Gray-box optimization employs Walsh decomposition to obtain non-linear variable dependencies and utilize them to propose masks of variables that have a joint non-linear influence on fitness value. These masks significantly improve the effectiveness of variation operators. In some problems, all variables are non-linearly dependent, making the aforementioned masks useless. We analyze the features of the real-world instances of such problems and show that many of their dependencies may have noise-like origins. Such noise-caused dependencies are irrelevant to the optimization process and can be ignored. To identify them, we propose extending the use of Walsh decomposition by measuring variable dependency strength that allows the construction of the weighted dynamic Variable Interaction Graph (wdVIG). wdVIGs adjust the dependency strength to mixed individuals. They allow the filtering of irrelevant dependencies and re-enable using dependency-based masks by variation operators. We verify the wdVIG potential on a large benchmark suite. For problems with noise, the wdVIG masks can improve the optimizer's effectiveness. If all dependencies are relevant for the optimization, i.e., the problem is not noised, the influence of wdVIG masks is similar to that of state-of-the-art structures of this kind.

Monitoring forest areas is of paramount importance to maintain environmental sustainability. The scalability of forest monitoring solutions is effectively offered by satellite imaging, where images of various modalities are acquired in orbit and cover large areas. However, building machine learning models for such downstream Earth observation (EO) tasks is challenging due to the limited amounts of ground-truth datasets. We tackle this issue and introduce an end-to-end deep learning pipeline to detect forest changes from Sentinel-2 time series of multispectral images (MSIs). It benefits from a foundation model (FM) fine-tuned over a small yet spatially diverse dataset. The experiments showed that not only does it outperform other deep models but also it requires minimal user intervention before the fine-tuning process.

Hyperspectral imaging brings important opportunities in a variety of fields due to the unprecedented amount of information it captures in numerous narrow and contiguous spectral bands. However, the high spectral and spatial dimensionality of hyperspectral images makes them challenging to transfer, store, and ultimately analyze, while only a subset of bands may be significant in specific downstream applications in Earth observation. In this article, we tackle this issue and introduce CANNIBAL---a band selection algorithm based on unsupervised clustering of inter-band dependencies captured in weighted Variable Interaction Graphs, which are a side-effect of the optimization performed by the Genetic Algorithm with Linkage Learning. We apply CANNIBAL to two downstream tasks of hyperspectral unmixing and segmentation. Our experimental study revealed that it outperforms other band selection algorithms and allows us to dramatically reduce the number of bands without negatively affecting the quality of downstream models. Finally, CANNIBAL offers a high level of flexibility, as it can be both parametric and non-parametric, depending on a use case.

Cardiovascular diseases remain one of the leading causes of death worldwide. Therefore, developing and validating automated tools to help identify high-risk patients are of paramount clinical utility. In this article, we tackle this task and introduce a genetic programming algorithm (called GIRAFFE) to build (deep) machine learning classification ensembles for arrhythmia classification from two-dimensional images of 12-lead electrocardiogram (ECG) tracings. GIRAFFE evolves the architecture, content, and fusion scheme of the ensemble, to obtain an accurate yet lightweight classification system. The experimental study performed over a large-scale dataset of ECG images revealed that our approach outperforms other ensemble methods and carefully fine-tuned deep models, elaborates compact heterogeneous ensembles, and does not require any user intervention hence it is easy to apply to other classification tasks.

Needle localization in ultrasound images is pivotal for the successful execution of ultrasound-guided core needle biopsies. Automating the needle detection process can decrease the procedure time and lead to a more precise diagnosis. In this article, we introduce an automatic method for detecting the core needle and determining its trajectory in 2D ultrasound images. In our approach, the Vision Transformer architecture, renowned for its self-attention mechanisms is used for needle detection and segmentation, and is followed by the analysis of the Radon transformed segmentation mask to identify the needle’s trajectory. The experiments, performed over two clinical datasets of more than 600 ultrasound images rigorously split into various training-test subsets and backed up with a variety of statistical analyses revealed that our approach offers highquality needle segmentation, and significantly outperforms other techniques in identifying the needle’s trajectory, with the trajectory localization errors reduced up to more than $5 \times$ when compared to the most competitive deep learning algorithm. We believe that our work may pave the way for more accurate and efficient ultrasoundguided procedures, ultimately improving patient outcomes.

Recent advances in remote sensing and artificial intelligence offer exciting opportunities in an array of fields, with precision agriculture being a notable use case. Here, estimating soil parameters from remotely-sensed hyperspectral imagery at a global scale can play a pivotal role in day-to-day operations, as it may help optimize agricultural management processes, hence positively affecting our planet. In this paper, we tackled the problem of estimating soil parameters from hyperspectral images and introduced heterogeneous regression ensembles for this task. They not only benefit from both classic and deep machine learning models but were also thoroughly investigated and fine-tuned in our rigorous experimental study, performed over a well-established HYPERVIEW benchmark dataset. The experiments showed that such heterogeneous ensembles outperform other techniques and offer a high level of model flexibility.

Current technological advancements bring exciting opportunities in hyperspectral image analysis across various Earth observation applications, with precision agriculture being one of their notable examples. However, gathering high-quality and representative ground-truth data for training supervised machine learners is extremely challenging in emerging use cases, as it is cost-inefficient and user-dependent. Thus, building (deep) machine learning models from very few training samples is of paramount practical importance. To tackle this issue, we propose a semi-supervised learning pipeline for elaborating deep learning models for multi-class hyperspectral image classification, while benefiting from the available unlabeled samples, leveraging pseudo-labeling and consistency regularization. Although our technique is model-agnostic, we exploit a deep residual network for classifying hyperspectral images according to the magnesium content in the imaged soil. The experimental study performed over a real-world dataset (HYPERVIEW) encompassing environmental tasks crucial to food sustainability showed that the proposed technique significantly outperforms the pre-trained models fine-tuned in a fully supervised way.

We investigate the runtime of quantum kernel estimation in the view of quantum kernel concentration effect. The study is performed for projected quantum kernel family evaluated on hyperspectral remote sensing data. The effect of exponential value concentration leads to the indistinguishability of the kernel matrix entries as the size of the quantum device grows. In order to prevent that, kernel values have to be estimated with a better precision. Increasing precision inevitably connects to an increasing number of circuit runs, which influences the runtime of quantum algorithm. This, in turn, frequently obstructs a possible advantage for quantum machine learning methods. We find that, against popular opinions, the effect of exponential value concentration does not rule out the utility of quantum kernel methods and the severity of the issue depends highly on the data used.

