An emotion EEG recognition method providing emotion recognition model time robustness, comprising: performing pre-processing on a collected 64-lead EEG signal comprising changing a reference to a binaural average, downsampling to 500 Hz, performing 1-100 Hz bandpass filtering, and using an independent component analysis algorithm to remove EOG interference finding an optimal discriminative frequency component in a pre-processed EEG signal by means of adaptive tracking of discriminative frequency components, and calculating a power spectral density of the optimal discriminative frequency component on each lead, respectively, forming an emotion characteristic matrix using principal component analysis to perform dimension reduction on the characteristic matrix using a support vector machine classifier to perform recognition on the dimension-reduced EEG power spectrum characteristics, establishing an emotion recognition model. The described solution finds an optimal discriminative frequency component by means of adaptive tracking of discriminative frequency components, strengthens emotion correlation characteristics by means of increasing training set sample days in an emotion recognition model, weakens a time specificity characteristic, and increases time robustness of an emotion recognition model.La présente invention concerne un procédé de reconnaissance démotion par EEG fournissant une robustesse temporelle de modèle de reconnaissance démotion, comprenant : la réalisation dun prétraitement sur un signal dEEG à 64 dérivations collecté comprenant le changement dune référence en une moyenne binaurale, un sous-échantillonnage à 500 Hz, la réalisation dun filtrage passe-bande de 1 à 100 Hz, et lutilisation dun algorithme danalyse de composante indépendant en vue déliminer linterférence EOG la découverte dune composante de fréquence discriminante optimale dans un signal dEEG prétraité au moyen dun suivi adaptatif de composantes de fréquence discriminantes, et le calcul d
