Data for crop diseases
marshallrichards opened this issue · comments
Searching through large databases like imagenet as well as academic databases there is a lack of specific and iterable data to go through to have an accurate tensorflow application. The image recognition is directly reliant on the amount of imagery supplied and that is currently the largest bottleneck of this project. A large effort needs to be made on compiling images into categories of diseases first by hand and then, hopefully later, run our algorithm on new content. But before we can supply user data we need our hand compiled sets of data to train on.
I would say focus on nutrient deficiencies. Easy to pull images for at least for what/how im farming... nutrient deficiencies seam to be more of a problem and have more of a pattern that can be seen with a pic of the full plant. This just made my day ill try to add where features where i can. example there are feature changes that location specific but the location in taring might not be reflected in the training set. it would be benifical keeping track of locations of blossoms in the affected area aswel as the top of the plant. my reasioning is i belive this is how a botanist would talk to a man who has gone blind. simple stimuli at first right? PS bing image>google image