Name: Daniela Zaharie
Affiliation: West University of Timișoara
Primary research interest: evolutionary algorithms, neural networks, data mining, stochastic modelling, image processing
Name: Teodora Selea
Affiliation: West University of Timișoara
Primary research interest: deep learning, distributed systems, image processing
Title of the lecture: Ensemble Models in Image Processing
Keywords: ensemble learning, image classification and regression
Summary: Ensemble learning proved to be an effective strategy in constructing performant predictive models by aggregating the results produced by several base models. The ensemble learning process consists of several steps:
(i) choice of the collection of base learners;
(ii) the construction of base models using specific strategies of selecting training data from the available dataset;
(iii) the aggregation of the results produced by the base models.
Different types of approaches for each of these steps led to various ensemble strategies (bagging, boosting, stacking) which proved to be effective in increasing the performance in classification and regression tasks. In the context of image processing, both traditional methods (Random Forests and AdaBoost) as well as recent ensembles of deep neural networks are popular approaches.
The lecture is organized in two parts. The first part will present an overview of design principles and practical aspects in constructing ensemble models with examples related to image classification tasks. The second part will focus on semantic segmentation for satellite hyperspectral images. Various aspects related to data pre-processing and the usage of ensemble models will be illustrated.