SoylentNews

takyon writes:

A Microsoft researcher has created software for identifying flower species:

The project came about after random cross-pollination between Microsoft Research Asia chief researcher Yong Rui and botanists at the Chinese Academy of Sciences. Rui told the academics Redmond could use image-matching to help the botanists identify the spread of flowers throughout China, albeit with some pruning of its vast image banks.

Microsoft Research Asia senior research program manager Guobin Wu says a 20-layer deep convolutional neural network was cultivated alongside learnable filters to identify slight variations between flowers. "During the forward pass, each filter is convolved across the width and height of the input volume, computing the dot product between the entries of the filter and the input," Wu says.

Some 800,000 flower snaps were planted into the UC Berkely and open source Caffe deep learning network, leading to an impressive 90 percent species identification accuracy rate.

Original Submission