TC4 is a representative titanium alloy that has been widely used in the manufacture of critical aerospace components. Its mechanical properties can be significantly enhanced through appropriate heat-treatment processes. These enhancements are fundamentally attributed to heat-treatment-induced changes in the microstructural characteristics. Conventional microstructural characterization techniques are capable of directly revealing microstructural morphology and phase constituents. However, these methods are typically destructive and suffer from low inspection efficiency. To enable rapid and in situ identification of different heat-treated microstructural states in titanium alloys, this study exploits the matrix effects inherent to fiber-optic laser-induced breakdown spectroscopy (FO-LIBS) as discriminative features and proposes a LIBS-based spectral classification framework combining principal component analysis (PCA) and support vector machine (SVM). Random pulse-to-pulse fluctuations are mitigated through intra-spot multi-pulse averaging, followed by feature extraction using PCA. On this basis, an SVM classifier is constructed, with model performance evaluated via cross-validation and quantitatively assessed using confusion matrices. The results demonstrate that, compared with pulse-level spectral classification, samples subjected to intra-spot averaging exhibit markedly improved separability in the low-dimensional principal component space. Using the proposed model, classification accuracies of 100% and 99.8% are achieved for titanium alloy samples subjected to four and eight different heat-treatment conditions, respectively. These findings indicate that, when integrated with data-driven modeling strategies, LIBS shows strong potential for the rapid identification of material microstructural states.