So I'm currently using SAGA GIS 9.0.2 software and one of its tools that I'm exploring is Object Based Image Segmentation (OBIA). For my use, I'm exploring of SAGA GIS power to segment UAV imagery. The object i'm trying to identify is Seagrass species and its shoot density. I have found SAGA GIS quite useful the hurdle i'm trying to overcome is an indepth explantion of the parameters that I need to input in SAGA software for doing OBIA analysis. The only tutorials that I have found are limited explanation and the software version is 7.0 below. So the only parameters that are used in the old tutorials are which neighborhood is used and option to do post-processing (unsupervised classification).
I have found a specific forum which linked a paper on indepth explanation but with my limited knowledge, its explanation on the parameters isn't explained fully. After combing to many pages and forums in sourceforger, google scholar, youtube and such other places, there isn't a specific text that explains in depth for example what effect if my value Feature in Position Space is bigger. Does the space between each segmented area increase or decrease? and many such more questions.
I'm currently still experimenting but this research is for my undergradute thesis so it will be a great help if there is an in-depth explanation of each parameter and the effects its has with different values inputed. To my understanding, for my version the explanation is also very minimal for each parameter in which is include in the SAGA application.
Any help is much appreciated!
*here are twoscreenshot orthomosaic UAV image in SAGA GIS 9.0.2. The second picture is the parameters in which i'm confused and need explanation for.
you don't provide a link to the paper you've found, so I don't know were to start ... the topic is too broad to start from the beginning. Please also have a look at the provided references.
But I expect that you did some research in OBIA and especially in segmentation/region growing, so your question seems to boil down to the segmentation/region growing parameters.
Band Width for Seed Point Generation (radius in grid cells): controls the number of seed values created, the larger the number the less seeds/segments will be created
Neighbourhood: this describes the kernel used; with "Neumann" only the four cardinal neighbor cells of each cell are analyzed, with "Moore" also the diagonal ones (full 3x3 window)
Distance: the analysis can be performed in "feature and position space" or "only in feature space"; the latter only takes your input feature grids into account. With position, you also take the seed position into account, which puts a stronger constraint on the growing to stay in the neighborhood of the seed
Variance in Feature/Position Space: these are a weighting factors for the variances, so you can put more emphasis on feature or on position; this influences the similarity between pixels and thus which pixels form a region around the seed point
Similarity Threshold: if the similarity of a neighbor pixel is below this threshold, it will not be added to the region
Generalization: this is a radius (in grid cells) used for the majority filter applied on the segmentation result; can be used to remove speckle from the segmentation
Best regards,
Volker
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Also maybe any papers which boils down to logical steps and process of the OBIA workflow is greatly apperciated because in my understanding, I've found that OBIA tool in SAGA is more like a set of tools in a workflow. Each result of a a different tool is then used for the next and I want to understand how does it specifically work because I'm currently trying to understand the unsupervised classification and to see if it is possible if I can do another unsupervised classification without running another OBIA process because it takes a while.
Thanks for any such help or comments.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
the bandwidth parameter has to be provided in "number of grid cells" to the tool. I think they just converted it to meters in the publication in order to make it easier to read/understand.
You can easily perform all steps reproduced in the tool chain on your own. Just have a look at the tool chain and the tool and parameter settings it uses. Then execute these tools one after the other on your own. This way you can also store intermediate results and use them for a modified workflow. To have a look at the tool chain, simply do a right-click on it in the "Tools" tab and then select "Edit" (or have a look here).
Best regards,
Volker
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
So I'm currently using SAGA GIS 9.0.2 software and one of its tools that I'm exploring is Object Based Image Segmentation (OBIA). For my use, I'm exploring of SAGA GIS power to segment UAV imagery. The object i'm trying to identify is Seagrass species and its shoot density. I have found SAGA GIS quite useful the hurdle i'm trying to overcome is an indepth explantion of the parameters that I need to input in SAGA software for doing OBIA analysis. The only tutorials that I have found are limited explanation and the software version is 7.0 below. So the only parameters that are used in the old tutorials are which neighborhood is used and option to do post-processing (unsupervised classification).