Bare soil detection is an important step in soil composition analysis, as it can prune the areas that should be excluded from more expensive processing aimed at extracting selected soil parameters from hyperspectral images acquired in orbit. This is of paramount importance for on-board applications, where hardware constraints of an edge device (a satellite), such as computational and memory requirements or energy consumption need to be considered while processing big data in space. In this paper, we present a simple yet effective bare soil detection algorithm exploiting vegetation indices that is ready for in-orbit deployment. Our experimental study performed over the airborne hyperspectral data shows that this approach can be robustly used for simulated bands, i.e., wide bands aggregating several narrow neighboring bands within the spectrum. Therefore, we can apply our technique to sensors with lower spectral resolution. Finally, it offers high-quality bare soil delineation reaching the Dice Index of 0.85.

An unprecedented amount of data generated in space missions triggers lots of practical challenges and concerns with its transfer, storage, and analysis. As Lunar and deep space missions emerge, we need to also face the challenges of distributed computing and big data analytics. In this paper, we outline these issues and discuss how to design and analyze Lunar data centers, being space data centers designed for distributed computing, and data analysis for (not only) Lunar missions. We investigate the opportunities and chances of such space architectures to lay the foundations for practical space data centers and real-life use cases.

Satellite imaging brings exciting opportunities in an array of fields, with precision agriculture being a notable example. Soil analysis at scale with the use of Earth observation satellites coupled with on-board and on-the-ground artificial intelligence algorithms offers actionable items that may be exploited by practitioners to optimize their operations, including the fertilization process. Here, bare soil detection is a pivotal step in the processing chain to limit the detailed analysis to the areas of interest. In this paper, we tackle this task with quantum-kernel support vector machines and verify the utility of quantum machine learning in practical Earth observation. Our experimental study, performed over a real-world hyperspectral scene, indicates that the proposed quantum-kernel models are competitive with well-established classical support vector machines, as well as with approaches based on thresholding spectral indices that are widely exploited in the field.

Maintaining proper water supply is crucial for agriculture and efficient plant cultivation. In-situ soil moisture measurements lack scalability, necessitating non-invasive systems to estimate soil moisture from remotely-sensed hyperspectral images (HSIs). We propose a root-zone soil moisture estimation system using deep architectures on HSIs acquired via unmanned aerial vehicles. Experiments conducted on a meticulously prepared dataset of in-situ measurements and HSIs, collected across multiple agronomic seasons with diverse plant varieties, soil profiles, and watering scenarios, revealed that our approach achieved a mean absolute error, mean squared error, and R2of 0.60, 0.64, and 0.80, outperforming sensor-based methods.

Abstract Designing routing schedules is a pivotal aspect of smart delivery systems. Therefore, the field has been blooming for decades, and numerous algorithms for this task have been proposed for various formulations of rich vehicle routing problems. There is, however, an important gap in the state of the art that concerns the lack of an established and widely-adopted approach toward thorough verification and validation of such algorithms in practical scenarios. We tackle this issue and propose a comprehensive validation approach that can shed more light on functional and non-functional abilities of the solvers. Additionally, we propose novel similarity metrics to measure the distance between the routing schedules that can be used in verifying the convergence abilities of randomized techniques. To reflect practical aspects of intelligent transportation systems, we introduce an algorithm for elaborating solvable benchmark instances for any vehicle routing formulation, alongside the set of quality metrics that help quantify the real-life characteristics of the delivery systems, such as their profitability. The experiments prove the flexibility of our approach through utilizing it to the NP-hard pickup and delivery problem with time windows, and present the qualitative, quantitative, and statistical analysis scenarios which help understand the capabilities of the investigated techniques. We believe that our efforts will be a step toward the more critical and consistent evaluation of emerging vehicle routing (and other) solvers, and will allow the community to easier confront them, thus ultimately focus on the most promising research avenues that are determined in the quantifiable and traceable manner.

Bringing autonomy on board edge devices is inevitable to accelerate the process of space exploration. Although there are various tasks that can be executed autonomously by such vehicles, detecting and segmenting rocks in on-board images of extraterrestrial landscapes is a critical step in the processing chain, as it can allow to navigate safely while avoiding collisions. We tackle this issue and introduce an end-to-end pipeline for building and validating resource-frugal machine learning techniques for this task, offering a high level of flexibility. Deploying such models on board edge devices poses numerous practical challenges, spanning across ensuring their memory and computational efficiency, and understanding their robustness against varying quality of acquired images. These aspects are often overlooked while building deep learning-powered on-board systems—we show that they can (and ultimately should) be a part of the deployment chain. Our extensive experimental study performed over several benchmark data sets shed more light on functional and non-functional capabilities of the investigated models, both in full-precision and compressed by quantisation, the latter delivering statistically same segmentation accuracy while being approximately 11× smaller. Additionally, we show that synthesised images can be utilised to quantify the robustness of deep learning models against on-board acquisition conditions which directly affect the quality of captured images—such simulations mimicking real-world acquisition settings could have a negative impact on the capabilities of the models trained over clean and high-quality image data. To ensure full reproducibility of this study, we made our implementation publicly available at https://github.com/danielmarek22/onboard-rock-segmentation.

The advent of machine learning (ML) has significantly advanced the recognition and translation of sign languages, bridging communication gaps for hearing-impaired communities. At the heart of these technologies is data labeling, crucial for training ML algorithms on a huge amount of consistently labeled data to achieve models that generalize well. The adoption of language-agnostic annotations is essential to connect different sign languages, as single-language databases often provide limited lexicon examples, insufficient for training robust ML algorithms. This study critically examines the Hamburg Sign Language Notation System (HamNoSys), which describes the signer’s initial position and body movements, in contrary to the meanings of glosses. Despite HamNoSys’s utility in standardizing transcriptions across various sign languages, our investigation uncovers inconsistencies within HamNoSys that may negatively impact the development of accurate and reliable ML models. By analyzing HamNoSys labels across five sign languages, we identified a lack of standardized annotation procedures and the complexities within HamNoSys that introduce biases and errors. Our findings underscore the urgent need for unified, standardized data annotation guidelines to enhance the accuracy and efficiency of sign language recognition technologies. This research highlights the importance of addressing annotation challenges and advocates for a comprehensive, diversified database to improve the generalization of ML models.

