The transition to renewable energy sources has generated increased interest in accurate forecasting methods for green hydrogen production. This work aims to predict green hydrogen production from solar energy generation using machine and deep learning models. The proposed approach includes data preprocessing from different sites, implementing AI-driven techniques, running hyperparameter optimization, and extrapolating these parameters for training with real data from other sites. In addition, the Time Delay Embedding technique is applied to capture the temporal dependencies of the data for supervised learning. The methods Random Forest, Support Vector Regression, Extreme Gradient Boosting, and Long Short-Term Memory are trained and properly tuned. The results demonstrate that Extreme Gradient Boosting model achieves the highest accuracy, with all models adapting well to data from the distinct stations analyzed. The extrapolation of optimized hyperparameters proved efficient, reducing computational costs without compromising accuracy. In conclusion, the proposed approach is robust and viable for predicting the production of green hydrogen at different locations, making it a scalable solution for supporting clean energy planning.

Deforestation remains a critical global environmental concern, requiring effective monitoring approaches. This letter presents a novel attention-powered encoder–decoder neural network designed to address the key challenges in deforestation mapping, including scale heterogeneity, temporal dynamics, and computational efficiency. The proposed framework integrates a modified YOLOv8 backbone, spatial attention (SA) mechanisms, and a conjugated Dice–Focal loss function to enhance sensitivity to small- and large-scale deforestation patterns in temporal remote sensing (RS) data. An extensive battery of tests was conducted using two datasets from the Amazon region, exploring both single-image and image-pair inputs under varying contextual and class balance conditions. The results attest to substantial improvements in accuracy and computational efficiency compared to 13 deep learning (DL) methods, establishing the proposed model as effective in deforestation monitoring scenarios, where accuracy, scalability, and computational cost are simultaneously critical.

Change detection is a fundamental task that involves assessing changes in a given region over multiple time periods. It has been widely applied across various fields, including monitoring deforestation, urban expansion, and natural disaster analysis. In this article, we address the critical and complex issue of automatically identifying types of changes in land cover using remotely sensed imagery. While conventional unsupervised change detection methods typically focus on comparing pairs of images and making a binary decision between “change” and “nonchange,” our approach tackles the challenge of analyzing long image series and identifying the kind of change. Under this condition, the unsupervised change detection process allows for a more informative identification of the land cover dynamics. Moreover, our approach transforms input data to a new representation, capturing the target’s spectral response changes over time. Through the utilization of stochastic distances and an optimized thresholding scheme, areas exhibiting minimal spectral response variance are classified as unchanged, effectively distinguishing them from regions undergoing modifications. Next, by applying autocorrelation analysis, regions exhibiting temporal modifications are segregated into periodic (i.e., seasonal) and aperiodic (i.e., permanent) change cases. Experimental validation using both simulated and real-world remote sensing image series demonstrates the effectiveness of the proposed approach.

Deep Learning (DL) has become one of the key approaches for dealing with many challenges in medical imaging, which includes lung segmentation in Computed Tomography (CT). The use of seeded segmentation methods is another effective approach to get accurate partitions from complex CT images, as they give users autonomy, flexibility and easy usability when selecting specific targets for measurement purposes or pharmaceutical interventions. In this paper, we combine the accuracy of deep contour leaning with the versatility of seeded segmentation to yield a semi-automatic framework for segmenting lung CT images from patients affected by COVID-19. More specifically, we design a DL-driven approach that learns label diffusion maps from a contour detection network integrated with a label propagation model, used to diffuse the seeds over the CT images. Moreover, the trained model induces the diffusion of the seeds by only taking as input a marked CT-scan, segmenting hundreds of CT slices in an unsupervised and recursive way. Another important trait of our framework is that it is capable of segmenting lung structures even in the lack of well-defined boundaries and regardless of the level of COVID-19 infection. The accuracy and effectiveness of our learned diffusion model are attested to by both qualitative as well as quantitative comparisons involving several user-steered segmentations methods and eight CT data sets containing different types of lesions caused by COVID-19.

In ophthalmology and vision science applications, the process of registering a pair of fundus images, captured at different scales and viewing angles, is of paramount importance to support the diagnosis of diseases and routine eye examinations. Aiming at addressing the retina registration problem from the Deep Learning perspective, in this paper we introduce an end-to-end framework capable of learning the registration task in a fully unsupervised way. The designed approach combines Convolutional Neural Networks and Spatial Transformation Network into a unified pipeline that takes a similarity metric to gauge the difference between the images, thus enabling the image alignment without requiring any ground-truth data. Once the model is fully trained, it can perform one-shot registrations by just providing as input the pair of fundus images. As shown in the validation study, the trained model is able to successfully deal with several categories of fundus images, surpassing other recent techniques for retina registration.

