This paper proposes a teleophthalmology support system in which we use algorithms of object detection and semantic segmentation, such as faster region-based CNN (FR-CNN) and SegNet, based on several CNN architectures such as: Vgg16, MobileNet, AlexNet, etc. These are used to segment and analyze the principal anatomical elements, such as optic disc (OD), region of interest (ROI) composed by the macular region, real retinal region, and vessels. Unlike the conventional retinal image quality assessment system, the proposed system provides some possible reasons about the low-quality image to support the operator of an ophthalmoscope and patient to acquire and transmit a better-quality image to central eye hospital for its diagnosis. The proposed system consists of four steps: OD detection, OD quality analysis, obstruction detection of the region of interest (ROI), and vessel segmentation. For the OD detection, artefacts and vessel segmentation, the FR-CNN and SegNet are used, while for the OD quality analysis, we use transfer learning. The proposed system provides accuracies of 0.93 for the OD detection, 0.86 for OD image quality, 1.0 for artefact detection, and 0.98 for vessel segmentation. As the global performance metric, the kappa-based agreement score between ophthalmologist and the proposed system is calculated, which is higher than the score between ophthalmologist and general practitioner.

No abstract

ResumenEn este artículo se presenta una comparación entre diferentes algoritmos de descripción de texturas de bajo nivel acoplados con una máquina de soporte vectorial versus el algoritmo de aprendizaje profundo, en la tarea de reconocimiento y clasificación de imágenes aéreas. Para dicha tarea, una base de datos de 1,200 imágenes es utilizada para realizar los entrenamientos supervisados. El objetivo consiste en clasificar las imágenes en seis categorías comúnmente encontradas en zonas urbanas, de tal manera que pueda ser utilizado en cualquier parte del mundo. Los resultados arrojan que con 150 muestras de cada clase, el algoritmo de aprendizaje profundo es capaz de clasificar imágenes de avenidas, edificios, industrias, zonas naturales, zonas residenciales y cuerpos de agua, con un 87% de exactitud. Los resultados experimentales presentados muestran que las imágenes etiquetadas como edificios e industrias son las más complejas de discernir entre ellas, tanto para descriptores de bajo nivel como para las técnicas de aprendizaje profundo. Palabras clave: aprendizaje profundo; máquina de soporte vectorial; imágenes aéreas; descriptores de texturas; base de datos Classification of Urban Aerial Images: A Comparison between Low-Semantic Descriptors and Deep Learning AbstractThis paper presents a comparison between different low-semantic descriptive algorithms coupled with a support vector machine and the deep learning algorithm, for the task of recognition and classification of aerial images. For this task, a database composed of 1200 images is used to fulfill the supervised trainings. The objective consists on classifying images in six categories that are commonly found on urban areas, in order to be used in any part of the world. The results show that with 150 samples of each class, the deep learning algorithm is capable of classifying images of avenues, buildings, industries, natural areas, residential areas and water bodies with an 87% of accuracy. Experimental results also prove that the labeled images as industry and buildings are the most complex ones to distinguish among these two classes, both for low-level descriptors and deep learning techniques.