Online Semisupervised Active Classification for Multiview PolSAR Data

Co-Training Broad Siamese-Like Network for Coupled-View Semi-Supervised Learning

Multiview semi-supervised learning is a popular research area in which people utilize cross-view knowledge to overcome the limitation of labeled data in semi-supervised learning. Existing methods mainly utilize deep neural network, which is relatively time-consuming due to the complex network structure and back propagation iterations. In this article, co-training broad Siamese-like network (Co-BSLN) is proposed for coupled-view semi-supervised classification. Co-BSLN learns knowledge from two-view data and can be used for multiview data with the help of feature concatenation. Different from existing deep learning methods, Co-BSLN utilizes a simple shallow network based on broad learning system (BLS) to simplify the network structure and reduce training time. It replaces back propagation iterations with a direct pseudo inverse calculation to further reduce time consumption. In Co-BSLN, different views of the same instance are considered as positive pairs due to cross-view consistency. Predictions of views in positive pairs are used to guide the training of each other through a direct logit vector mapping. Such a design is fast and effectively utilizes cross-view consistency to improve the accuracy of semi-supervised learning. Evaluation results demonstrate that Co-BSLN is able to improve accuracy and reduce training time on popular datasets.

Online Active Continual Learning for Robotic Lifelong Object Recognition

In real-world applications, robotic systems collect vast amounts of new data from ever-changing environments over time. They need to continually interact and learn new knowledge from the external world to adapt to the environment. Particularly, lifelong object recognition in an online and interactive manner is a crucial and fundamental capability for robotic systems. To meet this realistic demand, in this article, we propose an online active continual learning (OACL) framework for robotic lifelong object recognition, in the scenario of both classes and domains changing with dynamic environments. First, to reduce the labeling cost as much as possible while maximizing the performance, a new online active learning (OAL) strategy is designed by taking both the uncertainty and diversity of samples into account to protect the information volume and distribution of data. In addition, to prevent catastrophic forgetting and reduce memory costs, a novel online continual learning (OCL) algorithm is proposed based on the deep feature semantic augmentation and a new loss-based deep model and replay buffer update, which can mitigate the class imbalance between the old and new classes and alleviate confusion between two similar classes. Moreover, the mistake bound of the proposed method is analyzed in theory. OACL allows robots to select the most representative new samples to query labels and continually learn new objects and new variants of previously learned objects from a nonindependent and identically distributed (i.i.d.) data stream without catastrophic forgetting. Extensive experiments conducted on real lifelong robotic vision datasets demonstrate that our algorithm, even trained with fewer labeled samples and replay exemplars, can achieve state-of-the-art performance on OCL tasks.

A Human-Machine Agent Based on Active Reinforcement Learning for Target Classification in Wargame

To meet the requirements of high accuracy and low cost of target classification in modern warfare, and lay the foundation for target threat assessment, the article proposes a human-machine agent for target classification based on active reinforcement learning (TCARL_H-M), inferring when to introduce human experience guidance for model and how to autonomously classify detected targets into predefined categories with equipment information. To simulate different levels of human guidance, we set up two modes for the model: the easier-to-obtain but low-value-type cues simulated by Mode 1 and the labor-intensive but high-value class labels simulated by Mode 2. In addition, to analyze the respective roles of human experience guidance and machine data learning in target classification tasks, the article proposes a machine-based learner (TCARL_M) with zero human participation and a human-based interventionist with full human guidance (TCARL_H). Finally, based on the simulation data from a wargame, we carried out performance evaluation and application analysis for the proposed models in terms of target prediction and target classification, respectively, and the obtained results demonstrate that TCARL_H-M can not only greatly save labor costs, but achieve more competitive classification accuracy compared with our TCARL_M, TCARL_H, a purely supervised model-long short-term memory network (LSTM), a classic active learning algorithm-Query By Committee (QBC), and the common active learning model-uncertainty sampling (Uncertainty).

Multi-View Feature Selection for PolSAR Image Classification via l₂,₁ Sparsity Regularization and Manifold Regularization

Feature is a crucial element of polarimetric synthetic aperture radar (PolSAR) image classification. Multiple types of Features, such as polarimetric features (PF) generated from the PolSAR data and various polarimetric target decompositions, texture features (TF) of the Pauli color-coded PolSAR images are used as features for PolSAR image classification. The obtained PF and TF often form the high-dimensional data, which leads to high computational complexity. Moreover, some features are irrelative and do nothing to improve the classification performance. Therefore, it is fairly indispensable to select a subset of useful features for PolSAR image classification. This paper proposes a multi-view feature selection method for PolSAR image classification. Firstly, two types of features, PF and TF are generated separately. Then the optimization model is built to pursue the feature selection matrices. Specifically, in order to maintain the consistency of different types of features, we search for the common representation of multiple types of features in the optimization problem. The l_2,1 norm sparsity regularization is imposed on the feature selection matrices to achieve feature selection. In addition, the manifold regularization on the common representation is utilized to preserve the structure information of the data. The effectiveness of the proposed method is evaluated on three real PolSAR data sets. Experimental results demonstrate the superiority of the proposed method.