Deep learning is recently showing outstanding results for solving a wide variety of robotic tasks in the areas of perception, planning, localization, and control. Its excellent capabilities for learning representations from the complex data acquired in real environments make it extremely suitable for many kinds of autonomous robotic applications. In parallel, Unmanned Aerial Vehicles (UAVs) are currently being extensively applied for several types of civilian tasks in applications going from security, surveillance, and disaster rescue to parcel delivery or warehouse management. In this paper, a thorough review has been performed on recent reported uses and applications of deep learning for UAVs, including the most relevant developments as well as their performances and limitations. In addition, a detailed explanation of the main deep learning techniques is provided. We conclude with a description of the main challenges for the application of deep learning for UAV-based solutions.

Lateral flow assay tests are nowadays becoming powerful, low-cost diagnostic tools. Obtaining a result is usually subject to visual interpretation of colored areas on the test by a human operator, introducing subjectivity and the possibility of errors in the extraction of the results. While automated test readers providing a result-consistent solution are widely available, they usually lack portability. In this paper, we present a smartphone-based automated reader for drug-of-abuse lateral flow assay tests, consisting of an inexpensive light box and a smartphone device. Test images captured with the smartphone camera are processed in the device using computer vision and machine learning techniques to perform automatic extraction of the results. A deep validation of the system has been carried out showing the high accuracy of the system. The proposed approach, applicable to any line-based or color-based lateral flow test in the market, effectively reduces the manufacturing costs of the reader and makes it portable and massively available while providing accurate, reliable results.

Personal drones are becoming part of every day life. To fully integrate them into society, it is crucial to design safe and intuitive ways to interact with these aerial systems. The recent advances on User-Centered Design (UCD) applied to Natural User Interfaces (NUIs) intend to make use of human innate features, such as speech, gestures and vision to interact with technology in the way humans would with one another. In this paper, a Graphical User Interface (GUI) and several NUI methods are studied and implemented, along with computer vision techniques, in a single software framework for aerial robotics called Aerostack which allows for intuitive and natural human-quadrotor interaction in indoor GPS-denied environments. These strategies include speech, body position, hand gesture and visual marker interactions used to directly command tasks to the drone. The NUIs presented are based on devices like the Leap Motion Controller, microphones and small size monocular on-board cameras which are unnoticeable to the user. Thanks to this UCD perspective, the users can choose the most intuitive and effective type of interaction for their application. Additionally, the strategies proposed allow for multi-modal interaction between multiple users and the drone by being able to integrate several of these interfaces in one single application as is shown in various real flight experiments performed with non-expert users.

This paper presents a complete system for automatic recognition and diagnosis of electrical insulator strings, which efficiently combines different deep learning-based components to build a versatile solution to the automation problem of the power line inspection process. To this aim, the proposed system integrates one component responsible for insulator string segmentation and two components in charge of its diagnosis. The insulator string segmentation component consists of a novel Fully Convolutional Network (FCN) architecture, termed Up-Net, which enhances the capabilities of the state-of-the-art U-Net network by introducing new skip connections at certain levels of the architecture. We further propose a second variant of the Up-Net network by training it within a Generative Adversarial Network (GAN) framework. The capabilities of the proposed Up-Net variants are incremented by the application of data augmentation and transfer learning techniques, achieving accurate segmentation of the insulator string elements (i.e., discs and caps). Regarding the insulator string diagnosis, we design a Convolutional Neural Network (CNN) which takes as input the mask generated by the insulator string segmentation component and is capable of identifying the absence of a variable number of discs. The second diagnosis component consists of a novel strategy which integrates a Siamese Convolutional Neural Network (SCNN) designed for modeling the similarity between adjacent discs, allowing the detection of several types of disc defects using the same model. The proposed system has been extensively evaluated in several video sequences from real aerial inspections of high voltage insulators, showing robust insulator recognition and diagnosis capabilities.

Inspection of power line infrastructures must be periodically conducted by electric companies in order to ensure reliable electric power distribution. Research efforts are focused on automating the power line inspection process by looking for strategies that satisfy the different requirements of the inspection: simultaneously detect transmission towers, check for defects, and analyze security distances. Following this direction, this paper proposes a supervised learning approach for solving the tower detection and classification problem, where HOG (Histograms of Oriented Gradients) features are used to train two MLP (multi-layer perceptron) neural networks. The first classifier is used for background-foreground segmentation, and the second multi-class MLP is used for classifying within 4 different types of electric towers. A thorough evaluation of the tower detection and classification approach has been carried out on image data from real inspections tasks with different types of towers and backgrounds. In the different evaluations, highly encouraging results were obtained. This shows that a learning-based approach is a promising technique for power line inspection.