Identification of underwater mines from electro-optical imagery using an operated-assisted reinforcement on-line learning

This paper presents a new approach for using an operated-assisted reinforcement on-line learning for mine identification from electro-optical images. The images acquired from Streak Tube Imaging Lidar (STIL) that constitute contrast and range maps are used. A reduced set of features using the Zernike moments is extracted from each preprocessed and detected/segmented object. This set is fed to a flexible network which uses a new on-line reinforcement learning based on expert operator´s votes. An important feature of this system is that it allows for the incorporation of new objects learning without deleting or modifying the previously learnt cases. The performance of this preliminary in-situ learning system will be demonstrated in this paper on several STIL images and the confusion matrix of the overall system will be presented.