Almond trees in Portugal are susceptible to aphid infestation, which can result in reduced fruit production. To effectively tackle this issue, the combination of remote sensing (RS) data and machine learning (ML) classifiers can be used to accurately detect the presence of aphids. This study focuses in the implementation of ML classifiers and RS data analysis to identify aphids on almond trees, using high-resolution multispectral data collected through an unmanned aerial vehicle (UAV) in a Portuguese almond orchard. Four ML classifiers, kNN, SVM, RF and XGBoost, were employed and fine-tuned using vegetation indices derived from spectral data. The results revealed that the SVM classifier achieved an overall accuracy (OA) of 77%, followed by kNN with an OA of 74%, while XGBoost and RF achieved OAs of 71% and 69%, respectively. Consequently, this study demonstrates the viability of employing RS data and ML classifiers for aphid identification in almond orchards.

The estimation of dendrometric parameters of tree crops is crucial to decision making support for ecological and economic reasons. However, traditional methods for its measurement are time-consuming and laborious. Remote sensing data acquired from unmanned aerial vehicles (UAVs) combined with computer vision and Structure from Motion (SfM) algorithms can provide an easier and reliable solution to estimate those parameters. Nevertheless, various UAV flight settings can influence the quality of parameters derived from these data (e.g., flight height, imagery overlap). Thus, the main goal of this study is to assess the impact of different flight configurations on the detection of olive trees and on height and crown diameter estimation. The results showed that not only the configuration of the flight affects the dendrometric results, but also the topography of the terrain. Automatic tree detection revealed to be insensitive to the different flight configurations, whereas the tree height estimation was strongly affected. Among the analysed flights, the plan in double grid at 60 m of flight altitude and 90% of frontal overlap showed the best performance.

The grape variety plays an important role in the wine production chain, thus identifying it is crucial for production control. Ampelographers, professionals who identify grape varieties through plant visual analysis, are scarce, and molecular markers are expansive to identify grape varieties on a large scale. In this context, Deep Learning models become an effective way to handle ampelographers scarcity. In this work, we explore the benefit of using deep learning vision transformers architecture relative to conventional CNN to identify 12 grapevine varieties using leaf-centred RGB images acquired in the field. We train an Xception model as a baseline and four different configurations of the ViT_B model. The best model achieved 0.96 of Fl-score, outperforming the state-of-the-art convolutional-based model in the used dataset.

The grapevine variety plays an important role in wine chain production, thus identifying it is crucial for control activities. However, the specialists responsible for identifying the different varieties, mainly through visual analysis, are disappearing. In this scenario, Deep Learning (DL) classification techniques become a possible solution to handle professionals' scarcity. Nevertheless, previous experiments show that trained classification models use the background information to make decisions, which should be avoided. In this paper, we present a study allowing the assessment of removing background regions from the grapevine images in the improvement classification using DL models. The Xception model is trained with a normal dataset and its segmented version. The Local Interpretable Model-Agnostic Explanations (LIME), Grad-CAM, and Grad-CAM++ approaches are used to visualize the segmentation impact in classification decisions. F1-score of 0.92 and 0.94 were achieved, respectively, for segmented-dataset and normal-dataset trained models. Despite the model trained with the segmented-dataset to achieve a worse performance, the Explainable Artificial Intelligence (XAI) approaches showed that it looks into more reliable regions when making decisions.

The detection of invasive plant species in aquatic ecosystems is important to help in the control or to mitigate its spread and impacts. Remote sensing (RS) can be explored in this context, helping to monitor this type of plants. This study intends to present a free to use and open-source software application that, through a graphical user interface, can process remote sensed data to monitor the spread of invasive plant species in aquatic environments, enabling a multi-temporal monitoring. Both unmanned aerial vehicle and satellite-based data were used to validate the potential of the proposed application. A site containing water hyacinth (Eichhornia crassipes) was selected as case study. Both RS platforms provided effective data to detect the areas containing water hyacinth. Thus, this tool provides an alternative and user-friendly way to include RS-based data in ecological studies allowing the detection of invasive plants in water channels.

In the last decade Unmanned Aerial Systems (UAS) have become a reference tool for agriculture applications. The integration of multispectral sensors that can capture near infrared (NIR) and red edge spectral reflectance allows the creation of vegetation indices, which are fundamental for crop monitoring process. In this study, we propose a methodology to analyze the vegetative state of almond crops using multi-temporal data acquired by a multispectral sensor accoupled to an Unmanned Aerial Vehicle (UAV). The methodology implemented allowed individual tree parameters extraction, such as number of trees, tree height, and tree crown area. This also allowed the acquisition of Normalized Difference Vegetation Index (NDVI) information for each tree. The multi-temporal data showed significant variations in the vegetative state of almond crops.

