Performing collaborative semantic mapping is a critical challenge for cooperative robots to enhance their comprehensive contextual understanding of the surroundings. This article bridges the gap between the advances in collaborative geometry mapping that relies on pure geometry information fusion, and single robot semantic mapping that focuses on integrating continuous raw sensor data. In this article, a novel hierarchical collaborative probabilistic semantic mapping framework is proposed, where the problem is formulated in a distributed setting. The key novelty of this work is the modeling of the hierarchical semantic map fusion framework and its mathematical derivation of its probability decomposition. At the single robot level, the semantic point cloud is obtained by combining information from heterogeneous sensors and used to generate local semantic maps. At the collaborative robots level, local maps are shared among robots for global semantic map fusion. Since the voxel correspondence is unknown between local maps, an expectation-maximization approach is proposed to estimate the hidden data association. Then, Bayesian rule is applied to perform semantic and occupancy probability update. Experimental results on the unmanned aerial vehicle and the unmanned ground vehicle platforms show the high quality of global semantic maps, demonstrating the accuracy and utility in practical missions.

Fusing 3-D maps generated by multiple robots in real/semi-real time distributed mapping systems are addressed in this paper. A 3-D occupancy grid-based approach for mapping is utilized to satisfy the real/semi-real time and distributed operating constraints. This paper proposes a novel hierarchical probabilistic fusion framework, which consists of uncertainty modeling, map matching, transformation evaluation, and map merging. Before the fusion of maps, the map features and their uncertainties are explicitly modeled and integrated. For map matching, a two-level probabilistic map matching (PMM) algorithm is developed to include high-level structural and low-level voxel features. In the PMM, the structural uncertainty is first used to generate a coarse matching between the maps and its result is then used to improve the voxel level map matching, resulting in a more efficient and accurate matching between maps with a larger convergence basin. The relative transformation output from PMM algorithm is then evaluated based on the Mahalanobis distance, and the relative entropy filter is used subsequently to integrate the map dissimilarities more accurately, completing the map fusion process. The proposed approach is evaluated using map data collected from both simulated and real environments, and the results validate the accuracy, efficiency, and the support for larger convergence basin of the proposed 3-D occupancy map fusion framework.

Operating multiple robots in an unstructured environment is challenging due to its high complexity and uncertainty. In such applications, the integration of individual maps generated by heterogeneous sensors is a critical problem, especially the fusion of sparse and dense maps. This paper proposes a general multi-level probabilistic framework to address the integrated map fusion problem, which is independent of sensor type and SLAM algorithm employed. The key novelty of this paper is the mathematical formulation of the overall map fusion problem and the derivation of its probabilistic decomposition. The framework provides a theoretical basis for computing the relative transformations amongst robots and merging probabilistic map information. Since the maps generated by heterogeneous sensors have different physical properties, an expectation-maximization-based map-matching algorithm is proposed which automatically determines the number of voxels to be associated. Then, a time-sequential map merging strategy is developed to generate a globally consistent map. The proposed approach is evaluated in various environments with heterogeneous sensors, which demonstrates its accuracy and versatility in 3-D multirobot map fusion missions.

Recently, deep semi-supervised hashing methods have attracted increasing attention, which can significantly improve retrieval performance by leveraging abundant unlabeled data. These methods usually generate surrogate supervision signals to learn with unlabeled data, such as neighborhood information and augmentation invariant requirements. However, an essential issue of these methods is that the supervised signals are not always reliable, which may damage the performance. In this paper, we propose a novel Uncertainty-Aware and Multi-Granularity Consistent Constrained Semi-Supervised Hashing (UMCSH) method to alleviate the negative effects of noisy supervised signals and enlarge the inter-class distance. Specifically, our UMCSH mainly consists of an Uncertainty-Aware Instance-Level Consistency (UAILC) model and a Cluster-Based Class-Level Consistency (CBCLC) model. UAILC introduces an uncertainty estimation method to select reliable supervised signals to extract discriminative features for each unlabeled data. CBCLC establishes connections between labeled data and unlabeled data by encouraging each unlabeled sample to be close to the hash center (calculated with the labeled data) according to its pseudo-label. Extensive experimental results demonstrate the superior performance of our proposed approach compared with several state-of-the-art semi-supervised hashing methods.

Ship detection in large-scale synthetic aperture radar (SAR) images has essential value in both military and civilian applications. However, due to the complexity of the background and the simplicity of the texture, ship detection in large-scale SAR images is prone to false alarms, such as similar-shaped reefs, islands, sea clutter, and inland buildings. This article proposes a multisize inference fusion framework to eliminate false alarms and improve the overall performance of ship detection in large-scale SAR images. In this framework, a multisize slicer is proposed to expand the scale range of image expression. Then, a detection model library is built to keep various types of models for different task scenarios and requirements. Finally, two subapproaches are proposed for false alarm elimination, namely, pixel feature filtering (FAE-pff) and multisource fusion (FAE-msf), to reduce false detection results in the output of the detection model. FAE-pff calculates how obvious each target is relative to the background and eliminates less obvious results. FAE-msf obtains bounding boxes and corresponding confidences from multiple inference sources and fuses them through weighting and updating them to achieve complementation and enhancement of information. Various experiments were conducted to evaluate the performance of each module qualitatively and quantitatively. It proves the effectiveness of the proposed framework, which can achieve more correct detections while greatly reducing erroneous detections.