Bringing artificial intelligence on board Earth observation satellites unlocks unprecedented possibilities to extract actionable items from various image modalities at the global scale in real time. This is of paramount importance nowadays, as downlinking large amounts of imagery is not only prohibitively expensive but also time-consuming. However, building deep learning solutions that could be deployed on board an edge device is challenging due to the limited manually-annotated satellite datasets and hardware constraints of an edge device. This paper addresses these challenges through harnessing a blend of data-centric and model-centric approaches to build a well-generalizing yet efficient and resource-frugal deep learning model for multi-class satellite image classification in the few-shot learning settings. This integrated strategy is formulated to enhance classification performance while accommodating the unique demands of an image analysis chain on board OPS-SAT, a nanosatellite operated by the European Space Agency. The experiments performed over a real-world dataset of OPS-SAT images delves into the interactions between data- and model-centric techniques, underscores the significance of synthesizing artificial training data and emphasizes the value of ensemble learning. However, they also caution against negative transfer in domain adaptation. This study sheds light on effective model training strategies and highlights the multifaceted challenges inherent in deep learning for practical Earth observation, contributing insights to the field of satellite image classification within the constraints of nanosatellite operations. To ensure reproducibility of our study, the implementation is available at https://github.com/ShendoxParadox/Few-shot-satellite-image-classification-OPS-SAT.

Aortic stenosis is the most common primary valvular pathology requiring surgical or transcatheter intervention in developed countries. Quantification of aortic valve calcification with cardiac computed tomography (CCT) is used for assessment of aortic stenosis severity, disease progression and prediction of major cardiovascular events. The calcium deposits, however, commonly appear in different regions of the aorta and heart, leading to false-positive regions, and to an incorrectly calculated aortic valve calcium score. We tackle the issue of pruning such false-positive regions from CCT scans, and introduce a particle swarm optimization (PSO) algorithm for this task. In our approach, PSO optimizes the radius while benefiting from the evolved position of a sphere which would embrace those calcifications that are positioned near the aortic valve. Our experimental study, performed over 30 non-contrast CCT scans, showed that our results are in strong agreement with the experienced human reader, and indicate the potential of PSO in data-driven pruning of false-positive calcifications which are positioned in other parts of the aorta and heart. Additionally, PSO outperformed a geometrical-based approach for this task.

Localization of a biopsy needle in ultrasound (US) images is an important medical image analysis task, as it may help clinicians reduce the risk of damaging the tissue surrounding the cancer and spreading cancerous cells. Despite numerous studies dedicated to segmenting the needle from US, virtually all of them build upon the strong assumption that the needle is present in the image, which does not hold in clinical settings. We address this research gap and propose an end-to-end machine learning approach for biopsy needle detection in US images. The rigorous experimental study revealed that our approach delivers high-quality and fast operation, while offering a high level of flexibility—not only does it allow to update all blocks of the pipeline, but also to build a detection cascade for multi-scale analysis which dramatically reduces the number of sub-images undergoing classification, hence speeds up the detection process.

The amount of data captured in the emerging satellite missions has been continuously growing. Thus, developing resource-efficient predictive models is of paramount importance in an array of onboard space applications, where downlinking the data for further analysis is extremely costly or impossible. In such scenarios, we should extract actionable items on board an edge device, using e.g., a machine learning model. Reducing the model’s complexity which may be significant in deep learning algorithms is not only about fitting a full-size neural net into resource-restrictive hardware, but it may result in decreasing the latency and energy consumption. We tackle this issue and exploit knowledge distillation to elaborate a simpler and memory-efficient version of the large-capacity learner, aiming to preserve the large model quality. The experimental study shows that knowledge distillation may not only improve the classification capability of the original model, but can also dramatically (up to 425×) reduce its size for the on-board classification of Mars imagery.

The rapid growth of the space industry is creating an increasing amount of data in orbit, which at the current time is not matched yet by a corresponding increase in data download capacity. Using data analytics at the edge, i.e., data processing on-board satellites, holds the promise to significantly mitigate this problem. Newly available and highly efficient artificial intelligence (AI) hardware accelerators and other off-the-shelf compute-hardware has already been successfully exploited for demonstrating satellite on-board deep learning. Aggregating such components and technologies at scale and introducing resource sharing concepts beyond individual spacecrafts, will yield the equivalent of a space data center—a space-based system that will collect, process, store, and relay data from own sensors or from "client" satellites and work in orchestration with other SDCs in a network. In this paper, we outline new opportunities for such data- and compute-sharing schemes in space and how current limitations of existing systems can be overcome.

Super-resolution is aimed at enhancing image spatial resolution and it has been intensively explored for many years. The recent advancements, underpinned with deep learning, also include techniques developed specifically for hyper-spectral data. However, most of the emerging methods are validated in application-independent scenarios, which often rely on an unrealistic experimental setup—the reconstruction is performed from simulated low-resolution images (degraded from an original image) with the goal of inverting the degradation process and restoring the original image. This leads to over-optimistic assessment of super-resolution capabilities and limits their practical applications. In this paper, we demonstrate task-based validation for different types of hyperspectral PRISMA image super-resolution, including pan-sharpening, fusion of multispectral and hyperspectral data, as well as single-image super-resolution. The obtained results reported in the paper are encouraging and they help better understand the value of super-resolved PRISMA images.

The optimization of Kernel-Target Alignment (TA) has been recently proposed as a way to reduce the number of hardware resources in quantum classifiers. It allows to exchange highly expressive and costly circuits to moderate size, task oriented ones. In this work we propose a simple toy model to study the optimization landscape of the Kernel-Target Alignment. We find that for underparameterized circuits the optimization landscape possess either many local extrema or becomes flat with narrow global extremum. We find the dependence of the width of the global extremum peak on the amount of data introduced to the model. The experimental study was performed using multispectral satellite data, and we targeted the cloud detection task, being one of the most fundamental and important image analysis tasks in remote sensing.

sifier effectively deployed in an array of pattern recognition and classification tasks. In this work, we consider extending classic SVMs with quantum kernels and applying them to satellite data analysis. The design and implementation of SVMs with quantum kernels (hybrid SVMs) is presented. It consists of the Quantum Kernel Estimation (QKE) procedure combined with a classic SVM training routine. The pixel data are mapped to the Hilbert space using ZZ-feature maps acting on the parameterized ansatz state. The parameters are optimized to maximize the kernel target alignment. We approach the problem of cloud detection in satellite image data, which is one of the pivotal steps in both on-the-ground and on-board satellite image analysis processing chains. The experiments performed over the benchmark Landsat-8 multi-spectral dataset revealed that the simulated hybrid SVM successfully classifies satellite images with accuracy on par with classic SVMs.

