Skip to content

Latest commit

 

History

History
62 lines (53 loc) · 2.98 KB

Progress.md

File metadata and controls

62 lines (53 loc) · 2.98 KB
Raw

Corn Disease Detection

2017

  • Idea origin on 2017(wished to do some AI project on Agriculture but i was novice back then).

2018- 2019

  • Did some of ML codes but skipped mathematics(so i am still having problems).

2020

March

  • Got an idea to do some detection project on hens.
  • The idea was to detect died hens and count them then warn the owner. The concept was to collect image dataset and use YOLO like model later.
  • But failed because data collection was not succesful.

April

  • Still no idea what to do next.

May

  • Still no idea what to do next.

June

  • Our corn got infected due to some swamp insects and pests.
  • Then i took first photo of corn's plant on June 12.
  • I was late to collect photos and our field has only few infected and other farmer's corn was already grown. So i was able to collect only 223 images.
  • Also electricity was gone for days and many climate changes.
  • I wrote some agumentation code to make 10 images from one image.
    • Flip x
    • Flip y
    • Flip xy
    • Brightness
    • Erosion
    • Dilation followed by Erosion
    • Dilation
    • Erosion followed by Dilation
    • Scale 0.75
    • Scale 0.5

July

  • Took 2009 images of freshly infected plants. There were more than 5k infected plants but my phone battery was low.
  • Those field of corn was totally ruined.
  • Found a bug of image read and annotation of VOTT. If we rotate image 90 degree then the annotation becomes true else annotation becomes wrong. (using np.rot90)
  • Wrote a code on 3rd day to find annotation of agumented images.
  • Wrote some codes to find region of images and also used some image segmentation codes but used selective-search algorithm.
  • Learned about some data generators for Keras.
  • 11th day, completed annotation. Still having question should i do agumentation?
  • Having electricity problem still on 11th day and heavy rain.
  • Should go to wifi zone and stay for days to train a detection model using GPUs.
  • July 22, about to sell classification of infected/normal corn leaf image to some MSc students.
  • As for july 26, nothing exciting happening. But studying about Region Proposal Network. It is hard to understand code of other peoples so i am thinking about writing my own version of RPN.

Agust

  • Wrote a code to classify whether a patch of corn is infected or not. Had to make custom data generator and then feed it to classifier.
  • Worte a code to predict possible Bounding Boxes on images by training it on images and bounding boxes. I think it doesn't work.
  • Both above codes were tested on local hardware and still dataset is not on cloud.

September

  • Got help on my internet by Vikash Krishna and trying to test YOLO for corn on colab.
  • Using TrainYourOwnYolo for training. My entire images has to be rotated by 90 and only the annotation works.
  • Train unsuccessful.

October