Inspiration

About a quarter of global emissions comes from agriculture, so does most of total water usage and a third of global energy. We were shocked to learn that 30% of all food goes to waste, and thus drew motivation from this to work on means of early detection of contamination, which might aid at preserving some of the food.

What it does

Our framework is based on 3 major steps: 1) it records images of samples; 2) it detects the presence of contaminants; 3) it quantifies the level and type of contamination. Additionally, we added two features: one of the measurement of atmospheric conditions, and one enabling early detection of pests.

how we built it

We used a raspberry pi and interfaced it with a camera for imaging and programmed it with via the command line. The detection of the presence of contamination is conducted with machine learning implemented in Microsoft Cognitive Services, with which we trained our samples. The degree of contamination is determined with image processing techniques, including image dilation and filtering. Additionally, we included two elements and interfaced them via a micro-controller with raspberry pi, namely a barometer and a light sensor. In this way, we can monitor the temperature and pressure of the container with samples, preventing degradation due to overheating and overpressurising. The light sensor measures the luminosity of the sample below it, and can detect changes in the brightness due to the movement of the grains (e.g. when they are pushed by bigger insects or mice), or a pest itself.

Challenges I ran into

MATLAB couldn't work on Raspberry Pi, we struggled with installing missing libraries, hence we had to divide the demo into a two parts: one based on the raspberry pi, one on the dell laptop. Another challenge was related to creating a website communicating with python via a web server but we eventually found a workaround/a hack. Deep learning will be the most suitable for contamination recognition tasks, however, we were not able to implement it due to time limitations.

Accomplishments we are proud of

We are proud of expanding the original project with new ideas (sensors). Our algorithm for image recognition.

What we learnt

new techniques in image recognition programming raspberry pi's sensors and a camera working under pressure with little sleep :)

What's next

We want to add a humidity sensor to prevent grains from developing mould. Our prototype could use some design. We need to migrate our distributed scripts into one OS. Our website should be running properly on the server and execute python/matlab scripts with a click on the button.