Machine learning with high-resolution aerial imagery and data fusion to improve and automate the detection of wetlands

•There is ecological interest in monitoring wetland loss, even small ones.•USA's wetland inventories are outdated and omit small but relevant wetlands.•A new method, WetSegNet, to segment wetlands in aerial imagery is presented.•We developed a new wetland dataset using images from NAIP and NHS labels.•The accuracy of WetSegNet (based on area under the ROC curve) reached 98%.