Hyperspectral unmixing is one of the most challenging tasks in the analysis of such data. There have been an array of algorithms proposed for this problem so far, but they are virtually always verified using random sampling, where training and test examples are drawn from the same image. Since such samples are spatially correlated and may be positioned close to each other, random sampling can induce the training-test information leak in the techniques that exploit spatial information during the unmixing process. We want to raise the attention of the community about this validation flaw in the context hyperspectral unmixing. We introduce the algorithm for unbiased validation of the unmixing techniques through splitting hyperspectral images into training and test samples that do not suffer from the training-test information leak. The experiments showed that the widely-used random sampling verification strategy leads to overly optimistic conclusions concerning the algorithm’s performance. This problem was mitigated with the proposed approach which allows us to rigorously validate unmixing techniques.

Detecting methane in satellite hyperspectral images (HSIs) can play a key role in environmental monitoring, as taking timely actions to reduce its emission and handle (unexpected) super emitters is of paramount importance. We tackle this issue and propose a machine learning pipeline for this task, with the ultimate goal of deploying it on board a satellite. Such solutions can offer global scalability, and they can act as a smart data prioritization step, as only those HSIs which contain methane can be downlinked for further analysis. However, the on-board deployment induces additional practical challenges—such algorithms should be resource-frugal, and should effectively operate on the target image data which may not be available during their development, since the satellite is not in orbit yet. Our experimental study revealed that the data-driven approaches can effectively detect methane in original airborne HSIs, as well as in HSIs emulating the target sensor and generated through data-level simulations.

In this paper, we present our study focused on applying a vision transformer-based pansharpening technique to enhance PRISMA satellite hyperspectral data. The PRISMA mission, launched by the Italian Space Agency, captures hyperspectral images comprising visible and near infra-red, as well as short-wave infra-red channels. By integrating the panchromatic image of high spatial resolution with the hyperspectral data of high spectral resolution, the pansharpening process consists in producing spatially-enhanced hyperspectral imagery. Our research involves modifying and adapting the state-of-the-art HyperTransformer architecture to effectively process real-life PRISMA data. The evaluation of our model’s performance utilizes PRISMA L2D data, encompassing simulated low-resolution data and real-life data. We employ quantitative metrics and visual examination to assess the results. We also highlight the importance of choosing right PRISMA data processing level for the pansharpening process. The proposed pansharpening model successfully enhances PRISMA data for practical applications, contributing to the advancement of Earth observation techniques.

Multispectral Sentinel-2 images are a valuable source of Earth observation data, however spatial resolution of their spectral bands limited to 10 m, 20 m, and 60 m ground sampling distance remains insufficient in many cases. This problem can be addressed with super-resolution, aimed at reconstructing a high-resolution image from a low-resolution observation. For Sentinel-2, spectral information fusion allows for enhancing the 20 m and 60 m bands to the 10 m resolution. Also, there were attempts to combine multitemporal stacks of individual Sentinel-2 bands, however these two approaches have not been combined so far. In this paper, we introduce DeepSent—a new deep network for super-resolving multitemporal series of multispectral Sentinel-2 images. It is underpinned with information fusion performed simultaneously in the spectral and temporal dimensions to generate an enlarged multispectral image. In our extensive experimental study, we demonstrate that our solution outperforms other state-of-the-art techniques that realize either multitemporal or multispectral data fusion. Furthermore, we show that the advantage of DeepSent results from how these two fusion types are combined in a single architecture, which is superior to performing such fusion in a sequential manner. Importantly, we have applied our method to super-resolve real-world Sentinel-2 images, enhancing the spatial resolution of all the spectral bands to 3.3 m nominal ground sampling distance, and we compare the outcome with very high-resolution WorldView-2 images. We will publish our implementation upon paper acceptance, and we expect it will increase the possibilities of exploiting super-resolved Sentinel-2 images in real-life applications.

Improving agricultural practices through exploiting the recent imaging and machine learning advancements plays a key role nowadays to ensure sustainable food security, and to help us deal with the climate change. Quantifying soil parameters can lead to optimizing the fertilization process but it is cumbersome, time-consuming and difficult to scale, as it requires performing in-situ soil measurements that are later analyzed in the laboratory settings. In the HYPER-VIEW challenge, we aim at automating the soil analysis thanks to the utilization of hyperspectral images that capture very detailed information about the scanned objects in hundreds of contiguous hyperspectral bands. Such imagery can be effectively analyzed using an array of classical and deep machine learning approaches. Also, the AI techniques can be deployed on-board the imaging satellites— it opens new doors related to the scalability of the solution. The winners of the challenge will be offered a unique opportunity to run their proposed solution in orbit, on-board the Intuition-1 satellite, equipped with a hyperspectral imager and on-board AI capabilities.

Magnetic resonance imaging is the modality of choice in brain tumors to diagnose and monitor the patients with such lesions. However, its manual analysis is user-dependent and cumbersome, and is affected by significant intra- and inter-rater variability which hampers the objective assessment of the progression of the disease. In this work, we tackle this issue and introduce an end-to-end pipeline to not only segment the brain tumors in a fully-automated and reproducible way and calculate their volumetric characteristics, but also to benefit from radiomic features extracted from the tumorous tissue for classifying the scan as low- or high-grade gliomas. Our extensive experiments revealed that the automatically calculated volumetric measurements are in almost perfect agreement with human readers, and that the ensemble classifiers utilizing radiomic features extracted from whole-tumor areas deliver high-quality classification for a range of segmentation approaches. To make the experiments fully reproducible and to encourage other research groups to exploit our end-to-end pipeline, we made the implementation open-sourced.

Automatic localization of the biopsy needle in ultrasound images has become an important medical image analysis task, because it can directly translate to reducing the risk of damage to the tissues surrounding the lesion and spreading cancer cells. Although the algorithms to tackle this problem has been emerging at a steady pace, we lack a standardized way of validating them. In this paper, we tackle this issue and raise the attention of the research community to this experimental flaw which is especially important for the applications that can be ultimately deployed in the clinical settings. Our study, which involves a range of different training-test dataset splits performed over heterogeneous image data, showed that the incorrectly designed validation procedures can easily lead to overly optimistic conclusions concerning the abilities of such algorithms. We believe that our efforts will be an important step toward designing reproducible, rigorous, and fair approach for confronting the needle localization techniques in an unbiased way.

Cloud detection is one of the most important image pre-processing steps that can be performed on-board satellites. It may allow us to reduce the amount of data to analyze or downlink by pruning the cloudy areas, or to make the satellites more autonomous through data-driven image acquisition re-scheduling of the areas obscured by clouds. Thus, building the cloud detection algorithms that can be ultimately deployed in orbit became an important research avenue. In this paper, we investigate the robustness of the fully-convolutional neural networks for cloud detection against the atmospheric conditions that resemble real acquisition settings of the Intuition-1 mission. Our experiments, performed over the original and simulated Landsat-8 images, with the latter reflecting target conditions, shed more light on the performance of deep models and showed how can we verify their robustness in Earth observation tasks for which real images do not exist yet.

