search

natesawant / tree-biomass

0 stars 0 forks source link

Investigate potential new tools like DeepForest and samgeo #5

Closed audreyromanik closed 3 months ago

audreyromanik commented 3 months ago

PointcloudITD- “Software tool for rapidly counting individual trees using cloud point data” chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://fwpa.com.au/wp-content/uploads/2017/12/pointcloudITD_information.pdf https://slideplayer.com/slide/13664245/

Pros: Uses lidar data to create a classification model of trees, then takes the classification model and the lidar data to identify trees Uses SVM to make the classification model Cons: Collection of lidar Finding the GUI and software???

-- currently waiting for the original research paper on the individual tree detection(Itd) algorithm

audreyromanik commented 3 months ago

DetecTree- opensource python package https://www.youtube.com/watch?v=USwF2KyxVjY&ab_channel=AppliedMachineLearningDays https://pypi.org/project/detectree/0.3.1/#:~:text=DetecTree%20is%20a%20Pythonic%20library,methods%20of%20Yang%20et%20al. https://github.com/martibosch/detectree

Takes an aerial image, splits it into tiles, then splits the pixels into part of a tree or not and returns those pixel values(binary pixel identification) Cannot identify tree type only if it is a tree, but it can be used well for the proportion of tree cover It can be noisy bc some pixels like grass and other non-trees are classified as trees, but they use a special algorithm with an adjustable/configurable parameter to adjust the noise

this gitHub repo shows a usage of detectree to map out the tree canopy coverage of a town in Switzerland https://github.com/martibosch/lausanne-tree-canopy

this repo uses detecTree and deepforest to sement and classify trees in munich https://github.com/easz/urban-tree

adityaa-chandna commented 3 months ago

YOLOv7 - appears to be a form of object detection software, industry-leading standard does samgeo use YOLOv7? very quick for real-time object detection, not sure how accurate it is

adityaa-chandna commented 3 months ago

sentinel-tree-cover Maps tree cover and extent using TensorFlow models https://github.com/wri/sentinel-tree-cover

pros: works on 10m resolutions, satellite images cons: maps out areas of trees instead of individual trees

workarounds: could we write a simple program to estimate number of trees in a given area assuming continuous cover?

adityaa-chandna commented 3 months ago

TreeDetection it's literally called tree detection it does exactly that pros: takes in google images cons: i havent used it yet and i dont know what the fuc it does https://github.com/AmirNiaraki/TreeDetection

audreyromanik commented 3 months ago

Detectron2: metaAI opensources detection software chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://arxiv.org/pdf/2301.12058#:~:text=It%20provides%20many%20state%2Dof,with%20its%20provided%20pretrained%20models. https://github.com/facebookresearch/detectron2?tab=readme-ov-file Another detection software that can detect objects and create bounding boxes and masks for front-facing and aerial images, has an extensive library to compare to as well

sjkkolasa commented 3 months ago

TreeDetection it's literally called tree detection it does exactly that pros: takes in google images cons: i havent used it yet and i dont know what the fuc it does https://github.com/AmirNiaraki/TreeDetection

I think we can run this directly in ArcGIS

https://www.youtube.com/watch?v=_isX3WKIk-E&t=161s

audreyromanik commented 3 months ago

Qgis has plugins that can seperate an aerial image into layers(ex. buildings woodlands, water etc) this plugin is called deepness and should be easy to install on anything but mac os https://www.youtube.com/watch?v=UTiILsy0Mt8&ab_channel=GeospatialSchool https://plugins.qgis.org/plugins/deepness/#plugin-versions