Abstract: Muskmelon has a wide planting area and manual picking is time-consuming and labor-intensive．Developing mechanized picking technology has been a hot topic in recent years, and target detection is one of the key links in mechanized picking．"Boyang 9" muskmelon images were used, and YOLOv5 and YOLOv7 models were used to study muskmelon target detection．The images of muskmelon fruits in different environments were collected for model training．Images sizes of input models were 640×640, number of samples per batch was 32, epochs were 100, and initial learning rate was 0.01．The precision（P）, recall（R）, F₁ score and average precision（AP）of YOLOv5 model were 86.1%, 82.5%, 84.3% and 90.2% respectively．The P, R, F₁ and AP of YOLOv7 model were 85.7%, 85.8%, 85.7% and 92.5% respectively．Detection accuracy of YOLOv7 model was high．Results showed that it was feasible to detect muskmelon based on YOLOv5 and YOLOv7, which solved problem of target detection in the early stage of muskmelon robot picking technology．