In this article, we propose a novel and simple automatic model based on multimodal anomaly feature learning in a residual space, aiming at solving the binary classification problem of temporal change detection (CD) between pairs of heterogeneous remote sensing images. The model starts by learning from image pairs the normal existing patterns in the before and after images to come up with a suitable representation of the normal (nonchange) class. To achieve this, we employ a stacked sparse autoencoder trained on a large number of temporal image features (training data) in an unsupervised manner. To classify pixels of new unseen image-pairs, the built anomaly detection model reconstructs the input from its representation in the latent space. First, the probe (new) image (i.e., the bitemporal heterogeneous image pair as the input request) is encoded in this compact normal space from a stacked hidden representation. The reconstruction error is computed using the L2 norm in what we call the residual normal space. In which, the nonchange patterns are characterized by small reconstruction errors as a normal class while the change patterns are quantified by high reconstruction errors categorizing the abnormal class. The dichotomic (changed/unchanged) classification map is generated in the residual space by clustering the reconstructed errors using a Gaussian mixture model. Experimental results on different real heterogeneous images, reflecting a mixture of imaging and land surface CD conditions, confirm the robustness of the proposed anomaly detection model.

Analysis of heterogeneous remote sensing image is a challenging and complex problem due to the fact that the local statistics of the data to be processed can be radically different. In this article, we present a novel and reliable unsupervised change detection (CD) method to analyze heterogeneous remotely sensed image pairs. The proposed method is based on an imaging modality-invariant operator that detects at different scale levels the differences in terms of high-frequency pattern of each structural region existing in the two heterogeneous satellite images. First, this new detector is based upon a dual-norm formulation that makes our underlying CD estimation particularly robust in terms of a sensitivity/specificity tradeoff. Second, the detection process, embedded in a multiresolution framework, allows us to estimate a robust similarity or difference map that is then filtered out by a superpixel-based spatially adaptive filter to further increase its reliability against noise. Finally, changes are then identified from this similarity map by a simple binary clustering process that also takes into account the spatial contextual information around each pixel. Experimental results involving different types of heterogeneous remotely sensed image pairs confirm the robustness of the proposed approach.

The proliferation of deepfake generation has become increasingly widespread. Current solutions for automatically detecting and classifying generated content require substantial computational resources, making them impractical for use by the average non-expert individual, particularly from edge computing applications. In this paper, we propose a series of techniques to accelerate the inference speed of deepfake detection on video data. We also draw inspiration from steganalysis approaches to expose deepfakes as any secret payloads encoded in the image. Furthermore, some key considerations were identified to significantly reduce the size of the core convolutional neural network. The experiment yielded competitive results when evaluated on two second-generation deepfake datasets, namely Celeb-DFv2 and DFDC, while requiring only a fraction of the typical computational cost and resources.

With the increasing development of advanced unmanned aerial vehicles (UAVs), communication between operators and these intelligent systems is becoming more stressful. For the safety of UAV flights, automatic psychological stress detection is becoming a key research topic for successful missions. Stress can be reliably estimated via some biological markers which are not appropriate in many cases of human-machine-interaction setups. In this paper, we propose a non-intrusive deep learning-based stress level estimation approach. The goal is to identify the region where the operator's emotional state projects in the space defined by the latent dimensional emotions of arousal and valence since the stress region is well delimited in this space. The proposed multimodal approach uses sequential temporal CNN and LSTM with an Attention Weighted Average layer in the vision modality. As a second modality, we investigate local and global descriptors such as Mel-frequency cepstral coefficients, i-vector embeddings as well as Fisher-vector encodings. The multimodal-fusion approach uses a strategy referred to as "late-fusion" that involves the combination of unimodal model outputs as inputs of the decision engine. Since we have to deal with more naturalistic behavior in operator-machine interaction contexts, the One minute Gradual Emotion Challenge dataset was used for predictive model validation.

Extracting and identifying objects in very high resolution imagery has been a popular research topic in remote sensing. Since the beginning of this decade, deep learning techniques have revolutionized computer vision providing significant performance gains compared to traditional “shallow” techniques in various challenging vision problems. The training of deep neural networks usually requires very large training datasets. The advantage of using deep features is to exploit already trained Convolutional Neural Networks (CNN) in order to produce high level features without the burden of having to train a CNN from scratch. In this paper, we are investigating the use of deep features for the detection of small objects (cars and individual trees) in high resolution Pleiades imagery. Preliminary results show good detection performance and are very encouraging for future applications.