Multi-image super-resolution is a branch of super-resolution reconstruction techniques aiming to resolve a set of lowresolution images into a high-resolution one. Unlike singleimage super-resolution, its goal is to fuse information embedded in different images depicting the same scene, usually captured at different times. Such data combination contains more high-resolution information than a single image, thus allowing for more accurate reconstruction results. One of the most critical challenges is to prepare training data for multi-image super-resolution since only a few datasets are available here, especially for satellite imaging applications. For this reason, many studies are conducted using simulated low-resolution images, but the results obtained for real-life data are often unsatisfactory. To overcome this problem, we propose a new semi-simulated approach of creating lowresolution images for training that resemble real-life ones much more accurately. We also investigate the performance of selected deep learning models trained with simulated and semi-simulated datasets and we show that the latter achieve better results when applied to real-world images.

The recent advances in Earth observation and artificial in-telligence allow us to improve the agricultural management practices through effectively exploiting the spectral, spatial, and temporal characteristics of the area of interest captured by satellite images. In this paper, we tackle the problem of extracting high-resolution (2.5-meter) cultivated land maps from Sentinel-2 multispectral images, and propose a machine learning algorithm for this task. It aggregates the spectral, spatial, and temporal features of the upsampled images, and is independent from the number of observations captured for a given scene. The experimental results, performed within the framework of the Enhanced Sentinel-2 Agriculture chal-lenge show that our technique manifests high generalization abilities over the unseen data and elaborates high-quality cul-tivated land maps. Finally, utilizing this algorithm led us to taking the $6^{\text{th}}$ place in the aforementioned challenge.

Super-resolution reconstruction consists in generating a high-resolution image from a single low-resolution image or multiple images presenting the same area of interest. Existing state-of-the-art approaches to single-image and multi-image super-resolution are based on deep learning that requires large amounts of training data. They are commonly obtained by simulating low-resolution images from an original image treated as a high-resolution reference, but such simulation may not reflect the real-life operating conditions. Therefore, a serious obstacle in deploying super-resolution in real-world cases results from the lack of training data that would encompass real low-resolution images coupled with a real high-resolution reference. In this paper, we propose a new data augmentation technique underpinned with learning the relation between high and low resolution. This helps reduce the requirements concerned with the amount of real-life data necessary to train a super-resolution network, while providing higher-quality data for training, compared with the simulated low-resolution images. Our initial experimental results reported in the paper confirm that the proposed approach is suitable for multi-image super-resolution.

Maintaining farm sustainability through optimizing the agricultural management practices helps build more planet-friendly environment. The emerging satellite missions can acquire multi- and hyperspectral imagery which captures more detailed spectral information concerning the scanned area, hence allows us to benefit from subtle spectral features during the analysis process in agricultural applications. We introduce an approach for extracting 2.5m cultivated land maps from 10m Sentinel-2 multispectral image series which benefits from a compact graph convolutional neural network. The experiments indicate that our models not only outperform classical and deep machine learning techniques through delivering higher-quality segmentation maps, but also dramatically reduce the memory footprint when compared to U-Nets (almost 8k trainable parameters of our models, with up to 31M parameters of U-Nets). Such memory frugality is pivotal in the missions which allow us to uplink a model to the AI-powered satellite once it is in orbit, as sending large nets is impossible due to the time constraints.

Hyperspectral unmixing remains one of the most challenging tasks in the analysis of such data. Deep learning has been blooming in the field and proved to outperform other classic unmixing techniques, and can be effectively deployed onboard Earth observation satellites equipped with hyperspectral imagers. In this letter, we follow this research pathway and propose a multi-branch convolutional neural network that benefits from fusing spectral, spatial, and spectral-spatial features in the unmixing process. The results of our experiments, backed up with the ablation study, revealed that our techniques outperform others from the literature and lead to higher-quality fractional abundance estimation. Also, we investigated the influence of reducing the training sets on the capabilities of all algorithms and their robustness against noise, as capturing large and representative ground-truth sets is time-consuming and costly in practice, especially in emerging Earth observation scenarios.

Since its outbreak reported in late 2019 in Wuhan, China, the novel coronavirus disease (COVID-19) has been the major challenge across the globe, affecting virtually all aspects of our lives. To effectively manage the pandemic, we need fast, non-invasive, and precise routines for detecting active COVID-19 cases. Although there exist deep learning approaches for detecting COVID-19 in medical image data, their generalization abilities remain unknown. We tackle this issue and introduce deep ensembles that benefit from a wide range of architectural advances, alongside a new fusing approach to deliver accurate predictions. Also, we evolve their content to not only accelerate the inference but also to boost the classification performance. Our experiments, performed on a number of datasets of chest X-ray images, show that the proposed technique renders high-quality classification and generalizes well over a variety of test scans.

Super-resolution consists in reconstructing a high-resolution image from single or multiple low-resolution observations. Deep learning has been reported extremely successful for single-image super-resolution, but its applications to the multiple-image scenarios are limited due to the challenges that arise from feeding a network with a stack of images with sub-pixel translations. In this paper, we introduce Magnet—a new graph neural network that benefits from representing the input low-resolution images as a graph. This enables us to exploit the sub-pixel shifts among the input images while preserving the original low-resolution pixel values for feature extraction and information fusion. Despite a relatively simple architecture, Magnet outperforms the state-of-the-art methods for multiple-image super-resolution, and due to the flexible graph representation, it allows for using a variable number of low-resolution images for reconstruction.

Cloud detection is an important pre-processing step that allows us to significantly reduce the amount of satellite imagery which should undergo further processing. In this paper, we investigate the impact of training set selection on the abilities of fully-convolutional neural networks for this task. Our experiments, performed over a range of Landsat-8 satellite images, show that the performance of deep models can substantially vary for different training samples, especially in the case of challenging scenes, such as those capturing snowy areas.

Super-resolution (SR) reconstruction is a common term for techniques aimed at generating a high-resolution image from a single low-resolution image or multiple images showing the same scene. Multiple-image SR benefits from data fusion which allows for more accurate reconstruction of the underlying high-resolution information. Deep learning is extensively used for single-image SR, but its application to multiple-image SR is much less explored. Recently, several deep networks were proposed to enhance Proba-V images, and in this paper, we focus on employing them to super-resolve the Sentinel-2 images. In particular, we investigate the influence of the training data, including real and simulated low-resolution images, on the final SR outcome. Also, we make the simulated data publicly available.

