Arrangement and method for the optical assessment of crop in a harvesting machine

A method for optical assessment of crop in a harvesting machine, the method comprising:

• cleaning harvested crop in a cleaning system of the harvesting machine;
• recording, with a camera, a supplied image of the harvested crop;
• accessing, with a first neural network, the supplied image of the harvested crop from the camera;
• accessing, with a second neural network, a reference image of cleaned crop, the first and second neural networks forming a Siamese twin network, the first and second neural networks are subjected to pseudo-supervised similarity learning;
• extracting, via the first and second neural networks, at least one feature map from at least one of the supplied image of the harvested crop and the reference image of the cleaned crop;
• determining a degree of similarity of the reference image of the cleaned crop and the supplied image of the harvested crop based on the at least one feature map;
• generating an output value relating to a property of the harvested crop using at least in part the determined degree of similarity;
• identifying a reference working parameter for the harvesting machine associated with the harvesting of the cleaned crop in the reference image;
• and
• causing presentation of advice to an operator on a display device based on the output value, the advice to adjust a working parameter of the harvesting machine to the reference working parameter.

An apparatus for optical assessment of crops in a harvesting machine comprising:

• machine-readable instructions;
• and
• programmable circuitry to execute the instructions:
• train first and second neural networks via pseudo-supervised similarity learning, in which the first and second neural networks are supplied with pairs of training images having similar and dissimilar pairs and associated similarity classes;
• extract a feature map from at least one of:
• an image supplied by a camera of harvested crop using the first neural network, the harvested crop cleaned by a cleaning system of the harvesting machine, or
• a reference image of cleaned crop using the second neural network;
• calculate a mathematical distance and a degree of similarity based on the feature map;
• generate an output value based on the mathematical distance and the degree of similarity;
• identify a reference working parameter for the harvesting machine associated with the harvesting of the cleaned crop in the reference image;
• and
• cause presentation of advice on a display device based on the output value, the advice to adjust a working parameter of the harvesting machine to the reference working parameter.