Cyanobacterial blooms emerge unpredictably on the surface of lentic water bodies, posing both ecological threats and public health risks. To effectively monitor these events, this study introduces the use of Machine Vision Systems (MVS) integrated into Autonomous Surface Vehicles (ASVs). These ASVs are capable of autonomous and safe navigation, enabling them to detect cyanobacterial blooms while avoiding obstacles. Convolutional Neural Networks (CNNs) are employed for early detection and continuous monitoring, but their effectiveness hinges on access to large, high-quality training datasets. Due to the sporadic and uncontrollable nature of bloom occurrences, acquiring sufficient real-world images for training and validating CNN models is a significant challenge. To overcome this, the Stable Diffusion XL (SDXL) text-to-image generative model is utilized to produce realistic synthetic images, ensuring a sufficient dataset for training. However, SDXL alone struggles to accurately depict cyanobacterial blooms. To address this limitation, DreamBooth is used to fine-tune SDXL with a small set of real bloom-specific image patches. To ensure the diversity of the synthetic dataset, detailed prompts for SDXL are generated using a Large Language Model (LLM). The combination of SDXL fine-tuning with LLM-driven prompts design applied to environmental monitoring and autonomous navigation in lentic environments represents the core innovation of this work. A dual-task CNN model is then trained on the synthetic dataset to simultaneously detect blooms and obstacles. Experimental results demonstrate the effectiveness and novelty of the proposed approach, showing improvements of up to 15.74% in object detection and 6.48% in semantic segmentation compared to the baseline dataset

The development of explainability methods for Convolutional Neural Networks (CNNs), under the growing framework of explainable Artificial Intelligence (xAI) for image understanding, is crucial due to neural networks success in contrast with their black box nature. However, usual methods focus on image visualizations and are inadequate to analyze the encoded contextual information (that captures the spatial dependencies of pixels considering their neighbors), as well as to explain the evolution of learning across layers without degrading the information. To address the latter, this paper presents a novel explanatory method based on the study of the latent representations of CNNs through their topology, and supported by Topological Data Analysis (TDA). For each activation layer after a convolution, the pixel values of the activation maps along the channels are considered latent space points. The persistent homology of this data is summarized via persistence landscapes, called Latent Landscapes. This provides a global view of how contextual information is being encoded, its variety and evolution, and allows for statistical analysis. The applicability and effectiveness of our approach is demonstrated by experiments conducted with CNNs trained on distinct datasets: (1) two U-Net segmentation models on RGB and pseudo-multiband images (generated by considering vegetation indices) from the agricultural benchmark CRBD were evaluated, in order to explain the difference in performance; and (2) a VGG-16 classification network on CIFAR-10 (RGB) was analyzed, showing how the information evolves within the network. Moreover, comparisons with state-of-the-art methods (Grad-CAM and occlusion) prove the consistency and validity of our proposal. It offers novel insights into the decision making process and helps to compare how models learn.

The remarkable performance of Convolutional Neural Networks (CNNs) has increased their use in real-time systems and devices with limited resources. Hence, compacting these networks while preserving accuracy has become necessary, leading to multiple compression methods. However, the majority require intensive iterative procedures and do not delve into the influence of the used data. To overcome these issues, this paper presents several contributions, framed in the context of explainable Artificial Intelligence (xAI): (a) two filter pruning methods for CNNs, which remove the less significant convolutional kernels; (b) a fine-tuning strategy to recover generalization; (c) a layer pruning approach for U-Net; and (d) an explanation of the relationship between performance and the used data. Filter and feature maps information are used in the pruning process: Principal Component Analysis (PCA) is combined with a next-convolution influence-metric, while the latter and the mean standard deviation are used in an importance score distribution-based method. The developed strategies are generic, and therefore applicable to different models. Experiments demonstrating their effectiveness are conducted over distinct CNNs and datasets, focusing mainly on semantic segmentation (using U-Net, DeepLabv3+, SegNet, and VGG-16 as highly representative models). Pruned U-Net on agricultural benchmarks achieves 98.7% parameters and 97.5% FLOPs drop, with a 0.35% gain in accuracy. DeepLabv3+ and SegNet on CamVid reach 46.5% and 72.4% parameters reduction and a 51.9% and 83.6% FLOPs drop respectively, with almost no decrease in accuracy. VGG-16 on CIFAR-10 obtains up to 86.5% parameter and 82.2% FLOPs decrease with a 0.78% accuracy gain.

