Parts-per-Object Count in Agricultural Images: Solving Phenotyping Problems via a Single Deep Neural Network

Solving many phenotyping problems involves not only automatic detection of objects in an image, but also counting the number parts per object. We propose a solution form single deep network, tested for three agricultural datasets pertaining to bananas-per-bunch, spikelets-per-wheat-spike, and berries-per-grape-cluster. The suggested network incorporates object detection, resizing, part as modules with several variants tested. module is based on Retina-Net architecture, whereas modules, two different architectures are examined: first direct regression predicted count, other explicit counting. results promising, mean relative deviation between estimated visible count range 9.2% 11.5%. Further inference count-based yield related statistics considered. For banana bunches, actual (including occluded bananas) inferred from bananas. robust estimation methods employed get average spikelet across field, which effective estimator.