The detection performance on the sintering state of the rotary kiln is mostly dependent on the features used in the recognition process. So an optimization approach of sintering feature parameters based on Fuzzy Support Vector Machines (SVM) is proposed. This method firstly uses many feature parameters to describe an image, and then reduce some useless features by Portfolio optimization algorithm which is mainly based on Relief theory. Secondly Fuzzy SVM technology is used for state recognition according to the effective retention features. Each feature is defined as a fuzzy degree, and the sintering state is got ultimately. The experiments show that this approach has strong robustness, high accuracy, and good feasibility.