Hyperspectral unmixing allows us to estimate the endmember abundances in each pixel of an input hyperspectral image. Although there exist deep learning-powered end-to-end methods for this task, the lack of labeled ground-truth data is a challenging problem which makes the adoption of such techniques extremely difficult in emerging practical use cases where the ground truth is costly to capture. In this paper, we investigate and quantify the impact of the training set size on the quality of unmixing provided by deep learning models of conceptually different architectures.

Although deep learning is gaining more widespread use in hyperspectral image analysis, it is challenging to train high-capacity models in a supervised way—ground-truth sets are expensive to obtain, and they are practically always extremely imbalanced. To deal with the problem of missing ground-truth data, its high dimensionality and potential redundancy, we introduce a novel unsupervised feature learning technique to extract discriminative features from the original data. It exploits recurrent neural network-based asymmetric autoencoders (AEs) to learn the compressed representation of unlabeled data, and can elaborate both spectral and spectral–spatial features. Our extractors can be incorporated into the unsupervised segmentation pipeline—they can be followed by any clustering algorithm. The experiments revealed that our approaches deliver high-quality segmentation without any prior class labels, and are one order of magnitude faster than 3-D convolutional AEs. Our algorithms outperform or work on par with other approaches while allowing for significant data reduction.

The use of deep-learned features has recently allowed for improving the performance of skin detection and segmentation. In particular, fully-convolutional U-Nets, proposed for segmenting medical images, occurred to be extremely effective here. However, the spatial context, which is rather narrow for U-Nets, may be more important for skin segmentation than for segmenting other image structures. We propose Skinny-a lightweight U-Net-based architecture that extends the range of multi-scale analysis. The results of our experiments indicate that Skinny outperforms the state-of-the-art skin segmentation techniques, rendering the F-score of 92.3% and 94.9% for the ECU and HGR datasets, respectively.

Support vector machines (SVMs) are a supervised learning technique that can be applied in both binary and multi-class classification and regression tasks. SVMs seamlessly handle continuous and categorical variables. Their training is, however, both time- and memory-costly for large training data, and selecting an incorrect kernel function or its hyperparameters leads to suboptimal decision hyperplanes. In this paper, we introduce a memetic algorithm for evolving SVM training sets with adaptive radial basis function kernels to not only make the deployment of SVMs easier for emerging big data applications, but also to improve their generalization abilities over the unseen data. We build upon two observations: first, only a small subset of all training vectors, called the support vectors, contribute to the position of the decision boundary, hence the other vectors can be removed from the training set without deteriorating the performance of the model. Second, selecting different kernel hyperparameters for different training vectors may help better reflect the subtle characteristics of the space while determining the hyperplane. The experiments over almost 100 benchmark and synthetic sets showed that our algorithm delivers models outperforming both SVMs optimized using state-of-the-art evolutionary techniques, and other supervised learners.

Super-resolution reconstruction is a process aimed at enhancing image spatial resolution. To evaluate the quality of super-resolution, the reconstruction outcome is compared with a ground-truth reference image, and the dissimilarity between them is commonly treated as a determinant of the reconstruction quality. While this is straightforward for simulated data, it becomes more challenging for real-world scenarios, in which reference images and the reconstruction inputs are acquired using different imaging sensors. In such cases, the dissimilarity also results from other factors concerned with different sensor characteristics. In a recently organized Proba-V Super Resolution Challenge, the reconstruction quality was assessed using a modified peak signal-to-noise ratio which compensates for small shifts and global changes in the brightness. In the study reported here, we investigate a number of image similarity metrics to verify their robustness against different levels of distortions applied to Proba-V images. We expect that the reported results will help in choosing appropriate metrics while developing new super-resolution solutions aimed at real-world scenarios.

Hyperspectrometers capture detailed information about the scanned objects, but they are susceptible to the presence of various kinds of noise. In this work, we quantify the impact of different levels of Gaussian and impulsive noise on the abilities of a spectral convolutional neural network applied for hyperspectral image segmentation. Our experiments showed that the noise may degrade the classification accuracy of the spectral convolutional nets, and can be a serious obstacle in deploying such models on-board a satellite, where the presence of noise is inevitable.

Hyperspectral images provide detailed information about the scanned objects, as they capture their spectral characteristics within a large number of wavelength bands. Classification of such data has become an active research topic due to its wide applicability. In this paper, we introduce a new spectral-spatial convolutional neural network, benefitting from a battery of data augmentation techniques which help deal with a real-life problem of lacking ground-truth training data. Our experiments showed that the proposed method works in real time and outperforms other spectral-spatial algorithms.

Super-resolution (SR) reconstruction is a process aimed at enhancing the spatial resolution of images, either from a single observation, based on the learned relation between low and high resolution, or from multiple images presenting the same scene. SR is particularly important, if it is not feasible to acquire images at the desired resolution, while there are single or many observations available at lower resolution - this is inherent to a variety of remote sensing scenarios. Recently, we have witnessed substantial improvement in single-image SR attributed to the use of deep neural networks for learning the relation between low and high resolution. Importantly, deep learning has not been widely exploited for multiple-image super-resolution, which benefits from information fusion and in general allows for achieving higher reconstruction accuracy. In this letter, we introduce a new approach to combine the advantages of multiple-image fusion with learning the low-to-high resolution mapping using deep networks. The results of our extensive experiments indicate that the proposed framework outperforms the state-of-the-art SR methods.

Hyperspectral image analysis has become an important topic widely researched by the remote sensing community. Classification and segmentation of such imagery help understand the underlying materials within a scanned scene since hyperspectral images convey detailed information captured in a number of spectral bands. Although deep learning has established the state-of-the-art in the field, it still remains challenging to train well-generalizing models due to the lack of ground-truth data. In this letter, we tackle this problem and propose an end-to-end approach to segment hyperspectral images in a fully unsupervised way. We introduce a new deep architecture which couples 3-D convolutional autoencoders with clustering. Our multifaceted experimental study-performed over the benchmark and real-life data-revealed that our approach delivers high-quality segmentation without any prior class labels.

Data augmentation helps improve generalization capabilities of deep neural networks when only limited ground-truth training data are available. In this letter, we propose test-time augmentation of hyperspectral data, which is executed during the inference rather than before the training of deep networks. We introduce two augmentation techniques, which can be applied at both training time and test time. The experiments revealed that our augmentations boost generalization of deep models and work in real time, and the test-time approach can be combined with training-time techniques to enhance the classification accuracy.

Deep learning has established the state of the art in multiple fields, including hyperspectral image analysis. However, training large-capacity learners to segment such imagery requires representative training sets. Acquiring such data is human-dependent and time-consuming, especially in earth observation scenarios, where the hyperspectral data transfer is very costly and time-constrained. In this letter, we show how to effectively deal with a limited number and size of available hyperspectral ground-truth sets and apply transfer learning for building deep feature extractors. Also, we exploit spectral dimensionality reduction to make our technique applicable over hyperspectral data acquired using different sensors, which may capture different numbers of hyperspectral bands. The experiments, performed over several benchmarks and backed up with statistical tests, indicated that our approach allows us to effectively train well-generalizing deep convolutional neural nets even using significantly reduced data.

Ensemble methods train multiple classifiers and combine their outcomes to improve the generalization ability of a learning system. Although there exist methods for automatic creation of ensembles, they require heavy fine-tuning and the impact of their hyper-parameters on the optimization process very often remains unknown. In this paper, we propose a genetic algorithm for evolving classification ensembles. It is coupled with a pre-processing routine which deals with potential data imbalance and redundancy within the training and/or feature sets. Our experimental study, performed over binary and multi-class benchmark datasets and backed up with statistical analysis, revealed that the proposed technique outperforms other algorithms for building multiple classifier systems. It also helped understand the impact of various (hyper-)parameters of our method on its exploration/exploitation capabilities.

Data augmentation helps improve generalization of deep neural networks, and can be perceived as implicit regularization. It is pivotal in scenarios in which the amount of ground-truth data is limited, and acquiring new examples is costly and time-consuming. This is a common problem in medical image analysis, especially tumor delineation-in this paper, we focus on brain-tumor segmentation from magnetic resonance imaging (MRI), and propose a novel augmentation technique which exploits image registration to benefit from subtle spatial and/or tissue characteristics captured within the training set. We used a set of MRI scans of 44 low-grade glioma patients, augmented it using the proposed technique, and exploited it to train U-Net-based deep networks. The results show that our augmentation delivers statistically important boost of performance without sacrificing inference speed.

The capabilities of super-resolution (SR) reconstruction (i.e., techniques for enhancing image spatial resolution) have been boosted recently by the use of deep convolutional neural networks. For SR, they are learned using huge training sets composed of original images, each of which is coupled with a low-resolution counterpart. In this paper, we explore how the SR performance depends on the procedure employed to obtain the training data. Up to date, this has not been given much attention-commonly, bicubic downsampling is used. Our extensive experimental study indicates that the training data characteristics have a large impact on the reconstruction accuracy, and the widely-adopted approach is not the most effective for dealing with satellite images. Overall, we argue that developing better training data preparation routines may be pivotal in making SR suitable for real-world applications.

Hyperspectral satellite imaging has been gaining enormous research attention due to the latest advancements in the sensor technology, and the amount of information it conveys. However, its efficient analysis, transfer, and storage are still big practical issues which need to be endured in on-board applications. In this paper, we verify if the simulated multispectral data (with significantly smaller number of bands) can be segmented with accuracy as high as obtained over its corresponding original hyperspectral imagery. Our experimental study, backed up with statistical tests, revealed that it is possible to dramatically decrease the transfer and storage requirements of the original hyperspectral data by simulating its multispectral counterpart without adversely affecting the classification performance of popular supervised learners.

Hyperspectral satellite imaging attracts enormous research attention in the remote sensing community, and hence, automated approaches for precise segmentation of such imagery are being rapidly developed. In this letter, we share our observations on the strategy for validating hyperspectral image segmentation algorithms currently followed in the literature, and show that it can lead to overoptimistic experimental insights. We introduce a new routine for generating segmentation benchmarks and use it to elaborate ready-to-use hyperspectral training-test data partitions. They can be utilized for fair validation of new and existing algorithms without any training-test data leakage.

Image segmentation is a pivotal task in hyperspectral image processing. Usually, it is performed in a setting where neither the environment nor the hardware pose an obstacle. If segmentation methods are deployed on embedded hardware, constraints in electrical power or computational resources can lead to the degradation of their performance, as well as to an increase in their inference time. This paper presents a supervised classification algorithm for hyperspectral image segmentation based on quantized convolutional neural networks. Compared with its non-quantized variant and other state-of-the-art approaches, the proposed method obtains accurate segmentation with a much lower memory size. By exploiting hardware instructions that are well suited to embedded devices (with or without GPU), energy requirements are reduced while high accuracy is maintained.

Deep neural networks (DNNs) have proven to be effective at solving challenging problems, but their success relies on finding a good architecture to fit the task. Designing a DNN requires expert knowledge and a lot of trial and error, especially as the difficulty of the problem grows. This paper proposes a fully automatic method with the goal of optimizing DNN topologies through memetic evolution. By recasting the mutation step as a series of progressively refined educated local-search moves, this method achieves results comparable to best human designs. Our extensive experimental study showed that the proposed memetic algorithm supports building a real-world solution for segmenting medical images, it exhibits very promising results over a challenging CIFAR-10 benchmark, and works very fast. Given the ever growing availability of data, our memetic algorithm is a very promising avenue for hands-free DNN architecture design to tackle emerging classification tasks.

Over the last years, research has been placed at the core of numerous software products resulting from the ubiquitous presence of multimedia devices in our lives. Besides the need of acquiring specialized hardware, performing reliable and reproducible research requires the exploitation of specialized tools that are not traditionally present in the agile enterprise ecosystem. This might be a factor that ultimately draws many organizations away from introducing research into their software products. In this work, we propose a hands-free workflow that exploits widely available tools in the enterprise, like source control (SC) and continuous integration (CI) systems. We demonstrate that this workflow acts as a single-step end-to-end solution, maximizing the usage of the available hardware, and ensuring the repeatability of the performed experiments. The probability of human errors is minimized by automating all file transfers, and feedback is provided at the end of every trial with the location of the results. Generated artifacts are automatically archived, alongside the initial conditions of the experiment allowing for its full recreation. Although in our solution we exploit Git and Jenkins, this workflow can also be implemented with any of the SC and CI tools typically available in the enterprise.

Deep neural networks (DNNs) have achieved unprecedented success in a wide array of tasks. However, the performance of these systems depends directly on their hyper-parameters which often must be selected by an expert. Optimizing the hyper-parameters remains a substantial obstacle in designing DNNs in practice. In this work, we propose to select them using particle swarm optimization (PSO). Such biologically-inspired approaches have not been extensively exploited for this task. We demonstrate that PSO efficiently explores the solution space, allowing DNNs of a minimal topology to obtain competitive classification performance over the MNIST dataset. We showed that very small DNNs optimized by PSO retrieve promising classification accuracy for CIFAR-10. Also, PSO improves the performance of existing architectures. Extensive experimental study, backed-up with the statistical tests, revealed that PSO is an effective technique for automating hyper-parameter selection and efficiently exploits computational resources.

Solving the pickup and delivery problem with time windows (PDPTW) is a vital research topic due to its NP-hardness and its numerous practical applications. In this paper, we propose an island-model parallel memetic algorithm for minimizing the distance in the PDPTW. In this algorithm, the processes execute the same memetic algorithm and co-operate to guide the optimization efficiently. An extensive experimental study revealed that the MPI implementation of the proposed approach retrieves very high-quality routing schedules. The analysis is coupled with appropriate statistical tests.

It has been reported in many works on skin detection and segmentation from color images that skin color models suffer from low specificity and high variance of the skin color, and this problem can be addressed by conforming the skin model to a presented scene. Here, we introduce a new hybrid adapta tion system which combines two strategies, namely (i) adaptation from a detected facial region and (ii) a self-adaptive scheme that creates a local model based on the response obtained using the global one. As a result of this hybrid adaptation, we obtain a local skin color model and we use it to extract seeds for the geodesic distance transform that determines the boundaries of skin regions. The results of our extensive experimental study confirm that the proposed algorithm outperforms several state-of-the-art methods, as well as our earlier adaptive skin detectors.

This paper presents an adaptive memetic algorithm to solve the vehicle routing problem with time windows (VRPTW). It is a well-known NP-hard discrete optimization problem with two objectives—to minimize the number of vehicles serving a set of geographically dispersed customers, and to minimize the total distance traveled in the routing plan. Although memetic algorithms have been proven to be extremely efficient in solving the VRPTW, their main drawback is an unclear tuning of their numerous parameters. Here, we introduce the adaptive memetic algorithm (AMA-VRPTW) for minimizing the total travel distance. In AMA-VRPTW, a population of solutions evolves with time. The parameters of the algorithm, including the selection scheme, population size and the number of child solutions generated for each pair of parents, are adjusted dynamically during the search. We propose a new adaptive selection scheme to balance the exploration and exploitation of the solution space. Extensive experimental study performed on the well-known Solomon’s and Gehring and Homberger’s benchmark sets confirms the efficacy and convergence capabilities of the proposed AMA-VRPTW. We show that it is very competitive compared with other state-of-the-art techniques. Finally, the influence of the proposed adaptive schemes on the AMA-VRPTW behavior and performance is investigated in a thorough sensitivity analysis. This analysis is complemented with the two-tailed Wilcoxon test for verifying the statistical significance of the results.

Solving the job shop scheduling problem (JSP) is a vital research topic due to its wide practical applicability. It is an NP-hard discrete optimization problem which was transformed into a plethora of variants reflecting other real-life scenarios. In this paper, we propose an adaptive memetic algorithm (MA) to solve the JSP. It consists in determining a schedule for completing jobs (divided into operations) on a set of available machines. MAs, which are the hybrids of genetic algorithms and refinement procedures, were shown to be very efficient in tackling complex problems in many fields of science and engineering. In the proposed algorithm (AMXMA), a number of children are generated for each pair of parents to exploit them intensively. We keep three solution representations (if necessary) within each individual to avoid the necessity of transforming one representation into another required by various crossover operators. In addition, we introduce the adaptive selection scheme which is dynamically controlled on the fly to effectively balance the exploration and exploitation of the search space. An extensive experimental study performed on a widely-used benchmark set of problems with various sizes shows that AMXMA is extremely efficient in terms of the computation time and allows for fast convergence to very high-quality solutions. We show that AMXMA is highly competitive compared with other state-of-the-art algorithms.

In this paper, we propose a parallel guided ejection search algorithm to minimize the fleet size in the NP-hard pickup and delivery problem with time windows. The parallel processes co-operate periodically to enhance the quality of results and to accelerate the convergence of computations. The experimental study shows that the parallel algorithm retrieves very high-quality results. Finally, we report 13 (22% of all considered benchmark tests) new world's best solutions.

In this paper we propose a new memetic algorithm (MASVM) for fast and efficient selection of a valuable training set for support vector machines (SVMs). This is a crucial step especially in case of large and noisy data sets, since the SVM training has high time and memory complexity. The majority of state-of-the-art methods exploit the data geometry analysis, both in the input and kernel space. Although evolutionary algorithms have been proven to be very efficient for this purpose, they have not been extensively studied so far. Here, we propose a new method employing an adaptive genetic algorithm enhanced by some refinement techniques. The refinements are based on utilizing a pool of the support vectors identified so far at various steps of the algorithm. Extensive experimental study performed on the well-known benchmark, real-world and artificial data sets clearly confirms the efficacy, robustness and convergence capabilities of the proposed approach, and shows that it is competitive compared with other state-of-the-art techniques.

In this paper we propose a new method for skin detection in color images which consists in spatial analysis using the introduced texture-based discriminative skin-presence features. Color-based skin detection has been widely explored and many skin color modeling techniques were developed so far. However, efficacy of the pixel-wise classification is limited due to an overlap between the skin and non-skin pixels reported in many color spaces. To increase the discriminating power of the skin classification schemes, textural and spatial features are often exploited for skin modeling. Our contribution lies in using the proposed discriminative feature space as a domain for spatial analysis of skin pixels. Contrary to existing approaches, we extract the textural features from the skin probability maps rather than from the luminance channel. Presented experimental study confirms that the proposed method outperforms alternative skin detection techniques, which also involve analysis of textural and spatial features.

This paper introduces a new method for adaptive skin detection in color images combined with spatial analysis of skin pixels. It has been reported in many works that adaptation of a skin color model to a particular image may decrease the false positives, however the false negatives are considerably high unless a local model is combined with the global one. Another possibility for improvement is to analyze spatial properties of the pixels classified as skin, but this operation strongly depends on the seed extraction technique. Our contribution lies in using a local dynamic skin model learned from the detected faces to extract seeds for the spatial analysis. We present an extensive experimental study confirming that our method outperforms alternative skin detection techniques.

This paper introduces a new parallel algorithm (PA) for fast hand shape classification. This problem is challenging as a hand is characterized by a high number of degrees of freedom. Our objective is to design and implement a robust algorithm suitable for real-time applications. We show how the analysis time can be decreased, together with the increase of the classification accuracy, by the means of parallelization. Also, we propose to combine the shape contexts approach with the appearance-based techniques to increase the efficacy of the PA. An extensive experimental study confirms the effectiveness of the proposed PA compared with other state-of-the-art methods.