Nicola Acito

Hyperspectral sensors provide high spectral resolution, enabling accurate material discrimination and effective target detection. However, their practical use is constrained by limited spatial resolution and high acquisition costs. This paper proposes a novel framework to enhance small-target detection in multispectral imagery by leveraging deep learning-based spectral reconstruction to generate high-resolution hyperspectral representations from multispectral inputs. Two state-of-the-art reconstruction networks, MST++ and MIRNet, are trained using paired multispectral–hyperspectral samples derived from AVIRIS-NG data through proper spectral response functions. To improve discriminative capability for the target of interest, a rapid, target-specific fine-tuning stage is introduced, allowing the models to adapt to spectral signatures that are poorly represented or absent in the original training data. Target detection is performed using a spectral signature-based detector applied to the reconstructed hyperspectral data. The proposed framework is evaluated in a real-world scenario involving known field-deployed targets and hyperspectral imagery acquired from an unmanned aerial vehicle. Experimental results demonstrate that the proposed approach significantly outperforms baseline detection applied directly to multispectral data. These findings underscore the effectiveness of spectral reconstruction for downstream tasks such as target detection, particularly in scenarios where hyperspectral data are expensive or unavailable.

The capability to generate realistic hyperspectral imagery plays a prominent role in applications to sensor and mission development as well as in the training of machine learning models. Yet, it is a challenging task due to the high dimensionality and complex spectral–spatial structure of the data. This paper proposes a novel unsupervised deep-learning framework for generating realistic hyperspectral imagery based on blind hyperspectral unmixing and denoising diffusion probabilistic models. First, the approach extracts both endmembers and abundance maps from hyperspectral data through a dictionary of hyperspectral unmixing algorithms. The extracted abundances are then used as inputs for a guided diffusion model, which serves as the generative framework with the goal of producing realistic synthetic abundance maps. Finally, the generation of synthetic hyperspectral images is accomplished by integrating the generated abundance maps with the extracted endmember set and by suitably conditioning the probabilistic formulation of the guided diffusion model as a function of the unmixing algorithms in the aforementioned dictionary. By combining a collection of blind linear unmixing techniques with the generative capabilities of diffusion models, the proposed methodology aims to address key challenges in simulating hyperspectral sensor outputs. The methodology was validated experimentally using real satellite hyperspectral imagery from the PRISMA mission of the Italian Space Agency. The results confirm the effectiveness of the approach in generating realistic synthetic hyperspectral images associated with various land-covers. The code is available at: https://github.com/martinapastorino/HSI_DDPM.

Agroforestry systems offer clear environmental and agronomic advantages, but their effect on plant–biotic stressor interactions remains poorly understood. Specifically, the shade from companion trees can create microclimates favorable to fungal diseases on herbaceous crops. This potential drawback may offset other benefits, highlighting the urgent need for advanced plant health monitoring in these systems. This study assessed the potential of hyperspectral reflectance to detect the single and combined effects of simulated tree shading and infection by the fungal pathogen Alternaria alternata on grain sorghum (Sorghum bicolor L. Moench) under rainfed field conditions. Sorghum was grown either under full light or 50% shading conditions. Half of the plots were artificially inoculated with an A. alternata spore suspension (2 × 10⁸ CFU mL⁻¹), while the others served as controls. Leaf and ground-canopy measurements were acquired with a full range spectroradiometer (VNIR-SWIR, 400–2,400 nm) and UAV imagery covered the VIS-NIR range (400–1,000 nm) before the onset of visible symptoms. Permutational multivariate analysis of variance of leaf and ground-canopy data revealed significant effects of shading (Sh), infection (Aa), and their interaction (p < 0.05), allowing early detection of infection two days before symptom appearance, while UAV data showed only singular significant effects. Partial least squares discriminant analysis accuracy reached 78% at the leaf level, 90% at the ground-canopy level, and 74% (Sh) and 75% (Aa) at the UAV scale. Furthermore, vegetation spectral indices derived from the spectra confirmed greater physiological stress in shaded and infected plants, consistent with disease incidence assessments. Our results establish scale-specific hyperspectral reflectance spectroscopy as a powerful, non-destructive technique for early plant health surveillance in agroforestry. This advanced optical sensing capability is poised to illuminate complex stressor interactions, marking a significant step forward for precision agroforestry management.

