Faster RCNN implementation over a custom classification CNN

Hello, I have some basic problem with implementing faster RCNN. I have a custom made binary classification CNN for trees, this network is trained with images of only one tree/no tree.(all my training images has only single tree image or no tree at all) How can I use this network for tree detection from bigger images using faster RCNN. In other words, how can I implement faster RCNN over my trained classification net?? is it possible ?? or do I need to make the faster RCNN from scratch and this classification net is of no use to me??

Thanks