Detecção de embarcações em imagens de sensores orbitais ópticos baseada em características contextuais e comparação de formato de contorno

Abstract

This paper presents a ship detection method based on context characteristics and contour shape comparison characteristics of targets (ships), detected on images from orbital optical sensors, by applying edge detection filters. The method uses these characteristics as descriptors and applies them to a classification process based on artificial neural network. It was evaluated by precision and recall and F1-measure. On a case study using images with 5 meters spatial resolution, the average method result was 94.9% for F1-measure, 92,19% for precision and 100% for recall, i.e., all ships of medium and large size were detected on the images. Failures on precision have occurred due to false positives caused by small islands. The method was able to detect ships in harbor and ships that are at anchor beside another one.