Multivariate time series classification supports applications from wearable sensing to biomedical monitoring and demands models that can capture both short-term patterns and multi-scale temporal dependencies. Despite recent advances, Transformer and CNN models often remain computationally heavy and rely on many parameters. This work presents PRISM (Per-channel Resolution Informed Symmetric Module), a lightweight fully convolutional classifier. Operating in a channel-independent manner, in its early stage it applies a set of multi-resolution symmetric convolutional filters. This symmetry enforces structural constraints inspired by linear-phase FIR filters from classical signal processing, effectively halving the number of learnable parameters within the initial layers while preserving the full receptive field. Across the diverse UEA multivariate time-series archive as well as specific benchmarks in human activity recognition, sleep staging, and biomedical signals, PRISM matches or outperforms state-of-the-art CNN and Transformer models while using significantly fewer parameters and markedly lower computational cost. By bringing a principled signal processing prior into a modern neural architecture, PRISM offers an effective and computationally economical solution for multivariate time series classification. Code and data are available at https://github.com/PRISM-TS/PRISM .

Apprentissage de représentations invariantes de domaines pour des données d'images hétérogènes L’apprentissage profond supervisé repose largement sur une grande quantité de données étiquetées, souvent difficile à obtenir. L'adaptation de domaine résout ce problème en appliquant les connaissances acquises à partir d'un jeu de données étiqueté à un autre jeu lié, mais non ou faiblement étiqueté. L'adaptation de domaine hétérogène est particulièrement complexe car les domaines se situent dans différents espaces. Ces méthodes sont très intéressantes pour les champs où une variété de capteurs est utilisée (comme la télédétection), capturant des images de différentes modalités. Cette thèse propose des approches novatrices pour l'Adaptation de Domaine d'Images Hétérogènes (ADIH) basées sur l'extraction de caractéristiques invariantes de domaines. La thèse examine différents scénarios de supervision dans le domaine cible : non supervisé, semi-supervisé et avec des contraintes. Les résultats montrent que l'approche proposée surpasse de manière cohérente les méthodes concurrentes.

Building footprint detection from remote sensing imagery remains an activate field of research. In particular, training models that generalise well remains challenging. A common approach to building generic models is to leverage domain adaptation approaches based on style transfer to adapt labelled datasets to an unlabelled domain. These assume the availability of labelled and unlabelled data during training. In applications such as emergency mapping however, the target domain is not always known in advance, and there is therefore a need to build models that generalise to domains unseen at training time. Using SpaceNet data various domain adaptation approaches are evaluated to improve model gen-eralisation. This article demonstrates that domain adaptation improves model generalisation, and the choice of source and target domain can be more significant than the choice of style transfer algorithm.

The application of supervised deep learning methods in digital pathology is limited due to their sensitivity to domain shift. Digital Pathology is an area prone to high variability due to many sources, including the common practice of evaluating several consecutive tissue sections stained with different staining protocols. Obtaining labels for each stain is very expensive and time consuming as it requires a high level of domain knowledge. In this article, we propose an unsupervised augmentation approach based on adversarial image-to-image translation, which facilitates the training of stain invariant supervised convolutional neural networks. By training the network on one commonly used staining modality and applying it to images that include corresponding, but differently stained, tissue structures, the presented method demonstrates significant improvements over other approaches. These benefits are illustrated in the problem of glomeruli segmentation in seven different staining modalities (PAS, Jones H&E, CD68, Sirius Red, CD34, H&E and CD3) and analysis of the learned representations demonstrate their stain invariance.

The advent of high-resolution instruments for time-series sampling poses added complexity for the formal definition of thematic classes in the remote sensing domain-required by supervised methods-while unsupervised methods ignore expert knowledge and intuition. Constrained clustering is becoming an increasingly popular approach in data mining because it offers a solution to these problems, however, its application in remote sensing is relatively unknown. This article addresses this divide by adapting publicly available k-Means constrained clustering implementations to use the dynamic time warping (DTW) dissimilarity measure, which is thought to be more appropriate for time-series analysis. Adding constraints to the clustering problem increases accuracy when compared to unconstrained clustering. The output of such algorithms are homogeneous in spatially defined regions.

Although agreement between the annotators who mark feature locations within images has been studied in the past from a statistical viewpoint, little work has attempted to quantify the extent to which this phenomenon affects the evaluation of foreground-background segmentation algorithms. Many researchers utilize ground truth (GT) in experimentation and more often than not this GT is derived from one annotator's opinion. How does the difference in opinion affects an algorithm's evaluation? A methodology is applied to four image-processing problems to quantify the interannotator variance and to offer insight into the mechanisms behind agreement and the use of GT. It is found that when detecting linear structures, annotator agreement is very low. The agreement in a structure's position can be partially explained through basic image properties. Automatic segmentation algorithms are compared with annotator agreement and it is found that there is a clear relation between the two. Several GT estimation methods are used to infer a number of algorithm performances. It is found that the rank of a detector is highly dependent upon the method used to form the GT, and that although STAPLE and LSML appear to represent the mean of the performance measured using individual annotations, when there are few annotations, or there is a large variance in them, these estimates tend to degrade. Furthermore, one of the most commonly adopted combination methods-consensus voting-accentuates more obvious features, resulting in an overestimation of performance. It is concluded that in some data sets, it is not possible to confidently infer an algorithm ranking when evaluating upon one GT.

Surface fissures are important indicators for slope instability and their patterns reveal information about the distribution of strain and mechanical processes. The increasing availability of sub-decimeter resolution aerial images may enable the detection and mapping of such features with imagery. This study combines Gaussian matched filters and first order derivates in a multi-scale framework to address the semi-automatic detection of surface cracks with variable sizes. The accuracy of the detector was assessed with two expert mappings and compared to a corresponding single scale approach.

Reliable feature detection is a prerequisite to higher level decisions regarding image content. In the domain of spectrogram track detection and classification, the detection problem is compounded by low signal-to-noise ratios and high variation in track appearance. Evaluation of standard feature detection methods in the literature is essential to determine their strengths and weaknesses in this domain. With this knowledge, improved detection strategies can be developed. This paper presents a comparison of line detectors and a novel, multi-scale, linear feature detector able to detect tracks of varying gradients. We outline improvements to the multi-scale search strategies which reduce run-time costs. It is shown that the Equal Error Rates of existing methods are high, highlighting the need for research into novel detectors. Results demonstrate that the proposed method offers an improvement in detection rates when compared to other, state of the art, methods whilst keeping false positive rates low. It is also shown that a multi-scale implementation offers an improvement over fixed scale implementations.

This paper proposes an extension to the active contour algorithm for the detection of linear patterns within remote sensing and vibration data. The proposed technique uses an alternative energy force, overcoming the limitations of the original algorithm, which relies upon simple energy formulations to extract intensity and gradient information from an image. We overcome these by forming a noise model, which is used to detect a feature¿s presence, and by integrating information from several locations within an image to strengthen the detection process.