Abstract: Acoustic method can detect stored grain insects quickly and plays a very important role in preventing grain insects early.Four kinds of active acoustic signals of two varieties of grain insects were studied.Using isometric feature mapping method of manifold learning,dimensions were reduced and manifold characteristics of acoustic signals were extracted.Then optimal separate hyperplane was trained by SVM with heavy tailed radial basis function as kernel function.It can test and recognize manifold characteristics.The result demonstratesd that isometric feature mapping method can extract manifold characteristics of acoustic signals of two kinds of grain insects.SVM classifies and recognizes manifold characteristics well.