Modelling of material appearance from reflectance measurements has become increasingly prevalent due to the development of novel methodologies in Computer Graphics. In the last few years, some advances have been made in measuring the light-material interactions, by employing goniometers/reflectometers under specific laboratory’s constraints. A wide range of applications benefit from data-driven appearance modelling techniques and material databases to create photorealistic scenarios and physically based simulations. However, important limitations arise from the current material scanning process, mostly related to the high diversity of existing materials in the real-world, the tedious process for material scanning and the spectral characterisation behaviour. Consequently, new approaches are required both for the automatic material acquisition process and for the generation of measured material databases. In this study, a novel approach for material appearance acquisition using hyperspectral data is proposed. A dense 3D point cloud filled with spectral data was generated from the images obtained by an unmanned aerial vehicle (UAV) equipped with an RGB camera and a hyperspectral sensor. The observed hyperspectral signatures were used to recognise natural and artificial materials in the 3D point cloud according to spectral similarity. Then, a parametrisation of Bidirectional Reflectance Distribution Function (BRDF) was carried out by sampling the BRDF space for each material. Consequently, each material is characterised by multiple samples with different incoming and outgoing angles. Finally, an analysis of BRDF sample completeness is performed considering four sunlight positions and 16x16 resolution for each material. The results demonstrated the capability of the used technology and the effectiveness of our method to be used in applications such as spectral rendering and real-word material acquisition and classification.

This study assesses machine learning methods for the classification of an intertidal reef using RGB and multispectral imagery acquired by an unmanned aerial vehicle (UAV). After the photogrammetric processing of the acquired data an orthophoto mosaic was generated, from the RGB imagery, and the reflectance of four bands (green, red, red edge and near infrared) from the multispectral data. Four machine learning classifiers were evaluated: support vector machines (SVM), artificial neural networks (ANN) naive Bayes (NB) and random forests (RF). The data was classified into four classes: sand; rock, barnacles, limpets; mussels, rock; and algae mixed. The classifiers were trained with RGB and with multispectral data. The pixel-based classification results demonstrated that when using multispectral data all classifiers overcame the performance achieved when using RGB data. NB classifier performed better in discriminating all classes and detecting submerged seaweeds. Such techniques present a valuable tool for accurately map the coastal zone.

In this paper we present a Deep Learning-based methodology to automatically classify 12 of the most representative grape-varieties existing in the Douro Demarked region, Portugal. The dataset used consisted of images of leaves at different stages of development, collected on their natural environment. The development of such methodologies becomes particularly important, in a scenario in which ampeleographers are disappearing, creating a gap in the task of inspection of grape varieties. Our approach was based on the transfer learning of the Xcepetion model, using Focal Loss, adaptive learning rate decay and SGD. The model obtained a F1 score of 0.93. To clearly understand the predictions of the model, and realize which regions of the image contributed the most to the classification, the LIME library was used. This way it was possible to identify the parts of the images that were considered for and against each prediction.

Materials represented by measured BRDF (Bidirectional Reflectance Distribution function) with reflectance data captured from real-world materials have become increasingly prevalent due to the development of novel measurement approaches. Nowadays, important limitations can be highlighted in the current material scanning process, mostly related to the high diversity of existing materials in the real-world and the tedious process for material scanning. Consequently, new approaches are required both for the automatic material acquisition process and for the generation of measured material databases. In this study, a novel approach is proposed for modelling the material appearance by sampling hyperspectral measurements on the BRDF domain. An unmanned aerial vehicle (UAV)-based hyperspectral sensor was used to capture high spatial and spectral resolution data. The generated hyperspectral data cubes were used to identify materials with a similar spectral behaviour. Then, a sparse mapping of collected samples is developed to study the appearance of natural and artificial materials in an urban scenario.

The development and implementation of a virtual environment that aims to support farmers in managing their land and crops in a more sustainable way is presented in this paper. It allows both textual and 3D visualization of crop-related biophysical parameters, such as height, volume and length. Moreover, the latter can be dynamically altered according to various criteria. A case study was conducted in a Portuguese vineyard. The application was developed using the Unity software, while a real agricultural data feed was provided by mySense interface. The virtual environment can be seen as a valuable decision support system to assist farmers.

With the continuous escalation of global warming and consequent water scarcity, techniques to optimize water use of irrigation in agriculture are needed. Thus, deficit irrigation strategies (DI) can be used for a sustainable water usage. However, it is necessary to recursively monitor plant response under DI to ensure their productivity and prevent from severe water stress. The goal of this study is to assesscanopy and soil surface temperatures of olive trees under different irrigation strategies, through thermal infrared images obtained by Unmanned Aerial Vehicle (UAV). The temperatures from the different irrigation strategies were analysed with three approaches using the difference between canopy and air temperatures (Tc-Ta). The use of UAV-based thermal infrared imagery has proven to be extremely useful to the estimation of olive canopy and soil surface temperatures, which allow to discriminate different irrigation treatments.

Developed focusing agriculture sustainability, mySense is a comprehensive close-range sensor-based data management environment to improve precision farming practices. It integrates discussion platforms for quick problem solving through experts support and a computational intelligence layer for multipurpose application (e.g. vine variety discrimination, plant disease detection and identification). Attending the need for keeping track of agricultural crops not only based on close-range sensing but also at a macro perspective, mySense was complemented with proper functionalities to unlock macro-monitoring features, through the implementation of a Web-based Geographical Information System (WebGIS) planned as a sidekick application that provides agriculture professionals with visual decision support tools over remote sensed data. This paper presents and discusses its specification and implementation.

Nowadays, many drone models are available, designed for the most diverse applications. However, the various models fall into one of two types of drone: multi-copter or fixed-wing. The first type of drone consumes a lot of energy, since motors have to turn during all the flight. The former type of drone, in general needs a runway to take-off and landing. Moreover, they fly fast and cannot be motionless, which is unsuitable for many applications. In this paper we present a new paramotor drone, conceived and designed to overcome the highlighted limitations and to be a low-cost solution adapted for most applications. The selection of the various components of the presented prototype was based on a very thorough study, considering aerodynamic and efficiency criteria.

Unmanned aerial vehicles (UAVs) are used nowadays as a standard tool to derive very high-resolution geospatial data. However, UAV payload limitation imposes the use of not such reliable hardware affecting the georeferencing precision. In the literature it is possible to find numerous studies investigating the parameters influencing UAV-based products quality. Even if new photogrammetry methods could, in theory, avoid the use of ground control points (GCPs), they still playa key role to assure quality products. Nevertheless, usually only the number and distribution of GCPs are taking into account, since both change the geometric accuracy of the final products. In order to improve the understanding of the actual influence of GCPs, in this study we evaluate how can different physical characteristics affect GCPs identification in aerial images. The results demonstrate that GCPs' color, material, size and shape, among others, may influence a precise identification in aerial imagery.

Individual tree segmentation is a challenging task due to the labour-intensive and time-consuming work required. Remote sensing data acquired from sensors coupled in unmanned aerial vehicles (UAV) constitutes a viable alternative to provide a quicker data acquisition, covering broader areas in a shorter period of time. This study aims to use UAV-based multispectral imagery to automatically identify individual trees in a chestnut stand. Tree parameters were estimated allowing its characterization. The leaf area index (LAI) was measured and was correlated with the estimated parameters. A good correlation was found for NDVI(R2= 0.76), while this relationship was less evident in the tree crown area and tree height. This way, our results indicate that the use of UAV-based multispectral imagery is a quick and reliable way to determine canopy structural parameters and LAI of chestnut trees.

In this study machine learning methods were applied to RGB data obtained by an unmanned aerial vehicle (UAV) to assess this effectiveness in vineyard classification. The very high-resolution UAV-based imagery was subjected to a photogrammetric processing allowing the generation of different outcomes: orthophoto mosaic, crop surface model and five vegetation indices. The orthophoto mosaic was used in an object-based image analysis approach to group pixels with similar values into objects. Three machine learning techniques-support vector machine (SVM), random forest (RF) and artificial neural network (ANN)-were applied to classify the data into four classes: grapevine, shadow, soil and other vegetation. The data were divided with 22% (n=240, 60 per class) for training purposes and 78% (n = 850) for testing purposes. The mean value of the objects from each feature were used to create a dataset for prediction. The results demonstrated that both RF and ANN models showed a good performance, yet the RF classifier achieved better results.

Machine Learning (ML) progressed significantly in the last decade, evolving the computer-based learning/prediction paradigm to a much more effective class of models known as Deep learning (DL). Since then, hyperspectral data processing relying on DL approaches is getting more popular, competing with the traditional classification techniques. In this paper, a valid ML/DL-based works applied to hyperspectral data processing is reviewed in order to get an insight regarding the approaches available for the effective meaning extraction from this type of data. Next, a general DL-based methodology focusing on hyperspectral data processing to provide farmers and winemakers effective tools for earlier threat detection is proposed.