Minimize false positive results doing some image processing in an object detection model

Question

I am using Mask-RCNN to solve an object detection problem. This is an implementation of Mask R-CNN on Python 3, cv2, Keras, and TensorFlow. I am trying to identify the damag