Purpose <p>Fine motor intention recognition technology can enhance rehabilitation outcomes in stroke patients. However, the limited spatial resolution of electroencephalogram (EEG) signals and the close proximity of somatosensory motor cortex regions make it challenging for brain–computer interface systems to distinguish between different fine motor intentions. Multi-parameter feature modulation technology could help to address these critical issues.</p> Methods <p>Speed and visual auxiliary stimulation feature modulation were integrated for fine motor intentions (elbow flexion/extension and hand grasping) of adjacent upper limb joints. First, we analyzed fine motor intention features under different experimental conditions using event-related spectral perturbation, brain topography, and <i>R</i>² separability methods; then, we combined the filter bank common spatial pattern (FBCSP) algorithm and minimum distance to Riemannian mean (MDRM) classifier for fine motor intention recognition and compared the performance of this model with that of traditional support vector machine (SVM) classification.</p> Results <p>Both dynamic visual auxiliary stimulation and fast speed parameters significantly modulated EEG features (<i>p</i> &lt; 0.05), and the combination of multiple parameters produced a coupled modulation effect, which significantly enhanced the separability of event-related desynchronization features. Under multi-parameter experimental conditions, the FBCSP + MDRM model achieved an average classification accuracy of 89.82%±3.90%, representing improvements of 18.22% and 9.84% compared with the baseline and single-parameter conditions, respectively, and outperforming FBCSP + SVM by 15.2% (<i>p</i> &lt; 0.01).</p> Conclusion <p>Multi-parameter feature modulation technology can enhance the strength of EEG features of fine motor intentions of adjacent upper limb joints and improve the separability of intentions. Our FBCSP + MDRM model significantly increased the accuracy of fine motor intention recognition.</p>