This paper presents a novel methodology for generating realistic abundance maps from hyperspectral imagery using an unsupervised, deep-learning-driven approach. Our framework integrates blind linear hyperspectral unmixing with state-of-the-art diffusion models to enhance the realism and diversity of synthetic abundance maps. First, we apply blind unmixing to extract endmembers and abundance maps directly from raw hyperspectral data. These abundance maps then serve as inputs to a diffusion model, which acts as a generative engine to synthesize highly realistic spatial distributions. Diffusion models have recently revolutionized image synthesis by offering superior performance, flexibility, and stability, making them well-suited for high-dimensional spectral data. By leveraging this combination of physically interpretable unmixing and deep generative modeling, our approach enables the simulation of hyperspectral sensor outputs under diverse imaging conditions--critical for data augmentation, algorithm benchmarking, and model evaluation in hyperspectral analysis. Notably, our method is entirely unsupervised, ensuring adaptability to different datasets without the need for labeled training data. We validate our approach using real hyperspectral imagery from the PRISMA space mission for Earth observation, demonstrating its effectiveness in producing realistic synthetic abundance maps that capture the spatial and spectral characteristics of natural scenes.

Hyperspectral (HS) satellite data are of considerable importance for applications such as environmental monitoring and precision agriculture, given the richness of the spectral information they contain. However, hyperspectral data typically exhibit limited spatial resolution and are less readily available than multispectral (MS) data. This study, which aims to simulate data with high spectral and spatial resolution, explores the use of attention-based spectral reconstruction (SR) techniques, specifically MST++, MIRNet, AWAN and Restormer, to derive HS data in the Visible Near InfraRed (VNIR) and Short-Wave InfraRed (SWIR) from MS imagery. High-resolution MS and HS image pairs are generated from AVIRIS-NG aerial data and employed for training procedures, thereby enabling the reconstruction of HS data that closely resembles the original measurements. The results indicate that SR techniques can considerably enhance the utility of existing MS datasets for HS-dependent applications. Such techniques can effectively be employed to synthesize high-resolution HS data from MS inputs, thereby facilitating the potential for developing a comprehensive end-to-end sensor simulator. This is particularly advantageous in the context of simulating data from a mission that has not yet become operational, as exemplified by the PRISMA-2G data, which could be simulated, for example, from Sentinel-2 data.

Satellite-based remote sensing images are essential for Earth surface analysis, serving diverse purposes in both civilian and military domains. Satellite images are used for analysis and decision making and are considered a reliable source of information. Recently, the field of image generation has developed increasingly sophisticated techniques, such as generative neural models, usually known as Generative Adversarial Neural Networks (GAN), to create synthetic images from scratch that appear almost real. Generative models have traditionally been applied to RGB or grayscale images and have been used for generating fake images of faces, animals, or objects. Currently, there are few studies regarding the application of GAN to multispectral satellite images. This work aims to test GAN models against the generation of multispectral satellite images, and in particular, the work explores the ability of the state-of-the-art StyleGAN3 model to produce high-quality synthetic Sentinel-2 images. The work delves into the configuration, training process, and evaluation of StyleGAN3 using custom Sentinel-2 datasets. StyleGAN3 results are compared with those provided by the proGAN model, the only GAN model tested so far on multispectral satellite data. Evaluation methods include visual inspection, spectral signature analysis, and a modified Fréchet Inception Distance for quantitative assessment. Results show that StyleGAN3 outperforms proGAN model exhibiting visually pleasing images. The quantitative comparison shows that StyleGAN3 provides the best results in terms of FID scores, in particular the improvement compared to proGAN increases as the spatial extent and spectral dimension of the generated images increases.

