With an increase in the amount of natural disasters, the combined use of cloud-penetrating Synthetic Aperture Radar and deep learning becomes unavoidable for their monitoring. This article proposes a methodology for forest fire detection using unsupervised location-expert autoencoders and Sentinel-1 SAR time series. The models are trained on SAR multitemporal images over a specific area using a reference period and extract any deviating time series over that same area for the test period. We present three variations of the autoencoder, incorporating either temporal features or spatiotemporal features, and we compare it against a state-of-the-art supervised autoencoder. Despite their limitations, we show that unsupervised approaches are on par with supervised techniques, performance-wise. A specific architecture, the fully temporal autoencoder, stands out as the best-performing unsupervised approach by leveraging temporal information of Sentinel-1 time series using one-dimensional convolutional layers. The approach is generic and can be applied to many applications, though we focus here on forest fire detection in Canadian boreal forests as a successful use case.

We present a fully unsupervised learning pipeline, which involves both a projection method and a clustering algorithm dedicated to the pixel-wise classification of multitemporal SAR images. We design a Convolutional Autoencoder as the method to project our time series onto a lower dimensional latent space, where semantically similar temporal signals are placed close together. The additional use of convolutional layers as feature extraction steps allows us to exploit the sequential nature of time series, exhibiting higher representation performance than fully connected layers. The extracted clusters can encapture different semantic levels to either separate classes or extract outlying temporal signals. The application of this method to crop-types mapping enables the extraction of major crop-types within a scene, without supervision. In a labeled context, this method also allows for the extraction of outlying profiles which can lead to the discovery of mislabeled time series.

This paper aims to quantify the errors in the provided agricultural crop types, estimate the possible error rate in the available dataset, and propose a correction strategy. This quantification could establish a confidence criterion useful for decisions taken on this data or to have a better apprehension of the possible consequences of using this data in learning downstream functions such as classification. We consider two agricultural label errors: crop type mislabels and mis-split crops. To process and correct these errors, we design a two-step methodology. Using class-specific convolutional autoencoders applied to synthetic aperture radar (SAR) time series of free-to-use and temporally dense Sentinel-1 data, we detect out-of-distribution temporal profiles of crop time series, which we categorize as one out of the three following possibilities: crop edge confusion, incorrectly split crop areas, and potentially mislabeled crop. We then relabel crops flagged as mislabeled using an Otsu threshold-derived confidence criteria. We numerically validate our methodology using a controlled disruption of labels over crops of confidence. We then compare our methods to supervised algorithms and show improved quality of relabels, with up to 98% correct relabels for our method, against up to 91% for Random Forest-based approaches. We show a drastic decrease in the performance of supervised algorithms under critical conditions (smaller and larger amounts of introduced label errors), with Random Forest falling to 56% of correct relabels against 95% for our approach. We also explicit the trade-off made in the design of our method between the number of relabels, and their quality. In addition, we apply this methodology to a set of agricultural labels containing probable mislabels. We also validate the quality of the corrections using optical imagery, which helps highlight incorrectly cut crops and potential mislabels. We then assess the applicability of the proposed method in various contexts and scales and present how it is suitable for verifying and correcting farmers’ crop declarations.

This paper introduces a tool for explaining the latent space generated by applying convolutional autoencoders to satellite image time series, entitled Grad-SLAM. We rely on backpropagated gradient interpretation combined with network activation localization. We use the proposed formula for multiple layers of the encoder, then scale and merge the results to generate a single date contribution metric for the generation of the latent space. We illustrate the potential of this method with the study of the unsupervised classification of agricultural Sentinel-1 time series. We show that critical characterizing dates for unsupervised retrieval of a given class are conditioned by the crop type’s radiometric signature and class count. We also present how Grad-SLAM can be used to enhance the understanding of unsupervised classification confusion.

Modeling user purchase behavior is a critical challenge in display advertising systems, necessary for real-time bidding. The difficulty arises from the sparsity of positive user events and the stochasticity of user actions, leading to severe class imbalance and irregular event timing. Predictive systems usually rely on hand-crafted ''counter'' features, overlooking the fine-grained temporal evolution of user intent. Meanwhile, current sequential models extract direct sequential signal, missing useful event-counting statistics. We enhance deep sequential models with self-supervised pretraining strategies for display advertising. Especially, we introduce Abacus, a novel approach of predicting the empirical frequency distribution of user events. We further propose a hybrid objective unifying Abacus with sequential learning objectives, combining stability of aggregated statistics with the sequence modeling sensitivity. Experiments on two real-world datasets show that Abacus pretraining outperforms existing methods accelerating downstream task convergence, while hybrid approach yields up to +6.1% AUC compared to the baselines.

Apprentissage profond non-supervisé pour le suivi de végétation à partir de séries temporelles SAR en bande C : de l'agriculture aux forêts boréales: L'imagerie satellite SAR (Synthetic Aperture Radar) est un outil critique pour le suivi des forêts boréales. Ces dernières souffrent d'une couverture nuageuse la plupart du temps de l'année, impactant les techlogies optiques de surveillance, mais pas les technologies SAR. Cependant, ces mêmes forêts représentent un milieu complexe, notamment pour le SAR; leur étude requiert donc des outils complexes, comme l'apprentissage profond. Une masse de données est alors nécessaire pour cette pratique. De part leur couverture globale, et leur régularité temporelle, les satellites SAR Sentinel-1 sont donc idéaux. Cependant ces séries temporelles sont pour beaucoup dépourvues d'étiquettes, nécessitant alors une approche non-supervisée. Le réseau de neurones profond autoencodeur est alors un outil de choix. Son usage dans un contexte de séries temporelles SAR est inédit: une adaptation du modèle classique au traitement de donnée temporelle fut requise. De plus, son application directe au cas forestier reste difficile. Un premier essai est alors fait sur de la classification non-supervisée agricole, mené avec succès. Une détection et correction d'erreurs de label est même possible. Sa généralisation au cas forestier s'ensuit alors, avec la classification non-supervisée de couverture forestiere, et la détection non-supervisée de feu de forêts. Egalement fructueuse, ces applications donnent alors lieu à la prédiction de paramètres forestiers, à partir de séries temporelles. La comparaison est d'abord faite avec des mesures parcellaires, puis LiDAR. Leur succès pousse à l'étude de la corrélation des variations temporelles de Sentinel-1 avec la physiologie des forêts observées, le tout lié à la température, et une relation est établie.