I am trying for automatic digitization of line features from imagery using deep learning in ArcGIS 2.9. After generating the buffer polygon from the line features, I used both imagery and this buffer polygon layer for Export Training Data for Deep Learning to train my model. The buffer width is 60m. After exporting, I trained a deep learning model using exported training samples using the tool Train Deep Learning Model. Then I applied the generated model to the same imagery by using Classify Pixels using Deep Learning. The result is good, however, I am getting overlapping with near features and the output width of the line features in the output imagery is nearly 300m. Please check the snapshots for understanding. My question is how to avoid overlapping and how to reduce the output width of the line feature in the output imagery. Your solution would be really appreciated. Thanks.
