Recent studies have shown evidence of a significant decline of the Posidonia oceanica (P.O.) meadows on a global scale. The monitoring and mapping of these meadows are fundamental tools for measuring their status. We present an approach based on a deep neural network to automatically perform a highprecision semantic segmentation of the P.O. meadows in sea-floor images, offering several improvements over the state-of-the-art techniques. Our network demonstrates outstanding performance over diverse test sets, reaching a precision of 96.57% and an accuracy of 96.81%, surpassing the reliability of labeling the images manually. Moreover, the network is implemented in an autonomous underwater vehicle, performing an online P.O. segmentation, which will be used to generate real-time semantic coverage maps.

Controlling and quantifying the presence of Posidonia Oceanica (P.O.) in the Mediterranean sea is crucial for the conservation of these endemic ecosystems and to underscore the negative impact of many anthropogenic activities. These activities, which include uncontrolled leisure anchoring or illegal drag fishing, directly affect the tourism and fishing industries. Nowadays, the control and quantification of P.O. is done by divers, in a slow and imprecise process achieved in successive missions of a duration limited by the capacity of the oxygen scuba tanks. This paper proposes the application of robotic and computer vision technologies to upgrade the current P.O. control methods, building large scale coverage maps using the imagery provided by an autonomous underwater vehicle endowed with a bottom-looking camera. The process includes four main steps: 1) training a classifier based on two different Gabor filter image patch descriptors and a support vector machine; 2) detecting P.O. autonomously, both on-line and off-line, in each individual image; 3) color photo-mosaicking the area explored by the vehicle to obtain a global view of the meadow structure; these mosaics are extremely useful to analyze the structure and extension of the meadow and to calculate some of the biological descriptors needed to diagnose its state; and 4) building a binary coverage map in which the classification results of areas with image overlap are fused according to four different strategies. The experiments, performed in coastal areas of Mallorca and Girona, evaluate and compare the proposed descriptors and fusion techniques, showing, in some cases, accuracies and precisions above 90% in the detection of different patterns of P.O., from video sequences at different locations, in different seasons and with different environmental conditions.

This paper proposes a method to perform on-line multi-class segmentation of Side-Scan Sonar acoustic images, thus being able to build a semantic map of the sea bottom usable to search loop candidates in a SLAM context. The proposal follows three main steps. First, the sonar data is pre-processed by means of acoustics based models. Second, the data is segmented thanks to a lightweight Convolutional Neural Network which is fed with acoustic swaths gathered within a temporal window. Third, the segmented swaths are fused into a consistent segmented image. The experiments, performed with real data gathered in coastal areas of Mallorca (Spain), explore all the possible configurations and show the validity of our proposal both in terms of segmentation quality, with per-class precisions and recalls surpassing the 90%, and in terms of computational speed, requiring less than a 7% of CPU time on a standard laptop computer. The fully documented source code, and some trained models and datasets are provided as part of this study.

We present a new vision-based localization system applied to an autonomous underwater vehicle (AUV) with limited sensing and computation capabilities. The traditional EKF-SLAM approaches are usually expensive in terms of execution time; the approach presented in this paper strengthens this method by adopting a trajectory-based schema that reduces the computational requirements. The pose of the vehicle is estimated using an extended Kalman filter (EKF), which predicts the vehicle motion by means of a visual odometer and corrects these predictions using the data associations (loop closures) between the current frame and the previous ones. One of the most important steps in this procedure is the image registration method, as it reinforces the data association and, thus, makes it possible to close loops reliably. Since the use of standard EKFs entail linearization errors that can distort the vehicle pose estimations, the approach has also been tested using an iterated Kalman filter (IEKF). Experiments have been conducted using a real underwater vehicle in controlled scenarios and in shallow sea waters, showing an excellent performance with very small errors, both in the vehicle pose and in the overall trajectory estimates.