I have found a specific forum which linked a paper on indepth explanation but with my limited knowledge, its explanation on the parameters isn't explained fully. After combing to many pages and forums in sourceforger, google scholar, youtube and such other places, there isn't a specific text that explains in depth for example what effect if my value Feature in Position Space is bigger. Does the space between each segmented area increase or decrease? and many such more questions.
I'm currently still experimenting but this research is for my undergradute thesis so it will be a great help if there is an in-depth explanation of each parameter and the effects its has with different values inputed. To my understanding, for my version the explanation is also very minimal for each parameter in which is include in the SAGA application.
Any help is much appreciated!
*here are twoscreenshot orthomosaic UAV image in SAGA GIS 9.0.2. The second picture is the parameters in which i'm confused and need explanation for.
Hi,
you don't provide a link to the paper you've found, so I don't know were to start ... the topic is too broad to start from the beginning. Please also have a look at the provided references.
But I expect that you did some research in OBIA and especially in segmentation/region growing, so your question seems to boil down to the segmentation/region growing parameters.
Band Width for Seed Point Generation (radius in grid cells): controls the number of seed values created, the larger the number the less seeds/segments will be created
Neighbourhood: this describes the kernel used; with "Neumann" only the four cardinal neighbor cells of each cell are analyzed, with "Moore" also the diagonal ones (full 3x3 window)
Distance: the analysis can be performed in "feature and position space" or "only in feature space"; the latter only takes your input feature grids into account. With position, you also take the seed position into account, which puts a stronger constraint on the growing to stay in the neighborhood of the seed
Variance in Feature/Position Space: these are a weighting factors for the variances, so you can put more emphasis on feature or on position; this influences the similarity between pixels and thus which pixels form a region around the seed point
Similarity Threshold: if the similarity of a neighbor pixel is below this threshold, it will not be added to the region
Generalization: this is a radius (in grid cells) used for the majority filter applied on the segmentation result; can be used to remove speckle from the segmentation
Best regards,
Volker
Hello Volker, thanks for the answer. So the refrences that I've found about Seeded Region Growing and OBIA parameters are https://mediatum.ub.tum.de/doc/1304679/document.pdf and https://www.geo.uni-hamburg.de/geographie/dokumente/personen/publikationen/bechtel/bechtel_et_al_2008.pdf. The second link is the one that shows up when I click the refrenced paper in the help section of the OBIA parameters in SAGA, basically the provided refrences.
Sorry for asking again, but there are papers which explains the units in the parameters for OBIA is in meters. Here is the link which explained that band width unit used is in meters (https://www.researchgate.net/profile/Tommaso-Sarzana/publication/344445457_Post-processing_of_Pixel_and_Object-Based_Land_Cover_Classifications_of_Very_High_Spatial_Resolution_Images/links/5f97da64a6fdccfd7b826ea4/Post-processing-of-Pixel-and-Object-Based-Land-Cover-Classifications-of-Very-High-Spatial-Resolution-Images.pdf. I thought the units used was per pixel in the image. I'm a bit confused so maybe any help with the explanation in detail would be great.
Also maybe any papers which boils down to logical steps and process of the OBIA workflow is greatly apperciated because in my understanding, I've found that OBIA tool in SAGA is more like a set of tools in a workflow. Each result of a a different tool is then used for the next and I want to understand how does it specifically work because I'm currently trying to understand the unsupervised classification and to see if it is possible if I can do another unsupervised classification without running another OBIA process because it takes a while.
Thanks for any such help or comments.
Hi,
the bandwidth parameter has to be provided in "number of grid cells" to the tool. I think they just converted it to meters in the publication in order to make it easier to read/understand.
You can easily perform all steps reproduced in the tool chain on your own. Just have a look at the tool chain and the tool and parameter settings it uses. Then execute these tools one after the other on your own. This way you can also store intermediate results and use them for a modified workflow. To have a look at the tool chain, simply do a right-click on it in the "Tools" tab and then select "Edit" (or have a look here).
Best regards,
Volker