Introduction
In this project, I will run a set of imagery and deep learning analysis to detect and identify palm trees in Kolovai and its health. ArcGIS Pro and Python were used to detect this analysis. This will help the locals to quickly adapt and find solutions to ensure that palm trees remain healthy and free of diseases or viruses. By identifying and measuring the health of the palm trees within the area, authorities and locals can make decisions based on the outcome of this analysis, for example, to stop the spread of a contagious palm tree that can infect other healthy palm trees.
Study area
Kolovai is a village in Tongatapu, Tongo and has a small population of 4,267 people that was recorded in the year 2006 and an area of 22.5 km². It is famous for its dance that is known as the Lakalaka. The national monument received a proposal to protect and preserve the palm trees of the area.
Data Source:
The data of the site were obtained from Open Arial Map (OAM), the data type is UAV, which means that this data was collected by an unmanned aerial vehicle (UAV). It allows you to focus on a smaller scale with a high resolution of the study area and I needed it for this project, see a sample of the TIFF file used. It will be processed in ArcGIS as Tagged Image File Format (TIFF), this format allows you to store the data as a raster graphics image. Landsat images are also a great option but usually, they cover bigger scales which is not needed for this project, however, both images can provide similar insights.
Methodologies:
To run the deep learning model and detect palm trees I needed first to make a customised scale of 1:800 for 5 different parts of the image, then set bookmarks. This will help me see with the naked eye the palm trees within the area. These bookmarks were then used to train the sample using the classification tool “Label Objects for Deep Learning”. I digitised over a 100 sample of the palm trees to train the model and extract image chips that later will be used to run the deep learning models, see the image below for a sample of the digitised palm trees. The more samples you make the more accurate the model will be.
I then used Python command prompts to install deep learning libraries to ArcGIS pro and then activated both tools "Detect Objects Using Deep Learning" and "Train Deep Learning Model". I then used the image chips that I exported using the previous method to train the deep learning model so that it can be used to detect the rest of the palm trees of the study area. The trained sample was then applied to the tool “Detect Object Using Deep Learning”, this process needed a powerful GPU and CPU to run, so if you plan to use the same analysis make sure that you have a strong graphic card same as the one usually used for gaming laptops, for example, GTX 1060Ti. After running this analysis I managed to detect all the other palm trees of the study area, see the image sample below.
This method was inspired by Esri documentation and guided tutorials(https://pro.arcgis.com/en/pro-app/help/analysis/image-analyst/deep-learning-in-arcgis-pro.htm), (https://pro.arcgis.com/en/pro-app/help/analysis/image-analyst/introduction-to-deep-learning.htm) and (https://pro.arcgis.com/en/pro-app/help/analysis/image-analyst/install-deep-learning-frameworks.htm)
Diagnosing palm trees health
To diagnose the health of the palm trees in Kolovai I calculated the Visible Atmospherically Resistant Index (VARI) that was developed for the Vegetation Fraction (VF) and indirect extent of Leaf Area Index (LAI). I used this formula to analyse the health of palm trees, (Rg - Rr) / (Rg + Rr - R(Rg - Rb)). (https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLII-3/1215/2018/isprs-archives-XLII-3-1215-2018.pdf). The Rg, Rr, and Rb represent the reflectance of the Green, Red and Blue bands of the image. The raster function tool was used to depict this formula, under the Math section I used Band Arithmetic raster function and then I used the model builder to create this model. I started by feature to point tool to highlight the palm trees on the image, Buffer tool to create a radius of 3 meters around each palm tree, Zonal Statistics as Table to extract the results of the equation and finally Join Field tool to join the equation results that translate to the palm tree health, which will be added to the buffered palm trees. This model will show the palm trees for each tress that was highlighted with a radius of 3 meters. At the end of this project, below you can have a look at the map created after running all the analysis.
Conclusion
Deep learning has been trending lately due to its various uses, in this project I have used samples to train the model to identify the palm trees within Kolovai, otherwise, I would have needed to digitise the whole area, which is not very practical and time-consuming. This method allowed me to analyse and find which palm tree is healthy and which is unhealthy and with that, I can conclude that the results of this project can be used to help locals of the area make better decisions in protecting the palm trees of Kolovai. This practice can be used for many other areas and in this project I displayed in the example of what it can be used for and I look forward to using it for other projects to help in making a better decision through the insights of data and maps.
Comments