pte20230130018 Technology/digitization, products/innovations
Software from Pennsylvania State University identifies buds that promise the highest yields
|
Apple blossoms in the original and in image processing by artificial intelligence (photos: psu.edu) |
State College (pte018/30.01.2023/12:30) –
researchers of Pennsylvania State University facilitate the artificial pollination of crops in times of rampant insect mortality with a new image processing system that clearly identifies the royal flower. This is the middle blossom of a bouquet of five blossoms on apple trees, which is the first to open. When pollinated, it produces by far the largest apple.
Hundreds of flower photos as a basis
PhD student Xinyang Mu uses “Mask R-CNN”. The deep learning computer program performs pixel-level segmentation to detect objects that are partially obscured – like the royal flower from the rest of the buds or blossoms. Mu photographed hundreds of apple blossom clusters and used them to develop an algorithm that can distinguish the royal flower from the rest of the blossoms and precisely locate it. With this information, a pollination robot can proceed in a targeted manner so that the largest fruit develops.
In 2014, Mu had Gala and Honeycrisp apple trees planted at intervals of a good 1.50 and almost two meters, respectively. When they had reached a height of almost four meters and were in bloom, the young researcher began his photo campaign. He mounted the camera on a small truck, which he drove between the rows of trees to take the pictures of the flowers. Then he trained his software to recognize the crucial bloom.
Hit rate around 98.7 percent
Training the vision system to locate royal flowers has been a major challenge, as they are the same size, color and shape as the side flowers, and because of their central position, the royal flowers are usually hidden by the surrounding flowers. He drilled the program to locate the center of a cluster of flowers. He achieved an accuracy rate of up to 98.7 percent, compared to the 100 percent achieved when identifying royals by human eyesight.
(End)
;Resize=(1200,627)&impolicy=perceptual&quality=medium&hash=2d4b303bfb5d61374c98c103aaaa6bbb4cebf92fd6100d0a0cbc82cd3e68ef97 "News about the corona pandemic: VdK for further protection of corona risk groups")