With the development of intelligent equipment and systems, pattern recognition technology has also been widely applied in these fields. Aiming at the complex pattern automatic recognition problem and in view of the excellent learning performance of support vector machine (SVM) in small sample situations, a multiclass classification method based on fuzzy least squares support vector machine (LSSVM) optimized by adaptive genetic algorithm (AGA) is proposed. Firstly, a robust LSSVM classifier model capable of recognizing unknown samples is constructed, it is based on binary tree two-classifiers and has fast learning and classification speed while meeting classification accuracy requirement. Secondly, an improved AGA is used to automatically optimize the parameters of the LSSVM classifier model for improving scientificity of parameter selection and modeling efficiency. Finally, the pattern recognition effect of the classifier is validated by the test samples in complex patterns. The recognition results and classification performance indicate that this proposed method can accurately recognize different complex fault modes, which also proves the effectiveness of this proposed method.