Image capturing setup
Bühler AG safe grain; workshop #13
Wheat represents one of the most important food crops and it is eaten by 2.5 billion people in 89 countries (http://wheat.org/wheat-in-the-world/). With nearly $50 billion worth of trading all over the world, it represents a challenging research subfield of the agricultural industry. On the other hand, wheat is wasted all over the world. I.e. India wastes 21 million tons of wheat per year and one of the reasons is lack of or late detection of contaminated storages (http://timesofindia.indiatimes.com/india/India-wastes-21-million-tonnes-of-wheat-every-year-Report/articleshow/17969340.cms). Machinery for detection of wheat contamination can be really expensive and often not available to developing countries. With the “Check my Grains” project, we aim to bridge that problem by offering a simple and effective way to detect contamination in wheat. Our project consists of IoT kit which takes photos of wheat containers, sends the photos to the server where after several image processing operations the objects that contaminate the wheat are detected and marked.
What it does
Detects contamination in wheat containers.
How We built it
IoT kit, which is consisted of Raspberry Pi 3 including camera sensor, sends an image to a web server written in Python. The image is processed by applying several filters, edge detection, and ultimately contour extraction. The contours are then sent to the ML algorithm (which uses Two-class Decision Forest method) and classify if the extracted grain is a contamination or not. The results are sent back to the server and after evaluating a report stating the level of contamination is presented.
Challenges I ran into
Calibrating the camera for taking images is the most challenging process. For best performance, controlled conditions with the right angle and intensity of light are crucial. Furthermore, the algorithm has to be re-calibrated every time the camera is changed. Preparing the training set for the ML algorithm was challenging due to the fact that generating images was heavily dependent on other conditions and hence the evaluation of the algorithm was dependent as well.
Accomplishments that We're proud of
All three of us learned a lot working on this project. We tackled a challenge that was at the same time interesting and visionary. Learning more about image processing, machine learning, and IoT while applying those technologies to real-world problems was a really nice way to spend the weekend.
What We Learned
That the project is as strong as its weakest link.
What's next for "Check my Grains"
Use more brains!