Jérôme Pasquet

Research activities

Object Detection & Semantic Segmentation

To recognize all objects in an image we search the localisation of each object using boundy box or predict the probability that each pixel of the image belong to a certain object category. These two approaches can be very powerful depending on the context.

Sonar Classification

Sonar classification has been a challenging pattern recognition problem for many years mainly due to the complexity of ocean environment. In our work we propose to use Deep Learning methods to detect very small objects in the entire image, to characterize the underwater ground and also generate data using a GAN process.

Multivariate time series classification

The main goal of this work is to classify time series of astrophysical objects. Indeed, the study of the variability of the flux emitted by a celestial object along time is a crucial information in Astrophysics. However due to the properties of the telescope which is designed to observe the entire sky very quickly (in four days), a given object is not observed very often, so the data are very sparse. So classify this object is a good challenge !

Manhole covers in high-resolution aerial images of urban areas

Currently many manholes are not listed or are badly positioned on maps. The main idea is to detect the manholes cover from aerial images and then to map the underground utility networks. However detect small objects is a difficult signal processing task.

Forensics and Security

Steganalysis is the study of detecting messages hidden in a support. Most steganalysis approaches use a learning methodology involving two steps, feature extraction and a classifier. Recently, we proposed a method using a convolutional neural network to determine whether an image contains a hidden message or not.

Google Sites

Report abuse