LVQ Neural Networks Species Identification Based On Echolocation Calls Of Bat

Different foraging habitats are chosen by different species bat in different terrain. So it is very important to develop protect work effectively for bat after knowing foraging habitats of different species bat. But there are some difficulties in catch and species distinguish because of the particularity of bat moving-behavior. At present, some people distinguish bat in different species main through observing modality characters of bat. So it is more difficulty to distinguish some bats having similar shape characters, and the distinguish result relies on the experience and ability of student.Study find that although the echolocation calls of bat constantly changes along with the variety of its emitting calls the collectivity trend is changeless. The echolocation calls of individual bat in each species can change evidently according to different circumstance and purpose. But the echolocation calls of each species bat is a sort of basic form of echolocation calls. Therefore, prodigious diversity exists among different species. So that we can classify to different species bat basis on the diversity of echolocation calls and neural network. In this way, not only some man-made factors can be reduced but also the species distinguish work of bat can be systematization,theorization and autoimmunization. At the same time, it can lay a foundation for setting up acoustic database of bat.LVQ neural network is a studying network which trains competition layers under having teacher state, having simple network structure. LVQ neural network is better than BP neural network in validity and stability, and found that the training velocity is quicker than BP neural network under the same input dimensionality at least.In this paper, the energy values extracted from echo orientation call signal of Rhinolophus lepidus, Rhinolophus rouxi, Rhinolophus ferrumequinum and Rhinolophus affinis in fly through three-layer wavelet packet transform are used as the feature parameter, then selecting parameter as eigenvector which input LVQ neural network to classify and compare with the result of BP neural network which be done by Zhang Xinna.In order to embody biology meaning fully in this paper, and combining with the characteristic which the input vector may not be attributed. Some essential feature parameters(main frequency, sound pulse width, sound pulse time interval and duty cycle) of echolocation calls of eight species bat as feature vector are input LVQ neural network to classify, and the result is much ideal.