Abstract:Band selection (BS) is an important technique in remote sensing hyperspectral image (HSI) classification that captures informative bands and removes redundant bands, thereby reducing the computational overhead of the model and improving classification performance. Recently, evolutionary algorithm (EA) with good search capability has been successfully extended as an efficient BS method for processing HSIs. However, most of the existing EA-based BS methods still face following challenges: 1) Most of them are designed based on a single search strategy, which may not be sufficient to find the promising subset of bands and tend to fall into the local optimum; 2) Most of them ignore the problem of potential duplicate bands existing in the solution vectors, which leads to some performance degradation. To address these issues, this paper proposes an effective unsupervised multiobjective BS approach through an adaptive particle swarm optimizer (PSO), called APSO-BS. First, a new robust unsupervised multiobjective BS model is designed, which applies information entropy to measure the amount of information contained in bands and uses the structural similarity measure to evaluate the correlation among bands. Second, to effectively solve the multiobjective BS model constructed above, an adaptive particle swarm optimizer is proposed, which first applies a pre-scoring mechanism to estimate the evolutionary states (i.e., diversity and convergence degree) of particles, and then adaptively selects different PSO search strategies for different particles according to their corresponding evolutionary states. In addition, a self-repair mechanism is designed to correct the violating individuals with duplicate bands. Therefore, our proposed method can reasonably address the two major challenges mentioned above. Finally, to evaluate the performance of APSO-BS, four common HSI datasets are adopted in our experiments, and extensive results under three different classifiers show that our proposed APSO-BS outperforms several state-of-the-art BS methods.

SRL-SOA: Self-Representation Learning with Sparse 1D-Operational Autoencoder for Hyperspectral Image Band Selection

A Band Selection Method with Masked Convolutional Autoencoder for Hyperspectral Image

Deep Reinforcement Learning for Band Selection in Hyperspectral Image Classification

A hyperspectral band selection strategy based on adjacency clustering and local structural correlation

Learning a Deep Ensemble Network with Band Importance for Hyperspectral Object Tracking.

A Cross-Scene Self-Representative Network for Hyperspectral Band Selection

A generic self-supervised learning (SSL) framework for representation learning from spectra-spatial feature of unlabeled remote sensing imagery

Unsupervised Band Selection Using Fused HSI and LiDAR Attention Integrating With Autoencoder

Hyperspectral Image Analysis with Subspace Learning-based One-Class Classification

Deep Latent Spectral Representation Learning-Based Hyperspectral Band Selection for Target Detection

Simulated annealing band selection approach for hyperspectral imagery

Multiobjective band selection approach via an adaptive particle swarm optimizer for remote sensing hyperspectral images

Self-Supervised Deep Multi-Level Representation Learning Fusion-Based Maximum Entropy Subspace Clustering for Hyperspectral Band Selection

Dropout Concrete Autoencoder for Band Selection on HSI Scenes

Structure-Preserved and Weakly Redundant Band Selection for Hyperspectral Imagery

Efficient Graph Convolutional Self-Representation for Band Selection of Hyperspectral Image

Double Deep Q-Network for Hyperspectral Image Band Selection in Land Cover Classification Applications

Hyperspectral Image Classification Based on Deep Auto-Encoder and Hidden Markov Random Field.

Hyperspectral Image Analysis in Single-Modal and Multimodal setting using Deep Learning Techniques

An Automated Machine Learning Framework for Adaptive and Optimized Hyperspectral-Based Land Cover and Land-Use Segmentation

Land use/land cover (LULC) classification using deep-LSTM for hyperspectral images