What are the best practices in using the Handheld?

 

When crop is stacked in several layers, the device is likely to detect partially visible crop in the lower layers. This leads to an overestimation of undersized crop in the sample and therefore biases the results. The example below illustrates the issues caused by stacking crop.

 

 

 

If the crop is too closely clustered (i.e. with no gaps between individual pieces), sizing accuracy can be reduced (see the first image below). In certain conditions, clustering may not only interfere with correct detection of each crop’s contour, but also prevent some crop from lying completely flat on the surface. As a result, the camera may not capture the full length of those crops (see the second image below – the crop highlighted in red will have less accurate measurements).

 

 

Capture images with the tablet positioned at a 90° angle to the crop (see the image below).

 

 

Ensure there is sufficient lighting so that the crop is clearly visible. Also make sure there is strong contrast between the crop and the background. If the crop and background are similar in colour, the system may find it more difficult to detect the crop contours. Where needed, we recommend placing a bag or board beneath the crop to improve contrast (see example below).

 

The confidence slider is located under the text "Adjust detections". It controls how confidently the system identifies potatoes. By default, it is set to 90%, meaning objects detected below this threshold are ignored. Sliding the control to the left lowers the threshold and increases detections, which may help in environments where potatoes are missed - but may also include non-potato objects. Sliding it to the right results in cleaner detections but may miss some potatoes. You may need to adjust this setting depending on conditions to optimise results. Please see below a practicable example.