In this paper we address the task of counting crop plants in a field using CNNs. The number of plants in an Unmanned Aerial Vehicle (UAV) image of the field is estimated using regression instead of classification. This avoids to need to know (or guess) the maximum expected number of plants. We also describe a method to extract images of sections or "plots" from an orthorectified image of the entire crop field. These images will be used for training and evaluation of the CNN. Our experiments show that we can obtain a Mean Absolute Percentage Error as low as 6.7 % with the Inception-v3 CNN architecture.