Classifying targets in satellite images is a nontrivial task which requires dealing with a large number of undesirable elements such as clouds, building shadows and other unexpected objects. Among these, a commonly found element refers to artificially inserted post-processing objects like textual content, as the added text usually takes the form of watermarks, sensor specifications, street and place location names, etc. Manually selecting text segments is tedious, time-consuming, and requires the familiarity with image editing tools to precisely delineate these writing areas. Therefore, in this paper, a new automatic approach for detecting textual elements in satellite images is presented. Our approach combines cartoon-texture decomposition, thresholding-based rules, morphological operations, and connected component analysis into a fully automated and concise framework. Experiments on real satellite images and comparisons against well-established text detection methods demonstrate the high accuracy and low false-positive rate achieved by our approach when detecting textual content.

Seeded segmentation methods have gained a lot of attention due to their good performance in fragmenting complex images, easy usability and synergism with graph-based representations. These methods usually rely on sophisticated computational tools whose performance strongly depends on how good the training data reflect a sought image pattern. Moreover, poor adherence to the image contours, lack of unique solution, and high computational cost are other common issues present in most seeded segmentation methods. In this work we introduce Laplacian Coordinates, a quadratic energy minimization framework that tackles the issues above in an effective and mathematically sound manner. The proposed formulation builds upon graph Laplacian operators, quadratic energy functions, and fast minimization schemes to produce highly accurate segmentations. Moreover, the presented energy functions are not prone to local minima, i.e., the solution is guaranteed to be globally optimal, a trait not present in most image segmentation methods. Another key property is that the minimization procedure leads to a constrained sparse linear system of equations, enabling the segmentation of high-resolution images at interactive rates. The effectiveness of Laplacian Coordinates is attested by a comprehensive set of comparisons involving nine state-of-the-art methods and several benchmarks extensively used in the image segmentation literature.

Change detection is a topic of great interest in remote sensing. A good similarity metric to compute the variations among the images is the key to high-quality change detection. However, most existing approaches rely on the fixed threshold values or the user-provided ground truth in order to be effective. The inability to deal with artificial objects such as clouds and shadows is a significant difficulty for many change-detection methods. We propose a new unsupervised change-detection framework to address those critical points. The notion of homogeneous regions is introduced together with a set of geometric operations and statistic-based criteria to characterize and distinguish formally the change and nonchange areas in a pair of remote sensing images. Moreover, a robust and statistically well-posed family of stochastic distances is also proposed, which allows comparing the probability distributions of different regions/objects in the images. These stochastic measures are then used to train a support-vector-machine-based approach in order to detect the change/nonchange areas. Three study cases using the images acquired with different sensors are given in order to compare the proposed method with other well-known unsupervised methods.

Remote Sensing has been of paramount importance to capture features of interest from the Earth's surface. In this context, extraction algorithms and classification methods can be applied to capture the response of the electromagnetic spectrum in different image bands in order to find out what kind of feature is more predominant in a satellite image. Therefore, in this paper, two different feature detection approaches are evaluated and compared: the first one based on mathematical morphology filtering, while the second one is built as a semi-supervised classification approach which applies the well-established Bhattacharyya distance. Mathematical Morphology is an important field of Digital Image Processing which aims at detecting and extracting image objects based on the set theory and convolution processes. Bhattacharyya distance is one of the most effective statistical tools for classifying image zones. In our experiments, both approaches are compared against each other by inspecting their classification results for two airport areas, which includes both visual as well as quantitative evaluations.

Optimal transport has emerged as a promising and useful tool for supporting modern image processing applications such as medical imaging and scientific visualization. Indeed, the optimal transport theory enables great flexibility in modeling problems related to image registration, as different optimization resources can be successfully used as well as the choice of suitable matching models to align the images. In this paper, we introduce an automated framework for fundus image registration which unifies optimal transport theory, image processing tools, and graph matching schemes into a functional and concise methodology. Given two ocular fundus images, we construct representative graphs which embed in their structures spatial and topological information from the eye's blood vessels. The graphs produced are then used as input by our optimal transport model in order to establish a correspondence between their sets of nodes. Finally, geometric transformations are performed between the images so as to accomplish the registration task properly. Our formulation relies on the solid mathematical foundation of optimal transport as a constrained optimization problem, being also robust when dealing with outliers created during the matching stage. We demonstrate the accuracy and effectiveness of the present framework throughout a comprehensive set of qualitative and quantitative comparisons against several influential state-of-the-art methods on various fundus image databases.

Image registration is an important pre-processing step in several computer vision applications, being crucial in medical imaging systems where patients are examined and diagnosed almost exclusively by images. For fundus images, in which microscopic differences are significant to better support medical decisions, an accurate registration is imperative. Historically, geometric transformations derived from quadratic models have been widely used as a benchmark to perform registration on fundus images, but in this paper, we demonstrate that quadratic and other high-order mappings are not necessarily the best choices for this purpose, even for well-established state-of-the-art registration methods. From a novel overlapping metric designed to determine the best image transformation that maximizes the registration accuracy, we improve the assertiveness of several methods of the literature while still preserving the same computational burden initially reached by those methods.

