Particle Swarm Optimization (PSO) is a popular and bionic algorithm based on the social behavior associated with bird flocking for optimization problems. To maintain the diversity of swarms, a few studies of multi-swarm strategy have been reported. However, the competition among swarms, reservation or destruction of a swarm, has not been considered further. In this paper, we formulate four rules by introducing the mechanism for survival of the fittest, which simulates the competition among the swarms. Based on the mechanism, we design a modified Multi-Swarm PSO (MSPSO) to solve discrete problems, which consists of a number of sub-swarms and a multi-swarm scheduler that can monitor and control each sub-swarm using the rules. To further settle the feature selection problems, we propose an Improved Feature Selection (IFS) method by integrating MSPSO, Support Vector Machines (SVM) with F-score method. The IFS method aims to achieve higher generalization capability through performing kernel parameter optimization and feature selection simultaneously. The performance of the proposed method is compared with that of the standard PSO based, Genetic Algorithm (GA) based and the grid search based methods on 10 benchmark datasets, taken from UCI machine learning and StatLog databases. The numerical results and statistical analysis show that the proposed IFS method performs significantly better than the other three methods in terms of prediction accuracy with smaller subset of features.

This paper proposes a novel multi-sensor-based indoor global localization system integrating visual localization aided by CNN-based image retrieval with a probabilistic localization approach. The global localization system consists of three parts: coarse place recognition, fine localization and re-localization from kidnapping. Coarse place recognition exploits a monocular camera to realize the initial localization based on image retrieval, in which off-the-shelf features extracted from a pre-trained Convolutional Neural Network (CNN) are adopted to determine the candidate locations of the robot. In the fine localization, a laser range finder is equipped to estimate the accurate pose of a mobile robot by means of an adaptive Monte Carlo localization, in which the candidate locations obtained by image retrieval are considered as seeds for initial random sampling. Additionally, to address the problem of robot kidnapping, we present a closed-loop localization mechanism to monitor the state of the robot in real time and make adaptive adjustments when the robot is kidnapped. The closed-loop mechanism effectively exploits the correlation of image sequences to realize the re-localization based on Long-Short Term Memory (LSTM) network. Extensive experiments were conducted and the results indicate that the proposed method not only exhibits great improvement on accuracy and speed, but also can recover from localization failures compared to two conventional localization methods.

The launch of the Chinese Gaofen-3 (GF-3) satellite will provide enough synthetic aperture radar (SAR) images with different imaging modes for land cover classification and other potential usages in the next few years. This paper aims to propose an efficient and practical classification framework for a GF-3 polarimetric SAR (PolSAR) image. The proposed classification framework consists of four simple parts including polarimetric feature extraction and stacking, the initial classification via XGBoost, superpixels generation by statistical region merging (SRM) based on Pauli RGB image, and a post-processing step to determine the label of a superpixel by modified majority voting. Fast initial classification via XGBoost and the incorporation of spatial information via a post-processing step through superpixel-based modified majority voting would potentially make the method efficient in practical use. Preliminary experimental results on real GF-3 PolSAR images and the AIRSAR Flevoland data set validate the efficacy and efficiency of the proposed classification framework. The results demonstrate that the quality of GF-3 PolSAR data is adequate enough for classification purpose. The results also show that the incorporation of spatial information is important for overall performance improvement.

Convolutional neural networks (CNN) have achieved great success in the optical image processing field. Because of the excellent performance of CNN, more and more methods based on CNN are applied to polarimetric synthetic aperture radar (PolSAR) image classification. Most CNN-based PolSAR image classification methods can only classify one pixel each time. Because all the pixels of a PolSAR image are classified independently, the inherent interrelation of different land covers is ignored. We use a fixed-feature-size CNN (FFS-CNN) to classify all pixels in a patch simultaneously. The proposed method has several advantages. First, FFS-CNN can classify all the pixels in a small patch simultaneously. When classifying a whole PolSAR image, it is faster than common CNNs. Second, FFS-CNN is trained to learn the interrelation of different land covers in a patch, so it can use the interrelation of land covers to improve the classification results. The experiments of FFS-CNN are evaluated on a Chinese Gaofen-3 PolSAR image and other two real PolSAR images. Experiment results show that FFS-CNN is comparable with the state-of-the-art PolSAR image classification methods.

No abstract