Machine learning applied on plant phenotyping for leaf counting

Valerio Giuffrida from the Alan Turing Institute in London is visiting Wageningen University & Research on 7-8 June. He is finishing a PhD under the supervision of Sotirios Tsaftaris (Univerity of Edinburgh). He is going to present his latest research in machine learning for plant phenotyping.

Organisator Wageningen University & Research

do 7 juni 2018 11:00

Locatie Gaia, building number 101
Droevendaalsesteeg 3
6708 PB Wageningen
+31 317 48 16 00
Zaal/kamer C0093 (2nd floor)

All who are interested are invited to join this exciting talk. Valerio Giuffrida's research is especially relevant for the agrofood sector.

The seminar is scheduled on Thursday, June 7th at 11am, in GAIA C0093 (2nd floor).

The abstract can be read below.

Plant phenotyping refers to the quantitative extraction and analysis of plant traits. It is important to study how plants grow under different environmental conditions, as such knowledge is useful for the development of sustainable agriculture. In the last decade, the use of image-based plant phenotyping has exponentially increased, as making manual measurements is a tedious and time-consuming task. In order to analyse plant images, robust and reliable algorithms are needed. For this reason, machine learning approaches have proven to be an excellent ally to the analysis of images of plants. One important plant trait is the number of leaves, as it is directly related to growth, flowering time, and yield potential. In this talk, I will present my work on leaf counting as a direct regression problem. The algorithm extracts visual features from patches of Arabidopsis plants to build a holistic per-plant descriptor. Plant images represented with this descriptor are used to train a direct regression model. Further, this work motivated us to explore the unsupervised learning of rotation-invariant features. I will also present my Explicit Rotation-Invariant RBM, which extends the original Restricted Boltzmann Machine model to learn rotation-invariant features.