In order to improve the performance of an electrooculogram ( EOG) -based human activity recognition ( HAR) system and increase the correct recognition ratio of multi-class saccade signals,a sample optimization algorithm is proposed on the basis of independent component analysis ( ICA) .In the algorithm,by taking a single saccade data as the object,an automatic selection method of saccade related independent components ( SRICs) is designed according to the independent components ( ICs) -to-electrode mapping mode,and a corresponding ICA spatial filter is established.Then,noisy saccade samples are deleted on the basis of the correct recognition ratio of saccade signals after the linear projection of original EOGs.In the lab environment,the ICA spatial filter is utilized to classify four types of saccade signals by“run-to-run test”and“session-to session test.The results show that,in the two tests,the correct recognition ratios of the data optimized by the proposed algorithm are respectively 99.57% and 98.82%,and they are 0.57% and 0.83% higher than those of original EOG signals,which means that the proposed algorithm can effectively optimize saccade signals and thus improve the correct recognition ratio.