The Digger3000 are using Custom Vision from Microsoft to analyse images and creates a dataset. That dataset can be used to detect similar objects in new images.
To create this dataset, the Custom Vision analyse hundreds of pictures and millions(!) of pixels. We created detection of plants and the dig mark.
We got 161 images that do not contain the object (Negative) and 135 images with the dig object. Shown in the images bellow:
We had to collect different kind of images to increase the precision and recall. Bad dataset equals a bad driver 😥
This architecture explain how it is included in the solution:
We apply for the badge [The analyst]