consuming. We developed a smartphone application leveraging deep learning techniques to automate yield prediction and maturity assessment for different blueberry cultivars under field conditions. State of the art YOLOv8 models were fine-tuned and evaluated using a dataset of side-view images of various southern highbush and rabbiteye blueberry cultivars. The best performing DL model of YOLOv8-x achieved a mean average precision of 0.708 and 0.372 under 0.5 and 0.5-0.95 Intersection over Union thresholds on validation datasets, respectively. Blueberry yield was predicted using non-linear regression-based machine learning models using the image-derived mature berry count multiplied by user-defined average berry weight and cultivar as explanatory variables with satisfactory accuracy. This proposed smartphone app can enable image-based yield prediction for blueberry growers and breeders, which is valuable for management decision making and accelerated selection for high-yielding cultivars.