Danielle Nuzillard

Partitional Clustering Techniques for Multi-Spectral Image Segmentation

Analyzing unknown data sets such as multispectral images often requires unsupervised techniques. Data clustering is a well known and widely used approach in such cases. Multi-spectral image segmentation requires pixel classification according to a similarity criterion. For this particular data, partitional clustering seems to be more appropriate. Classical K-means algorithm has important drawbacks with regard to the number and the shape of clusters. Probability density function based methods overcome these drawbacks and are investigated in this paper. Two main steps in data clustering are dimension reduction and data representation. Methods like PCA and ICA often perform dimension reduction step. To achieve a complete and more reliable representation of the data, a magnitude-shape representation is described, it takes into account both the magnitude and shape similarities between pixels vectors. The bases of PCA and magnitude-shape representation are explored to highlight the main differences and the advantages of our method over PCA. Experimental results confirm that this method is a reliable alternative to classical linear projection methods for dimension reduction.

A Novel Capacitive Safety Device for Target Localization and Identification

A novel capacitive device dedicated to the safety of people working on automated production sites has been developed. It can be implemented on an articulated mechanical system (robot, press) to estimate the distance between sensors that are fixed on such a system and a mobile target. An interdigital sensor has been fabricated to identify the target nature among them (human beings or objects). The variable geometry of the system is taken into account as well as its motion relatively towards the ground by means of an ingenious sensitive element design. Furthermore, the sensors are made of conducting paint, thus providing low weight and negligible space hindrance devices. Results show this novel low-cost sensing setup has a great potential to estimate the distance and to discriminate targets.

Static Stress Measurement Based on Capacitive Probes Integrated Inside Rolling Rings

An experimental setup has been conceived to measure the deformation of a rolling ring subjected by stress. This deformation is measured by means of a capacitive sensor. Such a sensor behaves like a displacement sensor that translates the displacement into an electrical signal. Mechanical and electrical operating points determine the sensitivity of the probe and its measurement range. Numerical simulations and tests performed on a prototype confirm the relevance of the device. Applications are numerous: i) mounted on the wheels of motorized vehicles, such sensors allow one to improve the quality information transmitted to integrated safety devices and ii) placed on devices of lift, they measure the load.

Denoising Using Blind Source Separation for Pyroelectric Sensors

This paper deals with a process of denoising based on a Blind Source Separation (BSS) method. This technique is inserted in an experimental device of nondestructive testing. Its excitation is a laser beam and its detectors are pyroelectric sensors. The latter are sensitive to the temperature. As they are also piezoelectric, they are particularly sensitive to the environmental noise. Therefore, it is necessary to denoise them. With this aim in view, a technique of blind source separation is implemented. One source corresponds to the incidental beam and the other sources are various noise. A judicious experimental device was designed in the laboratory. It fits to the requirements of the BSS technique, and it allows indeed a restoration of the incident signal.

BSS, Classification and Pixel Demixing

In the framework of the analysis of remote sensing images, the pixel mixture is a difficult task to solve. As it is considered that a mixture of pure elements is observed, it is necessary to identify them and to determine their proportions. Thus we associate statistical methods of Blind Source Separation (BSS) to complementary techniques of classification. Our purpose is developed and illustrated through an application on images for which a ground analysis was carried out. A comparison between a statistical approach and a clustering one is performed. Even if the BSS approach does not provide the classes associated to the ground analysis, it allows us to refind these classes from a simple learning.

A New Geometrical BSS Approach for Non Negative Sources

A new blind source separation method for non-negative sources based on geometrical evidences of the linear mixing model is presented. We show that the proposed method is able to find the mixing matrix as well as the original sources from an observation matrix under the assumption that for every source there is at least one instance where the underlined source is active and all the others are not. One major advantage of our proposal is that the number of sources is found automatically as being the number of extreme data in a set of points. Under the assumption mentioned above, our approach outperforms two well known implementations for NNMF BSS (ALS and multiplicative update algorithms).

Two neural network based strategies for the detection of a total instantaneous blockage of a sodium-cooled fast reactor

The total instantaneous blockage (TIB) of an assembly in the core of a sodium-cooled fast reactor (SFR) is investigated. Such incident could appear as an abnormal rise in temperature on the assemblies neighbouring the blockage. Its detection relies on a dataset of temperature measurements of the assemblies making up the core of the French Phenix Nuclear Reactor. The data are provided by the French Commission of Atomic and Alternatives Energies (CEA). Here, two strategies are proposed depending on whether the sensor measurement of the suspected assembly is reliable or not. The proposed methodology implements a time-lagged feed-forward neural (TLFFN) Network in order to predict the one-step-ahead temperature of a given assembly. The incident is declared if the difference between the predicted process and the actual one exceeds a threshold. In these simulated conditions, the method is efficient to detect small gradients as expected in reality.

Non-negative matrix factorization approach to blind image deconvolution

A novel approach to single frame multichannel blind image deconvolution is formulated recently as non-negative matrix factorization (NMF) problem with sparseness constraint imposed on the unknown mixing vector. Unlike most of the blind image deconvolution algorithms, the NMF approach requires no a priori knowledge about the blurring kernel and original image. The experimental performance evaluation of the NMF algorithm is presented with the degraded image by the out-of-focus blur. The NMF algorithm is compared to the state-of-the-art single frame blind image deconvolution algorithm: blind Richardson-Lucy algorithm and single frame multichannel independent component analysis based algorithm. It has been demonstrated that NMF approach outperforms mentioned blind image deconvolution methods.

Dynamic detection of nuclear reactor core incident

Surveillance, safety and security of evolving systems are a challenge to prevent accident. The dynamic detection of a hypothetical and theoretical blockage incident in the Phenix nuclear reactor is investigated. Such an incident is characterized by abnormal temperature rises in the neighbourhood of the concerned reactor core assembly. The dataset is the output temperature map of the reactor, it is provided by the Atomic Energy and Alternative Energies Commission (CEA). A real time approach is proposed, based on a sliding temporal window, it is divided into two steps. The first one behaves like a sieve, its function is to detect simultaneous temperature evolutions in a close neighbourhood which may induce a potential incident. When such evolutions are detected, the second step computes the temperature contrast between each assembly having these evolutions and its neighbourhood. This method permits to monitor the system evolution in real time while only few observations are required. Results are validated on various noisy realistic simulated perturbations.

Capacitive protection system for robot dedicated to the safety of humans

This protection system is dedicated to the safety of people working on automated sites of production. The system is implemented on an articulated mechanical system, (robot, press...). Capacitive sensors exploit the properties of an electric field to estimate the distance between a detected target and themselves. By the way, it makes it possible to discriminate the presence of human beings or objects in their environment and to give information about their distance. Moreover, the system takes into account the capacitive variations which are induced on itself due to its variable geometry (interactions between the various elements of the articulated mechanical system and the relative approach of the ground). The sensitive elements of detection are of a low cost because they are made with a conducting paint put on the dangerous parts of the machine. Their weight and their space hindrance are negligible, this is essential for an embarked tool of a robot.