Molecular networks provide a powerful tool for the study of biomedical systems, in particular several studies have detected alterations of the network structure associated to disease states. Here we propose that diseases cannot only alter the structure of the network but also its stability. To evaluate network stability we have developed a new methodological framework. Our approach is an adaptation of the classical Deterministic Simulated Annealing algorithm to work with discrete states. Adjusted energy values are used to compare the network stability in disease and control states. The results show that cancer networks are less stable than the Alzheimer’s disease (AD) ones. These results can be interpreted in terms of our previous observations on cancer and AD inverse comorbidity, i.e. AD patients have lower than expected risk to suffer cancer.

This study presents a new unsupervised hybrid classifier for natural texture identification in aerial images. The proposed strategy combines through the fuzzy integral (FI) six well‐tested base supervised classifiers. This automation is based on the generation of a general rule inferred through decision tree learning, ID3 strategy from the training data. This rule allows generation of a partition of the set of images that the base classifiers use to estimate automatically their parameters. These parameters are the inputs to calculate the relative importance of each classifier in their combination by the FI. The resulting classifier has been compared with related techniques getting an improvement of 8.04% average. The study includes discussion on this comparison.

This paper formulates and proposes a discrete solution for the problem of finding a lost target under uncertainty in minimum time (Minimum Time Search). Given a searching region where some information about the target is known but uncertain (i.e. location and dynamics), and a searching agent with constrained dynamics, we provide two decision making algorithms that optimizes the agent actions to find the target in a minimum time. The problem is faced as a discrete optimization: the actions and the sensor are discrete, and the target probabilistic model is described over a graph, where each vertex contains the target's location probability information and each edge defines the agent possible actions. We revisit the mathematical model of the optimal search problem and we propose a novel approach to include the time into the decision making by reinterpreting it as a maximum discounted time reward problem. The optimal decision plan for the agent is obtained by solving this non convex discrete problem using the cross entropy method. By performing statistical simulations we show how the target is found in minimum time.

This work presents a strategy to track and describe the boundary of an environmental surface using a sequence of incomplete boundary images coming from an UAV with a camera onboard. The algorithm uses all captured images to feed the trajectory, which drives the UAV towards the next point to take the next shot, until the identification is completed. The multiple patterns of the input image make the problem difficult. Our approach orders the points of the image to identify the existent edges, and computes the trajectory as a minimization problem using a multi-objective function that gathers information about the boundary and the polynomial that fit in these edges.

The development of image processing methods to discriminate between weed, crop and soil is an important step for precision agriculture, the main goal of which is the site-specific management of crops. The main challenge in terms of image analysis is to achieve an appropriate discrimination in outdoor field images under varying conditions of lighting, soil background texture and crop damage. This work presents several developed computer-vision-based methods for the estimation of percentages of weed, crop and soil in digital images of a crop field. These methods are interchangeable among them, having each one of them a set of input parameters that need to be adjusted differently for each image. Two different evolutionary methods (standard genetic algorithm and NSGA-II) have been used to adjust these parameters and find the best method combinations. The proposed approach can reach a correlation with real data of up to 97% for a set of images acquired from different fields and under different conditions.

Super-resolution (SR) techniques produce a high-resolution image from a set of low-resolution undersampled images. In this paper, we propose a new method for super-resolution that uses sampling theory concepts to derive a noniterative SR algorithm. We first raise the issue of the validity of the data model usually assumed in SR, pointing out that it imposes a band-limited reconstructed image plus a certain type of noise. We propose a sampling theory framework with a prefiltering step that allows us to work with more general data models and also a specific new method for SR that uses Delaunay triangulation and B-splines to build the super-resolved image. The proposed method is noniterative and well posed. We prove its effectiveness against traditional iterative and noniterative SR methods on synthetic and real data. Additionally, we also prove that we can first solve the interpolation problem and then make the deblurring not only when the motion is translational but also when there are rotations and shifts and the imaging system Point Spread Function (PSF) is rotationally symmetric.

This paper presents an approach to trajectory generation for unmanned aerial vehicles (UAV) by using mixed integer linear programming (MILP) and a modification of the A* algorithm to optimize paths in dynamic environments, particularly having pop-ups with a known future probability of appearance. Each pop-up leads to one or several possible evasion maneuvers, characterized with a set of values used as decision making parameters in an integer linear programming (ILP) model that optimizes the final route by choosing the most suitable alternative trajectories, according to the imposed constrains such as maximum fuel consumption and spent time. The model of the system in MILP and A* algorithms is presented, as well as the ILP formulation for decision making. Results and discussions are given to promote future real time implementations.

This paper presents an approach to trajectories optimization for unmanned aerial vehicle (UAV) in presence of obstacles, waypoints, and threat zones such as radar detection regions, using mixed integer linear programming (MILP). The main result is the linear approximation of a nonlinear radar detection risk function with integer constraints and indicator 0-1 variables. Several results are presented to show that the approach can yields trajectories depending on the acceptable risk of detection.

This paper outlines an optimization relaxation approach based on the analog Hopfield neural network (HNN) for solving the image change detection problem between two images. A difference image is obtained by subtracting pixel by pixel both images. The network topology is built so that each pixel in the difference image is a node in the network. Each node is characterized by its state, which determines if a pixel has changed. An energy function is derived, so that the network converges to stable states. The analog Hopfield's model allows each node to take on analog state values. Unlike most widely used approaches, where binary labels (changed/unchanged) are assigned to each pixel, the analog property provides the strength of the change. The main contribution of this paper is reflected in the customization of the analog Hopfield neural network to derive an automatic image change detection approach. When a pixel is being processed, some existing image change detection procedures consider only interpixel relations on its neighborhood. The main drawback of such approaches is the labeling of this pixel as changed or unchanged according to the information supplied by its neighbors, where its own information is ignored. The Hopfield model overcomes this drawback and for each pixel allows a tradeoff between the influence of its neighborhood and its own criterion. This is mapped under the energy function to be minimized. The performance of the proposed method is illustrated by comparative analysis against some existing image change detection methods.

This paper outlines a method for solving the stereovision matching problem using edge segments as the primitives. In stereovision matching, the following constraints are commonly used: epipolar, similarity, smoothness, ordering, and uniqueness. We propose a new strategy in which such constraints are sequentially combined. The goal is to achieve high performance in terms of correct matches by combining several strategies. The contributions of this paper are reflected in the development of a similarity measure through a support vector machines classification approach; the transformation of the smoothness, ordering and epipolar constraints into the form of an energy function, through an optimization simulated annealing approach, whose minimum value corresponds to a good matching solution and by introducing specific conditions to overcome the violation of the smoothness and ordering constraints. The performance of the proposed method is illustrated by comparative analysis against some recent global matching methods.

This paper presents an approach to the local stereovision matching problem using edge segments as features with four attributes. From these attributes we compute a matching probability between pairs of features of the stereo images. A correspondence is said true when such a probability is maximum. We introduce a nonparametric strategy based on Parzen's window (1962) to estimate a probability density function (PDF) which is used to obtain the matching probability. This is the main finding of the paper. A comparative analysis of other recent matching methods is included to show that this finding can be justified theoretically. A generalization of the proposed method is made in order to give guidelines about its use with the similarity constraint and also in different environments where other features and attributes are more suitable.

This paper presents an approach to the local stereo matching problem using edge segments as features with several attributes. We have verified that the differences in attributes for the true matches cluster in a cloud around a center. The correspondence is established on the basis of the minimum distance criterion, computing the Mahalanobis distance between the difference of the attributes for a current pair of features and the cluster center (similarity constraint). We introduce a learning strategy based on the Hebbian Learning to get the best cluster center. A comparative analysis among methods without learning and with other learning strategies is illustrated.