The Application Of Probabilistic Neural Network In The Amur Tiger Fur's Texture Identification

The Amur tiger is one of the rare and endangered species in the world which belongs to the national level protected wild animals. The research on the Amur tiger's individual identification is of great benefit to the survey and research on the rare and endangered species. Images of one hundred Amur tigers were collected in Heilongjiang Amur Tiger Park. Each left and right side of Amur tiger was taken for ten images. There are two thousands images of Amur tigers were as experimental subjects for research.With the application of conventional methods of image processing and the methods of dilation and erosion, opening and closing, and the Hit-or-Miss transformation in mathematical morphology, the Amur tiger fur's texture images were processed, after the processing, the Amur tiger fur's texture images were clearly and conveniently for the tiger fur's texture's extraction. A standpoint was proposed that the processing method to the Amur tiger fur's texture images is not unique, different images should choose fitting processing method based on the shooting conditions in order to get the best processing result.The area of the black texture whose shape is like diamond was used as the eigenvalues. The rule of eigenvalues'extraction was from head to tail and from top to bottom in the Amur tiger fur's texture image. Three eigenvalues were extracted from each side of the Amur tiger images based on the extraction rule. The groups which contain six eigenvalues were used to identify the Amur tiger individual. The six eigenvalues were first extract from the right side and then the left side. The software of Image J was used to extract the eigenvalues from the images of eighty-nine Amur tigers.A concept of standard area was proposed which was a triangle area and its vertices are the spine's highest point, abdomen's lowest point and tail root's central point. The ratio of the eigenvalues we extracted and the standard area were the last eigenvalues. After that the eigenvalues were unique to each Amur tiger.The probabilistic neural network was established with the function of newpnn and it was used to identify the Amur tiger individual. The network parameter of SPREAD was fixed by trying. Training the network use the input samples of the eigenvalues from fifty Amur tigers we measured and the output samples of the types of the Amur tigers. The testing samples of the other fifty eigenvalues were used to test the network and after that carried out the simulation experiments. It is got by analysis to the results that the probabilistic neural network can identify the Amur Tiger fur's texture and identification is at a high speed and accuracy rate.