Kernel functions have revolutionized theory and practice in the field of pattern recognition, especially to perform image classification. Besides giving rise to nonlinear variants of the well-known support vector machine (SVM), these functions have also been successfully used to classify nonvectorial data (e.g., graphs and collection of sets), in which customized metrics are created to precisely measure the similarity among such contextual data entities. This letter introduces two context-inspired kernel functions as new SVM-driven methods for remote sensing image classification. In contrast to the existing SVM-based approaches that assume only multiattribute vectors as representative features in a high-dimensional space, the proposed models formally establish comparisons between the entire sets of context-given data, thus employing these contextual measurements to drive the classification. More precisely, stochastic distances as well as hypothesis tests are conveniently handled and “kernelized” to build our models. A complete battery of experiments involving both remote sensing and real-world images is conducted to validate the performance of the proposed kernels against various well-established SVM-based methods.

Existing algorithms for building layouts from geometric primitives are typically designed to cope with requirements such as orthogonal alignment, overlap removal, optimal area usage, hierarchical organization, among others. However, most techniques are able to tackle just a few of those requirements simultaneously, impairing their use and flexibility. In this work we propose a novel methodology for building layouts from geometric primitives that concurrently addresses a wider range of requirements. Relying on multidimensional projection and mixed integer optimization, our approach arranges geometric objects in the visual space so as to generate well structured layouts that preserve the semantic relation among objects while still making an efficient use of display area. Moreover, scalability is handled through a hierarchical representation scheme combined with navigation tools. A comprehensive set of quantitative comparisons against existing geometry-based layouts and applications on text, image, and video data set visualization prove the effectiveness of our approach.

Image inpainting is a challenging topic in computer vision that seeks to recover the natural aspect of an image where data has been partially damaged or occluded by undesired objects. A common drawback not addressed by most inpainting methodologies is that the user must manually provide the inpainting mask as input data to the method. Selecting the inpainting mask is tedious, time consuming and it often requires artistic skills to precisely determine the mask. In this work we design a new tool that allows users to easily select the desirable mask. The proposed framework combines the high-adherence on image contours of the Laplacian Coordinates segmentation approach with the efficiency of a recent inpainting technique that unifies anisotropic diffusion, inner product-based filling order mechanism and exemplar-based completion. The user can interact with the object that he/she intends to edit by stroking small parts of the object so as to proceed with the segmentation and inpainting task. Our comparisons show that the proposed framework has good performance in terms of applicability and effectiveness when compared against other existing techniques in the literature.

Seed-based image segmentation methods have gained much attention lately, mainly due to their good performance in segmenting complex images with little user interaction. Such popularity leveraged the development of many new variations of seed-based image segmentation techniques, which vary greatly regarding mathematical formulation and complexity. Most existing methods in fact rely on complex mathematical formulations that typically do not guarantee unique solution for the segmentation problem while still being prone to be trapped in local minima. In this work we present a novel framework for seed-based image segmentation that is mathematically simple, easy to implement, and guaranteed to produce a unique solution. Moreover, the formulation holds an anisotropic behavior, that is, pixels sharing similar attributes are kept closer to each other while big jumps are naturally imposed on the boundary between image regions, thus ensuring better fitting on object boundaries. We show that the proposed framework outperform state-of-the-art techniques in terms of quantitative quality metrics as well as qualitative visual results.

Internet users are very familiar with the results of a search query displayed as a ranked list of snippets. Each textual snippet shows a content summary of the referred document (or webpage) and a link to it. This display has many advantages, for example, it affords easy navigation and is straightforward to interpret. Nonetheless, any user of search engines could possibly report some experience of disappointment with this metaphor. Indeed, it has limitations in particular situations, as it fails to provide an overview of the document collection retrieved. Moreover, depending on the nature of the query--for example, it may be too general, or ambiguous, or ill expressed--the desired information may be poorly ranked, or results may contemplate varied topics. Several search tasks would be easier if users were shown an overview of the returned documents, organized so as to reflect how related they are, content wise. We propose a visualization technique to display the results of web queries aimed at overcoming such limitations. It combines the neighborhood preservation capability of multidimensional projections with the familiar snippet-based representation by employing a multidimensional projection to derive two-dimensional layouts of the query search results that preserve text similarity relations, or neighborhoods. Similarity is computed by applying the cosine similarity over a "bag-of-words" vector representation of collection built from the snippets. If the snippets are displayed directly according to the derived layout, they will overlap considerably, producing a poor visualization. We overcome this problem by defining an energy functional that considers both the overlapping among snippets and the preservation of the neighborhood structure as given in the projected layout. Minimizing this energy functional provides a neighborhood preserving two-dimensional arrangement of the textual snippets with minimum overlap. The resulting visualization conveys both a global view of the query results and visual groupings that reflect related results, as illustrated in several examples shown.