I need to classify land covers from 1950s aerial photos in US. I can distinguish water, forest, agriculture and developed areas visually but I did not get good result from supervised classification, iso cluster or support vector machine with ArcGIS Pro. For example, both forest and water look dark and classes mixing, so are agriculture and buildings and roads. Is there a way to handle this problem? Thank you!
I work once with these aerial images, at the end the best way was to enhance the images to create contrast before classify them.
And the supervised way works better
I hope this will help you
Object-oriented classification was satisfactory in our case (Dai et 2017). Maybe proceeded in your case by supervised classification/exclusion of water bodies and settlements as provided on topographic maps.
Yes, I tried both object oriented and spectral approach together in ArcGIS Pro but I am not sure if I did it correctly. Now, I will adjust brightness, contrast etc. in photoshop, and remove water bodies then I will classify. Thank you!
I used Adobe Photoshop to merge images and increase contrast. I had to use partial contrasting/brightness etc. because center part was brighter than edges. For water I replace the water bodies with NLCD 2011 water coverage because in my case it did not make much difference. Finally, I used unsupervised classification that accounts also object orientation to produce forest/wetland, developed and crop land covers. Unfortunately, even after all these processes I need to do some manual classification. Thank you Ivan Lizaga and Hein Van Gils for your help and guidance.
Dear Oguz,
with my reasearch group we also classified panchromatic (or monochrome) photographs in many works, mostly in mountain areas. You already received very good advices from Ivan Lizaga and Hein Van Gills.
I can add that one thing that really helped us was to create shadow maps using sun irradiation models taking into account obstacles (this can be done if you know the hour and the date of each photograph).
for example: r.sun - Solar irradiance and irradiation model in GRASS GIS.
I suppose something similar exists for ARGIS.
In this way we divided each orthophotograph in different areas that had a different shadow level.
Then we repeated the classification in each area and the results where better than the classifications carried out on the whole orthophoto.
Anyway, the quality of the original photographs is crucial, because if the contrast or the light conditions or sometimes the season when the photographs were taken make them unclear, the results from automatic classification is generally poor.
All the best,
Marco
I did not have shadow issue specifically but whole image was brighter on the center and dark at the edges. However, I did have shadow problems in the past. I wish I knew solar irradience and irradiation model before. Marco Ciolli, thank you for your answer.
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity