Unsupervised Wood Species Identification Based on Multiobjective Optimal Clustering and Feature Fusion

This paper proposes an unsupervised wood species identification approach utilizing multiobjective optimization clustering and feature fusion. To address the limitations of single-band spectra in comprehensively capturing wood characteristics, we integrated preprocessed low-dimensional terahertz (THz) and hyperspectral data. Additionally, to tackle the issue of determining the optimal k-value in clustering, we developed an unsupervised wood clustering algorithm that employs multiobjective optimization and evolutionary algorithms. This algorithm utilized a prototype coding method for initialization, density peak clustering, and an improved firefly optimization algorithm for cross-variation to ensure population diversity. Furthermore, a selection operator was designed based on grid division and fast, non-dominated sorting. We evaluated the model's performance using a dataset consisting of hyperspectral and THz spectra from 400 samples representing ten wood species, comprising five coniferous and five broadleaf species. Experimental results demonstrated a 3.5% enhancement in clustering purity with fused data compared to individual datasets. Our proposed algorithm outperformed comparative methods such as DBSCAN, OPTICS, and peak density clustering, achieving a maximum clustering purity of 91.25% in both the internal and external clustering metrics.