+36 70 203-3120 | agrowebsystem@gmail.com | Blog | Contact
ROI eper mezőgazdasági képfeldolgozás

Highlighting ROI for strawberries

  • 254

One of the concepts in image processing is ROI. This is nothing more than the areas of the image that are of interest to us.

ROI – Region Of Interest.

It’s no different in agricultural image processing.

We have our pictures: in our case, a picture of a strawberry field. And we’re wondering if there are any ripe strawberries in it. If there are, where they are. We highlight these strawberries from the picture and pass them on to a strawberry grading algorithm, which determines whether the strawberry is healthy or not, or how ripe it is based on color.

We can send this to an expert. It is expensive to have an expert visit 10-20 orchards every 2 days. Taking a series of high-resolution images of the area and sending those high-resolution images, typically via mobile internet, is a slow process. Especially since only a 5-10% of the images are of interest to the expert. Therefore, having an ROI image processing algorithm in place is very useful in such cases.

Another example is to send a peach leaf curl with taffy in this way. Someone takes 2-2 shots of the trees. Each picture will show at least 200-200 leaves, but maybe only every 10th tree has 1-2 diseased leaves at the beginning of an epidemic.

In such a case, if the algorithm quickly looks through the leaves, even several 10s or 100s of thousands, and highlights the 50 that are suspect cases in a gallery and sends only those to the expert, that’s a done win. And the expert decides what diseases he sees or maybe it’s just chewing or something else that caused the leaf to become deformed.

So in our example, let’s take an image that includes strawberries and highlight only those regions – ROI – that contain mature strawberries.

For H, we now also pass the image in the HSV domain and create a mask from the H domain.

On the mask, we perform a contour detection.

We filter out contours that are too short, because they are typically noise in the background.

The inclusion squares of the remaining contours are determined. Then these parts are highlighted from the image. And we can send it to the experts…

And finally, here are the 3 little pictures we were wondering about: