Explore the vast range of titles available.

Maritime traffic and fishing activities have accelerated considerably over the last decade, with a consequent impact on the environment and marine resources. Meanwhile, a growing number of ship-reporting technologies and remote-sensing systems are generating an overwhelming amount of spatio-temporal and geographically distributed data related to large-scale vessels and their movements. Individual technologies have distinct limitations but, when combined, can provide a better view of what is happening at sea, lead to effectively monitor fishing activities, and help tackle the investigations of suspicious behaviors in close proximity of managed areas. The paper integrates non-cooperative Synthetic Aperture Radar (SAR) Sentinel-1 images and cooperative Automatic Identification System (AIS) data, by proposing two types of associations: (i) point-to-point and (ii) point-to-line. They allow the fusion of ship positions and highlight “suspicious” AIS data gaps in close proximity of managed areas that can be further investigated only once the vessel—and the gear it adopts—is known. This is addressed by a machine-learning approach based on the Fast Fourier Transform that classifies single sea trips. The approach is tested on a case study in the central Adriatic Sea, automatically reporting AIS-SAR associations and seeking ships that are not broadcasting their positions (intentionally or not). Results allow the discrimination of collaborative and non-collaborative ships, playing a key role in detecting potential suspect behaviors especially in close proximity of managed areas.

One of the most challenging problems in precision agriculture is to correctly identify and separate crops from the soil. Current precision farming algorithms based on artificially intelligent networks use multi-spectral or hyper-spectral data to derive radiometric indices that guide the operational management of agricultural complexes. Deep learning applications using these big data require sensitive filtering of raw data to effectively drive their hidden layer neural network architectures. Threshold techniques based on the normalized difference vegetation index (NDVI) or other similar metrics are generally used to simplify the development and training of deep learning neural networks. They have the advantage of being natural transformations of hyper-spectral or multi-spectral images that filter the data stream into a neural network, while reducing training requirements and increasing system classification performance. In this paper, to calculate a detailed crop/soil segmentation based on high resolution Digital Surface Model (DSM) data, we propose the redefinition of the radiometric index using a directional mathematical filter. To further refine the analysis, we feed this new radiometric index image of about 3500 × 4500 pixels into a relatively small Convolution Neural Network (CNN) designed for general image pattern recognition at 28 × 28 pixels to evaluate and resolve the vegetation correctly. We show that the result of applying a DSM filter to the NDVI radiometric index before feeding it into a Convolutional Neural Network can potentially improve crop separation hit rate by 65%.

During the last decade vessel-position-recording devices, such as the Vessel Monitoring System and the Automatic Identification System, have increasingly given accurate spatial and quantitative information of industrial fisheries. On the other hand, small-scale fisheries (vessels below 12 m) remain untracked and largely unregulated even though they play an important socio-economic and cultural role in European waters and coastal communities and account for most of the total EU fishing fleet. The typically low-technological capacity of these small-scale fishing boats—for which space and power onboard are often limited—as well their reduced operative range encourage the development of efficient, low-cost, and low-burden tracking solutions. In this context, we designed a cost-effective and scalable prototypic architecture to gather and process positional data from small-scale vessels, making use of a LoRaWAN/cellular network. Data collected by our first installation are presented, as well as its preliminary processing. The emergence of a such low-cost and open-source technology coupled to artificial intelligence could open new opportunities for equipping small-scale vessels, collecting their trajectory data, and estimating their fishing effort (information which has historically not been present). It enables a new monitoring strategy that could effectively include small-scale fleets and support the design of new policies oriented to inform coastal resource and fisheries management.

The identification, classification and mapping of different plant communities and habitats is of fundamental importance for defining biodiversity monitoring and conservation strategies. Today, the availability of high temporal, spatial and spectral data from remote sensing platforms provides dense time series over different spectral bands. In the case of supervised mapping, time series based on classical vegetation indices (e.g., NDVI, GNDVI, …) are usually input characteristics, but the selection of the best index or set of indices (which guarantees the best performance) is still based on human experience and is also influenced by the study area. In this work, several different time series, based on Sentinel-2 images, were created exploring new combinations of bands that extend the classic basic formulas as the normalized difference index. Multivariate Functional Principal Component Analysis (MFPCA) was used to contemporarily decompose the multiple time series. The principal multivariate seasonal spectral variations identified (MFPCA scores) were classified by using a Random Forest (RF) model. The MFPCA and RF classifications were nested into a forward selection strategy to identify the proper and minimum set of indices’ (dense) time series that produced the most accurate supervised classification of plant communities and habitat. The results we obtained can be summarized as follows: (i) the selection of the best set of time series is specific to the study area and the habitats involved; (ii) well-known and widely used indices such as the NDVI are not selected as the indices with the best performance; instead, time series based on original indices (in terms of formula or combination of bands) or underused indices (such as those derivable with the visible bands) are selected; (iii) MFPCA efficiently reduces the dimensionality of the data (multiple dense time series) providing ecologically interpretable results representing an important tool for habitat modelling outperforming conventional approaches that consider only discrete time series.

The classification of plant associations and their mapping play a key role in defining habitat biodiversity management, monitoring, and conservation strategies. In this work we present a methodological framework to map Mediterranean forest plant associations and habitats that relies on the application of the Functional Principal Component Analysis (FPCA) to the remotely sensed Normalized Difference Vegetation Index (NDVI) time series. FPCA, considering the chronological order of the data, reduced the NDVI time series data complexity and provided (as FPCA scores) the main seasonal NDVI phenological variations of the forests. We performed a supervised classification of the FPCA scores combined with topographic and lithological features of the study area to map the forest plant associations. The supervised mapping achieved an overall accuracy of 87.5%. The FPCA scores contributed to the global accuracy of the map much more than the topographic and lithological features. The results showed that (i) the main seasonal phenological variations (FPCA scores) are effective spatial predictors to obtain accurate plant associations and habitat maps; (ii) the FPCA is a suitable solution to simultaneously express the relationships between remotely sensed and ecological field data, since it allows us to integrate these two different perspectives about plant associations in a single graph. The proposed approach based on the FPCA is useful for forest habitat monitoring, as it can contribute to produce periodically detailed vegetation-based habitat maps that reflect the “current” status of vegetation and habitats, also supporting the study of plant associations.

Sentiment analysis on social media such as Twitter is a challenging task given the data characteristics such as the length, spelling errors, abbreviations, and special characters. Social media sentiment analysis is also a fundamental issue with many applications. With particular regard of the tourism sector, where the characterization of fluxes is a vital issue, the sources of geotagged information have already proven to be promising for tourism-related geographic research. The paper introduces an approach to estimate the sentiment related to Cilento’s, a well known tourism venue in Southern Italy. A newly collected dataset of tweets related to tourism is at the base of our method. We aim at demonstrating and testing a deep learning social geodata framework to characterize spatial, temporal and demographic tourist flows across the vast of territory this rural touristic region and along its coasts. We have applied four specially trained Deep Neural Networks to identify and assess the sentiment, two word-level and two character-based, respectively. In contrast to many existing datasets, the actual sentiment carried by texts or hashtags is not automatically assessed in our approach. We manually annotated the whole set to get to a higher dataset quality in terms of accuracy, proving the effectiveness of our method. Moreover, the geographical coding labelling each information, allow for fitting the inferred sentiments with their geographical location, obtaining an even more nuanced content analysis of the semantic meaning.

In retail environments, understanding how shoppers move about in a store’s spaces and interact with products is very valuable. While the retail environment has several favourable characteristics that support computer vision, such as reasonable lighting, the large number and diversity of products sold, as well as the potential ambiguity of shoppers’ movements, mean that accurately measuring shopper behaviour is still challenging. Over the past years, machine-learning and feature-based tools for people counting as well as interactions analytic and re-identification were developed with the aim of learning shopper skills based on occlusion-free RGB-D cameras in a top-view configuration. However, after moving into the era of multimedia big data, machine-learning approaches evolved into deep learning approaches, which are a more powerful and efficient way of dealing with the complexities of human behaviour. In this paper, a novel VRAI deep learning application that uses three convolutional neural networks to count the number of people passing or stopping in the camera area, perform top-view re-identification and measure shopper–shelf interactions from a single RGB-D video flow with near real-time performances has been introduced. The framework is evaluated on the following three new datasets that are publicly available: TVHeads for people counting, HaDa for shopper–shelf interactions and TVPR2 for people re-identification. The experimental results show that the proposed methods significantly outperform all competitive state-of-the-art methods (accuracy of 99.5% on people counting, 92.6% on interaction classification and 74.5% on re-id), bringing to different and significative insights for implicit and extensive shopper behaviour analysis for marketing applications.

The mapping and monitoring of natural and semi-natural habitats are crucial activities and are regulated by European policies and regulations, such as the 92/43/EEC. In the Mediterranean area, which is characterized by high vegetational and environmental diversity, the mapping and monitoring of habitats are particularly difficult and often exclusively based on in situ observations. In this scenario, it is necessary to automate the generation of updated maps to support the decisions of policy makers. At present, the availability of high spatiotemporal resolution data provides new possibilities for improving the mapping and monitoring of habitats. In this work, we present a methodology that, starting from remotely sensed time-series data, generates habitat maps using supervised classification supported by Functional Data Analysis. We constructed the methodology using Sentinel-2 data in the Mediterranean Special Area of Conservation “Gola di Frasassi” (Code: IT5320003). In particular, the training set uses 308 field plots with 11 target classes (five forests, two shrubs, one grassland, one mosaic, one extensive crop, and one urban land). Starting from vegetation index time-series data, Functional Principal Component Analysis was applied to derive FPCA scores and components. In particular, in the classification stage, the FPCA scores are considered as features. The obtained results out-performed a previous map derived from photo-interpretation by domain experts. We obtained an overall accuracy of 85.58% using vegetation index time-series, topography, and lithology data. The main advantages of the proposed approach are the capability to efficiently compress high dimensional data (dense remote-sensing time series) providing results in a compact way (e.g., FPCA scores and mean seasonal time profiles) that: (i) facilitate the link between remote sensing with habitat mapping and monitoring and their ecological interpretation and (ii) could be complementary to species-based approaches in plant community ecology and phytosociology.

Accurate vegetation mapping is essential for monitoring biodiversity and managing habitats, particularly in the context of increasing environmental pressures and conservation needs. Ground truthing plays a crucial role in ensuring the accuracy of supervised remote sensing maps, as it provides the high-quality reference data needed for model training and validation. However, traditional ground truthing methods are labor-intensive, time-consuming and restricted in spatial coverage, posing challenges for large-scale or complex landscapes. The advent of drone technology offers an efficient and cost-effective solution to these limitations, enabling the rapid collection of high-resolution imagery even in remote or inaccessible areas. This study proposes an approach to enhance the efficiency of supervised vegetation mapping in complex landscapes, integrating Multivariate Functional Principal Component Analysis (MFPCA) applied to the Sentinel-2 time series with drone-based ground truthing. Unlike traditional ground truthing activities, drone truthing enabled the generation of large, spatially balanced reference datasets, which are critical for machine learning classification systems. These datasets improved classification accuracy by ensuring a comprehensive representation of vegetation spectral variability, enabling the classifier to identify the key phenological patterns that best characterize and distinguish different vegetation types across the landscape. The proposed methodology achieves a classification accuracy of 92.59%, significantly exceeding the commonly reported thresholds for habitat mapping. This approach, characterized by its efficiency, repeatability and adaptability, aligns seamlessly with key environmental monitoring and conservation policies, such as the Habitats Directive. By integrating advanced remote sensing with drone-based technologies, it offers a scalable and cost-effective solution to the challenges of biodiversity monitoring, enabling timely updates and supporting effective habitat management in diverse and complex environments.

In the rapidly evolving field of remote sensing, Deep Learning (DL) techniques have become pivotal in interpreting and processing complex datasets. However, the increasing reliance on these algorithms necessitates a robust ethical framework to evaluate their trustworthiness. This paper introduces a comprehensive ethical framework designed to assess and quantify the trustworthiness of DL techniques in the context of remote sensing. We first define trustworthiness in DL as a multidimensional construct encompassing accuracy, reliability, transparency and explainability, fairness, and accountability. Our framework then operationalizes these dimensions through a set of quantifiable metrics, allowing for the systematic evaluation of DL models. To illustrate the applicability of our framework, we selected an existing case study in remote sensing, wherein we apply our ethical assessment to a DL model used for classification. Our results demonstrate the model’s performance across different trustworthiness metrics, highlighting areas for ethical improvement. This paper not only contributes a novel framework for ethical analysis in the field of DL, but also provides a practical tool for developers and practitioners in remote sensing to ensure the responsible deployment of DL technologies. Through a dual approach that combines top-down international standards with bottom-up, context-specific considerations, our framework serves as a practical tool for ensuring responsible AI applications in remote sensing. Its application through a case study highlights its potential to influence policy-making and guide ethical AI development in this domain.