The study presents a novel approach to forest canopy height estimation using the ResUNet model, a deep learning architecture, combined with Sentinel-2 satellite imagery. Focused on the diverse forest landscapes of southern Finland, this research leverages LiDAR data provided by the Finnish Forest Center for model training and validation. The ResUNet model, an advancement over the conventional U-Net model, is tailored to address the complexities of forest canopy structures and the nuances of satellite data. Our findings demonstrate that the ResUNet model outperforms existing methods in terms of F1 score and Jaccard coefficient, indicating superior accuracy in canopy height estimation. The study highlights the potential challenges in remote sensing and environmental analysis, such as the interpretation of canopy heights in mixed vegetation areas and the impact of image variability across different times of the year. Despite these challenges, the visual examination of the model's output aligns closely with expert interpretations and LiDAR-based ground truth data, affirming the model's effectiveness. This research not only underscores the efficacy of the ResUNet model in environmental remote sensing tasks but also suggests potential future directions, including the use of multi-temporal data and the integration of different spectral bands for enhanced accuracy. The ResUNet model emerges as a promising tool for forest canopy height estimation, offering implications for broader forest management and ecological monitoring systems.

This study investigates the efficacy of two feed-forward neural networks (MLP and RBF) and an SVM algorithm in monitoring grasslands. Employing six input parameters — Sentinel-1 SAR intensity, texture, temperature, precipitation, global radiation, and evapotranspiration — we aimed to determine their influence on model accuracy. Our research highlights the superior stability and accuracy of MLP and RBF NNs over SVM, with MLP NNs marginally outperforming RBF NNs. The most effective results were achieved using a comprehensive input set. Nevertheless, reducing the number of inputs led to a decrease in data dimensionality and consequently, model accuracy. The research indicates potential accuracy improvements with additional factors like soil type and grassland management. Conducting field studies concurrent with image acquisition is crucial for understanding the diverse grassland conditions and their effects on scattering mechanisms.

More than 70 years after the end of World War II, remaining military material and ordnance is still a problem in many European countries. Unexploded bombs left in the ground pose a particularly serious hazard, especially in heavily bombarded areas. Given the danger of accidental triggering of unexploded bombs, careful examination is required before construction or other activities can be permitted. For this purpose, bomb disposal teams analyze historical aerial reconnaissance imagery from World War II to evaluate risks. This is done mostly manually via 2D and 3D visual inspection and GIS. For each examination, a large number of images has to be evaluated. A (semi-)automation of procedures is therefore of great interest.We present preliminary results of an end-to-end process using Convolutional Neural Networks (CNNs) we propose for automatic detection and localization of bomb craters and impacts in historical aerial imagery. The goal is to create a service that can help bomb disposal teams to assess risks fast and reliably through a synergistic use of human and AI capacities.

Measuring the predictability and complexity of time series using entropy is an essential tool for designing and controlling a nonlinear system in the remote sensing field. However, the existing methods have some drawbacks related to their strong dependence on method parameters. To overcome these difficulties, this study proposes a new method for estimating the two- dimensional neural network entropy (NNetEn2D) for evaluating the regularity or predictability of images using the LogNNet neural network model.

Forest Canopy mapping is one of the significant factors in the evaluation of forest status and is an essential indicator of possible management interventions. Forest canopy cover, also known as canopy coverage or crown cover, is defined as the proportion of the forest floor covered by the vertical projection of the tree crowns. Estimation of forest canopy cover has recently become an important part of forest inventories. Using satellite imagery to estimate crown coverage has a long history. In this paper, an attempt has been made to demonstrate the potential of Pulse Coupled Neural Networks (PCNN) model for forest canopy mapping.

Accurate and rapid mapping of fire scares is fundamental to support fire management, account for environmental loss, define planning strategies and monitor the restoration of vegetation in post fire management. The new MSI sensor aboard Sentinel-2 satellites provides spectral information recorded in the near-infrared and shortwave infrared spectral region, opening the way to - applying bands ratio-based indices for burned area mapping. We have used a difference of Normalized Bum Ratio (dNBR) in order to develop a service based on Sentinel-2 data. The proposed approach has been tested on various study cases in Portugal for summer 2017 fires, and results show a good performance of the index and highlighted critical issues related to the Sentinel-2 data processing.

In this paper a new approach from the combination of band ratioing function and MLP Neural Networks technique is proposed to differentiate between clouds and background in Landsat ETM+ and MSG SEVIRI data. First, in order to increase the contrast of the clouds and background, a band ratioing function is applied to each sub-image. Second, the sub-images are segmented by MLP Neural Networks technique. The proposed approach was tested on 40 Landsat ETM+ sub-images of Gulf of Mexico and on 40 MSG SEVIRI sub-images over Italy. The same parameters were used in all tests. For the overall dataset, the average accuracy of 89 % was obtained for Landsat ETM+ images and the average accuracy of 85 % was obtained for MSG SEVIRI images. Our experimental results demonstrate that the proposed approach is robust and effective.

Clouds are one of the most important meteorological phenomena affecting the Earth radiation balance. The increasing development of whole-sky images enables temporal and spatial high-resolution sky observations and provides the possibility to understand and quantify cloud effects more accurately. In this letter, an attempt has been made to examine the machine learning [multilayer perceptron (MLP) neural networks and support vector machine (SVM)] capabilities for automatic cloud detection in whole-sky images. The approaches have been tested on a significant number of whole-sky images (containing a variety of cloud overages in different seasons and at different daytimes) from Vigna di Valle and Tor Vergata test sites, located near Rome. The pixel values of red, green, and blue bands of the images have been used as inputs of the mentioned models, while the outputs provided classified pixels in terms of cloud coverage or others (cloud-free pixels and sun). For the test data set, the overall accuracies of 95.07%, with a standard deviation of 3.37, and 93.66%, with a standard deviation of 4.45, have been obtained from MLP neural networks and SVM models, respectively. Although the two approaches generally generate similar accuracies, the MLP neural networks gave a better performance in some specific cases where the SVM generates poor accuracy.

Dark-spot detection is a critical step in oil-spill detection. In this paper, a novel approach for automated dark-spot detection using synthetic aperture radar imagery is presented. A new approach from the combination of Weibull multiplicative model (WMM) and pulse-coupled neural network (PCNN) techniques is proposed to differentiate between the dark spots and the background. First, the filter created based on WMM is applied to each subimage. Second, the subimage is segmented by PCNN techniques. As the last step, a very simple filtering process is used to eliminate the false targets. The proposed approach was tested on 60 Envisat and ERS2 images which contained dark spots. The same parameters were used in all tests. For the overall data set, an average accuracy of 93.66% was obtained. The average computational time for dark-spot detection with a 512 × 512 image is about 7 s using IDL software, which is the fastest one in this field at present. Our experimental results demonstrate that the proposed approach is very fast, robust, and effective. The proposed approach can be applied on any kind of synthetic aperture radar imagery.