One of the most challenging computer vision problems in the plant sciences is the segmentation of roots and soil in X-ray tomography. So far, this has been addressed using classical image analysis methods. In this paper, we address this soil-root segmentation problem in X-ray tomography using a variant of supervised deep learning-based classification called transfer learning where the learning stage is based on simulated data. The robustness of this technique, tested for the first time with this plant science problem, is established using soil-roots with very low contrast in X-ray tomography. We also demonstrate the possibility of efficiently segmenting the root from the soil while learning using purely synthetic soil and roots.

One of the most challenging computer vision problem in plant sciences is the segmentation of root and soil from X-ray tomography. So far, this has been addressed from classical image analysis methods. In this paper, we address this root/soil segmentation problem from X-ray tomography using a new deep learning classification technique. The robustness of this technique, tested for the first time on this plant science problem, is established with root/soil presenting a very low contrast in X-ray tomography. We also demonstrate the possibility to segment efficiently root from soil while learning on purely synthetic soil and root.

No abstract

Seed imbibition is a very important process in plant biology by which, thanks to a simple water income, a dry seed may turn into a developing organism. In natural conditions, this process occurs in the soil, e.g., with difficult access for a direct observation. Monitoring the seed imbibition with non-invasive imaging techniques is therefore an important and possibly challenging task if one tries to perform it in natural conditions. In this report, we describe a set of four different imaging techniques that enable to addressing this task either in 3D or in 2D. For each technique, the following items are proposed. A detailed experimental protocol is provided to acquire images of the imbibition process. With the illustration of real data, the significance of the physical quantities measured in terms of their relation to the income of water in the seed is presented. Complete image analysis pipelines are then proposed to extract dynamic information on the imbibition process from such monitoring experiments. A final discussion compares the advantages and current limitations of each technique in addition to elements concerning the associated throughput and cost. These are criteria especially relevant in the field of plant phenotyping where large populations of plants are imaged to produce quantitatively significative traits after image processing.

No abstract

The computed tomography imaging spectrometer (CTIS) is a snapshot hyperspectral imaging system. Its output is a 2D image of multiplexed spatiospectral projections of the hyperspectral cube of the scene. Traditionally, the 3D cube is reconstructed from this image before further analysis. In this paper, we show that it is possible to learn information directly from the CTIS raw output, by training a neural network to perform binary classification on such images. The use case we study is an agricultural one, as snapshot imagery is used substantially in this field: the detection of apple scab lesions on leaves. To train the network appropriately and to study several degrees of scab infection, we simulated CTIS images of scabbed leaves. This was made possible with a novel CTIS simulator, where special care was taken to preserve realistic pixel intensities compared to true images. To the best of our knowledge, this is the first application of compressed learning on a simulated CTIS system.

Dans cet article, nous présentons un retour d’expérience sur un module d’initiation à l’apprentissage automatique de 3 crédits ECTS que nous avons créé pour des élèves ingénieurs de l’INSA de Lyon. Nous présentons la structuration de ce module ainsi que les spécificités pédagogiques liées au faible nombre d’heures de face à face au regard de la complexité du sujet. Nous montrons des exemples de cas d’usage proposés aux étudiants et faisons état de leurs réactions. L’ensemble est complété par des liens bibliographiques vers les sites des outils numériques libres et références pédagogiques utilisées.

The computed tomography imaging spectrometer (CTIS) is a snapshot hyper-spectral imaging system which has recently been demonstrated of value when used in a compressed learning mode. In such a mode, the raw data are not reconstructed in an hyperspectral cube but are directly transmitted to a neural network to perform classification. While the previous investigations on this topic were limited to a simulation perspective, we extend these results to real images and demonstrate the possibility to train the network on simulated data and apply this trained model successfully on real images.