hello everyone
I want to get experienced in the machine learning and deep learning in agriculture application based on remotely sensed images
In this way, I want to explore all the process from beginning (acquiring required images to provide next measurements) to end (extract final targets like yield estimation, product identity etc.)
Can everyone summarize the exact process flow work?
For example,
- Which images should I acquire with drone or what satellite images (like orbview/aster) could be enough?
- How do I should apply measurements to estimate yield or define crop type, perhaps need to be performed over time
- What is the common widely used measurement equations for yield estimation?
- Finally, if I want to enhance the results using deep learning in MATLAB, have anyone any suggestion?
Dear Solmaz Babakan
There are many different approaches that can be found through literature review and every single researcher can design his/her own research path in order to estimate the agricultural products by remote sensing imagery and machine learning tool. Thus, there is no straight and specific method that can be found here.
However, I myself as a researcher who works with remote sensing images in different areas such as air pollution, solid waste management, water resources, and forestlands, suggest you first define your study area. For example, if you are looking for a provincial estimation, probably it could be better for you to use products such as MODIS on aqua or terra satellites. On the other hand, if you are assessing a very specific region, Landsat8's first and second level product for example would be suitable. Lastly, if you are searching for something special in a very local area, probably Sentinel 2 satellite would be a better choice. Sentinel 2 satellite offers very great estimation in terms of vegetation indices in red edge regions and sometimes authors recognize it as a special satellite for agricultural and plant health monitoring satellite. In conclusion, there are different options on the table that has to be considered based on your site selection.
For the machine learning section, I would say again there are different approaches that can be implemented in your analysis. I use regularly the Random forest model as a machine learning algorithm since it is known as a method that does not include overfitting, has higher accuracy and better performance in environmental applications. However, it again depends on your objectives. Sometimes it is being used for classification, sometimes temporal predictions, and sometimes spatial predictions.
down the road, I suggest an integrated literature review in your specific research area, which dramatically can help to figure out your future research path and helps to detect your mind map.
I hope it helps,
So thanks for your kindly response
Absolutely you're right but there is such a problem that I exactly work on mineralogical mapping and I am unfamiliar with agricultural issues
I search in thesis and articles about that, but they do not provide my required information
for example, they construct a regression between NDVI, etc. and yield
but how this yield was obtained
if there are recorded governmentally and could be extracted there or direct survey on the agricultural regions gives this data
in different seasons this was conducted
but if these measurements provided seasonally by farmers or corresponding officers
how this data is collected?
in fact, I could perform the required estimations or coding process to obtain last results
but I can’t find how should I provide my required raw data about agricultural properties
as well, appropriate sattelite images are mentioned in the sources but if there need for imaginary by drones, how does its conditions considered
for example, yield in training regions how could be calculated in practice
I offer my respect for you attended the discussion
There are a lot of Agri Startups in India doing this kind of work. Highly Technical as also proprietary.
- Badges
- Science topic
- Similar topics
- Agriculture