Object Classification in Semi Structured Enviroment Using Forward-Looking Sonar

Abstract

The submarine exploration using robots has been increasing in recent years. The automation of tasks such as monitoring, inspection, and underwater maintenance requires the understanding of the robot’s environment. The object recognition in the scene is becoming a critical issue for these systems. On this work, an underwater object classification pipeline applied in acoustic images acquired by Forward-Looking Sonar (FLS) are studied. The object segmentation combines thresholding, connected pixels searching and peak of intensity analyzing techniques. The object descriptor extract intensity and geometric features of the detected objects. A comparison between the Support Vector Machine, K-Nearest Neighbors, and Random Trees classifiers are presented. An open-source tool was developed to annotate and classify the objects and evaluate their classification performance. The proposed method efficiently segments and classifies the structures in the scene using a real dataset acquired by an underwater vehicle in a harbor area. Experimental results demonstrate the robustness and accuracy of the method described in this paper.

Object Recognition in Forward Looking Sonar Images using Transfer Learning

“…In total the dataset contains 531 labeled segments. The same dataset was also used by [8], to which we later compare our results in section V.…”

“…The segmented objects has then been described by Gaussians over their locations in the image. These steps were performed by the open source acoustic imaging tool [8]. A full FLS scan with the Gaussians marked in red is seen in Fig.…”

“…The dataset is unbalanced in the sense that it contains a different number of samples for the different classes. The size of the dataset is a little bit smaller but of comparable size to the unbalanced dataset in [8], which allows us to compare our results. The evaluation metrics used are precision, recall, and the F1-score (the harmonic average between precision and recall).…”

“…Our method can be compared to the classification results of the original work [8]. The hand-crafted features are computed from segments described by Gaussian probabilistic functions.…”

“…We compare the new classifier with a classifier designed by using hand-crafted features [8]. We show that the DL approach of transfer learning can be a good alternative to using hand-crafted features.…”