produce sort

• 

  The control

• 

  A rotting clementine (not the one we grabbed at lunch yesterday)

• 

  It also picks up on highlights/crowns sometimes

• 

  An apple

• 

  A plum

• 

  Works with non-circular food too!

• 

  Edge detection is sometimes imperfect if there isn't enough contrast in the original image

Inspiration

We wanted to make a computer vision user app that detected if a fruit was good to eat or not based on its discolouration/irregularities after picking up a few discoloured/bruised oranges at lunch on the first day of McHacks.

What it does

It uses scikit-image to detect edges using the canny algorithm, which it then filters with a Gaussian distribution to subtract noise. It uses the edges to create a mask to filter out the background, which it feeds into a blob detection (difference of Gaussian) method with specific parameters to extract the moldy blobs/irregularities. The final result plots the original image, the edges detected on it, the mask applied to the edge detection and the blobs found using the doG method. The backend is done in Python and the frontend is a basic user UI to upload jpegs/jpgs/pngs to.

Challenges we ran into

Edge inconsistencies are harder to detect than we thought. We originally wanted to determine how far a bruised orange deviates from a 'perfect' orange shape.

What's next for produce sort

Making a better user interface (an actual landing page) and maybe using tensorflow to get a better idea of food that is safe to consume based on its appearance.

Built With

• cv
• python
• scikit