The rapid growth of hyperspectral satellite missions and the subsequent availability of hyperspectral images have encouraged the remote sensing community to investigate their potential in estimating the concentration of gases, as CH ₄ and CO ₂ , which are related to the greenhouse effect. Though satellite hyperspectral sensors are not specifically designed for this purpose, they are expected to complement more specific satellite missions, such as NASA’s OCO-2 and OCO-3, both in terms of enriched temporal sampling and improved spatial resolution. In this work, we present a new method to estimate the column-averaged dry-air mole fraction of CO ₂ from hyperspectral data on a per-pixel basis. The method, which is here tailored to PRISMA images, leverages the spectral radiance samples collected in the SWIR spectral region around the CO ₂ absorption band at 2000 nm. By assuming a linear model to describe the dependence of the observed radiance on the CO ₂ concentration, the estimation problem is reduced to matched filtering and can be effectively implemented in compliance with the low computational burden required to perform a pixel by pixel analysis. The performance of the presented method is investigated by means of a rigorous, physically based simulator that accurately reproduces the at-sensor radiance allowing one to check the validity of the assumptions and to assess the algorithm accuracy. The results show that the presented algorithm outperforms a benchmark CIBR based approach, which has been proposed in the literature to get fast per-pixel estimates of CO ₂ concentration.

Moving target detection (MTD) is a crucial task in computer vision applications. In this paper, we investigate the problem of detecting moving targets in infrared (IR) surveillance video sequences captured using a steady camera in a maritime setting. For this purpose, we employ robust principal component analysis (RPCA), which is an improvement of principal component analysis (PCA) that separates an input matrix into the following two matrices: a low-rank matrix that is representative, in our case study, of the slowly changing background, and a sparse matrix that is representative of the foreground. RPCA is usually implemented in a non-causal batch form. To pursue a real-time application, we tested an online implementation, which, unfortunately, was affected by the presence of the target in the scene during the initialization phase. Therefore, we improved the robustness by implementing a saliency-based strategy. The advantages offered by the resulting technique, which we called “saliency-aided online moving window RPCA” (S-OMW-RPCA) are the following: RPCA is implemented online; along with the temporal features exploited by RPCA, the spatial features are also taken into consideration by using a saliency filter; the results are robust against the condition of the scene during the initialization. Finally, we compare the performance of the proposed technique in terms of precision, recall, and execution time with that of an online RPCA, thus, showing the effectiveness of the saliency-based approach.

In satellite hyperspectral sensors, a standard procedure based on homogeneous reference sources is used to regularly update the map of defective pixels (DPs). Unfortunately, this procedure often fails to detect some DPs both because they may arise in the interval between two standard calibration steps and because they may be characterized by subtle or unexpected signal values. The resulting hyperspectral image is affected by residual nonuniformity noise which correlates in the along track direction. This noise source reduces the quality of hyperspectral products, such as classification, unmixing and material detection. In this paper, we present a new procedure to find the location of the residual DPs in the detection matrix and we also propose an effective method to estimate the missing radiance values inferring them from the image pixels by leveraging both spatial and spectral correlation. The procedure, here tailored to PRISMA hyperspectral images, is quite general and can be easily adapted to process images recorded by any satellite pushbroom hyperspectral sensor. The improved image quality yielded by the proposed procedure is first demonstrated qualitatively , by comparing the GRX maps on the original image with those obtained after detection of the DPs and correction of their radiance values. The image quality improvement is then quantified on a set of nine PRISMA images, recorded in an interval of about two years, using two ad-hoc defined indexes. The analysis is carried out separately for the VNIR and SWIR spectrometers of the PRISMA mission.

This article deals with the problem of improving the spatial resolution of hyperspectral (HS) data from the PRecursore IperSpettrale della Missione Applicativa (PRISMA) mission. For this purpose, higher spatial resolution data from the Sentinel-2 (S2) mission are exploited. Particularly, 10 S2 bands at 10 and 20 m spatial resolution are used to accomplish the PRISMA super-resolution (SR) task. The article presents a new end-to-end procedure, called PRISMA-SR, that starting from the S2 data and the low-resolution PRISMA image, provides a super-resolved image with a spatial resolution of 10 m and the same spectral resolution as the PRISMA HS sensor. The first step of the PRISMA-SR procedure consists in fusing S2 data at different spatial resolutions to obtain a synthetic MS image with 10 m spatial resolution and 10 spectral bands. Then, an unsupervised procedure is applied to coregister the fused S2 image and the PRISMA image. Finally, the two images at different spatial resolutions are properly combined in order to obtain the super-resolved HS image. Solutions for each step of the PRISMA-SR processing chain are proposed and discussed. Simulated data are used to show the effectiveness of the PRISMA-SR scheme and to investigate the impact on its performance of each step of the processing chain. Real S2 and PRISMA images are finally considered to provide an example of the application of the PRISMA-SR.