0
points
Existing software to Diagnose plant disease from pictures

Big Data in Agriculture    Belgium

Which are the existing software options available to use for the recognition on pictures of plant diseases?

Sometimes a specific option does not recognise the disease in question while another one could, so it's interesting to know different possibilities.
Also, we have a lot of pictures that we would like to put for the learning process, but we have no knowledge of independently working API to process pictures in bulk.


Posted by: Elias (1 point) Elias
Posted: May 24, 2019




Answers

2
points
Tensorflow is a great machine learning library for image recognition/object detection. I suggest beginning with image recognition as it is simpler than object detection, but you can still get great results. Here's a tutorial from Tensorflow on how to use transfer learning with the Inception V3 classifier to do image recognition with your own dataset.

https://www.tensorflow.org/hub/tutori...

This particular tutorial only fine tunes the final layer on the convolutional neural network, instead of tuning the entire network. This allows you to run the training session with minimal hardware capacity, while still giving good accuracy. Here's a paper our group did in 2017 demonstrating that retraining the final layer of Inception v3 with our cassava disease dataset gives >95% classification accuracy for some diseases. We also show that training an SVM and KNN classifiers on the feature vectors gives good (sometimes better) results as well.

https://doi.org/10.3389/fpls.2017.01852


Posted by: Peter McCloskey (1 point) Peter McCloskey
Posted: May 25, 2019


Elias commented,
Thank you for the comment.
Tensorflow seems to be going a lot in the direction of what we are looking for. However we are not programming specialists so it might take a whole while to find out the workings.. So we make progress with an already finished program.
Plantix does not put its API at disposition of extenals anymore.
Any opinion about Agrio from Saillog?

6 months ago.



1
point
The most common is tensorflow object detection. That is where one trains a machine to correctly identify diseases. One can do it by classifying the whole leaf or use object detection to identify an area of disease in the leaf.

We do wish to get some demo blogs up to help people navigate this work and learn. Our main message is to work with agronomists in lock step to train the well-functioning models that perform as well or better than humans.


Posted by: David Hughes (43 points) David Hughes
Posted: May 24, 2019


Elias commented,
thank you. We are agronomists, so, for us it is to make the software part work that is difficult :-).
that's why we are looking for a ready-to-use solution.
Plantix was doing such a thing, but they have retracted their API use for externals. Maybe in the future it will come back.
Any idea about the working of Agrio from saillog?
Or other alternative software that is user-friendly.
However i take serious note of tensorflow, maybe we can find a skilled person that can help us to make it work.
Good continuation to you all.
many thanks

6 months ago.



You need to log in if you'd like to add an answer or comment.