Hybrid NGO–PSO optimized random forest combined with multi-energy LIBS for enhanced accurate classification of tea

Abstract

Accurately measuring and analyzing the chemical composition and spectral characteristics of tea is of great significance for improving the sorting accuracy and preventing tea adulteration and variety misjudgment. We propose a tea identification and classification system that combines multiple settings of laser-induced breakdown spectroscopy (LIBS) and a Random Forest model that combines the Northern Eagle and Particle Swarm Optimization algorithm (NGO–PSO–RF), which provides a new idea for efficient and accurate classification and recognition of tea. The experiment increased the spectral difference between similar samples by measuring the spectral data of eight tea samples at three different energies. Principal Component Analysis (PCA) was performed on the spectral data at each energy, and the first three principal components were extracted to construct the fusion feature set. PCA dimension reduction was used again to obtain the core feature vector. The original complex multi-energy spectral data were simplified by performing two analyses. Finally, the fusion spectral data were analyzed using the NGO–PSO–RF model, and the classification accuracy of the test set was 99.22%, which was 3.13 percentage points higher than the average classification accuracy (96.09%) at a single energy, and the core indicators, such as the recall rate (+3.11%) and F₁ value (+3.19%), were synchronously optimized. In addition, compared with LSTM, SVM, RF and PSO–RF, the classification accuracy of the NGO–PSO–RF model improved by 6.77%, 5.47%, 5.34% and 3.52% respectively. This method provides an innovative and efficient method with high universality and robustness for material classification and provides important technical support and reliable detection prospects for food production safety and chemical analysis.