This paper demonstrates how simulated images can be useful for object detection tasks in the agricultural sector, where labeled data can be scarce and costly to collect. We consider training on mixed datasets with real and simulated data for strawberry detection in real images. Our results show that using the real dataset augmented by the simulated dataset resulted in slightly higher accuracy.