One important issue emerging strongly in agriculture is related with the automatization of tasks, where the optical sensors play an important role. They provide images that must be conveniently processed. The most relevant image processing procedures require the identification of green plants, in our experiments they come from barley and corn crops including weeds, so that some types of action can be carried out, including site-specific treatments with chemical products or mechanical manipulations. Also the identification of textures belonging to the soil could be useful to know some variables, such as humidity, smoothness or any others. Finally, from the point of view of the autonomous robot navigation, where the robot is equipped with the imaging system, some times it is convenient to know not only the soil information and the plants growing in the soil but also additional information supplied by global references based on specific areas. This implies that the images to be processed contain textures of three main types to be identified: green plants, soil and sky if any. This paper proposes a new automatic approach for segmenting these main textures and also to refine the identification of sub-textures inside the main ones. Concerning the green identification, we propose a new approach that exploits the performance of existing strategies by combining them. The combination takes into account the relevance of the information provided by each strategy based on the intensity variability. This makes an important contribution. The combination of thresholding approaches, for segmenting the soil and the sky, makes the second contribution; finally the adjusting of the supervised fuzzy clustering approach for identifying sub-textures automatically, makes the third finding. The performance of the method allows to verify its viability for automatic tasks in agriculture based on image processing.

Accurate and real-time forecasting of the price of oil plays an important role in the world economy. Research interest in forecasting this type of time series has increased considerably in recent decades, since, due to the characteristics of the time series, it was a complicated task with inaccurate results. Concretely, deep learning models such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) have appeared in this field with promising results compared to traditional approaches. To improve the performance of existing networks in time series forecasting, in this work two types of neural networks are brought together, combining the characteristics of a Graph Convolutional Network (GCN) and a Bidirectional Long Short-Term Memory (BiLSTM) network. This is a novel evolution that improves existing results in the literature and provides new possibilities in the analysis of time series. The results confirm a better performance of the combined BiLSTM-GCN approach compared to the BiLSTM and GCN models separately, as well as to the traditional models, with a lower error in all the error metrics used: the Root Mean Squared Error (RMSE), the Mean Squared Error (MSE), the Mean Absolute Percentage Error (MAPE) and the R-squared (R²). These results represent a smaller difference between the result returned by the model and the real value and, therefore, a greater precision in the predictions of this model.

An important objective in weed management is the discrimination between grasses (monocots) and broad-leaved weeds (dicots), because these two weed groups can be appropriately controlled by specific herbicides. In fact, efficiency is higher if selective treatment is performed for each type of infestation instead of using a broadcast herbicide on the whole surface. This work proposes a strategy where weeds are characterised by a set of shape descriptors (the seven Hu moments and six geometric shape descriptors). Weeds appear in outdoor field images which display real situations obtained from a RGB camera. Thus, images present a mixture of both weed species under varying conditions of lighting. In the presented approach, four decision-making methods were adapted to use the best shape descriptors as attributes and a choice was taken. This proposal establishes a novel methodology with a high success rate in weed species discrimination.

The dynamics of a skid-steering robot present intrinsic non-linearities that make the design and implementation of a controller a very complex task, time-consuming, and difficult to implement into an embedded system with limited resources. This paper presents a simplified first order digital model approximation and an optimal observer-based control approach for the tracking of the lateral position of such robots. In order to verify the validity of this proposal, 3D real-time interactive simulations and real validations with an agricultural skid-steering robot were performed with satisfactory results.

In recent years, proximal sensing data has increasingly been used to optimize forest inventories. In this paper we present a forest inventory methodology based on stereoscopic hemispherical images. An automated pixel-based approach and a user-guided “region growing” approach have been developed for image matching. To estimate the basal area, number of trees and mean diameter, the sampling probability is determined for each tree. The accuracy and precision of the estimates derived from stereoscopic hemispherical images was analyzed for a set of National Forest Inventory plots. The results revealed that tree matching depends on the species, the distance to the target tree and the diameter. The Pearson correlation coefficient was 0.86 for the mean diameter and 0.89 for the basal area, whereas for the number of trees per hectare it was 0.59. The proposed methods may be used in large scale forest inventories as a cost-efficient way of obtaining data on diameter distribution and basal area from field surveys following a two-stage scheme combined with remote sensing techniques.