The Artificial Neural Network And Its Application In X Fluorescence Archaeology

In order to overcome problems arisen from the application of X fluorescence analysis into complex spectrum produced by archaeological ceramic fragments with multi-element, low content and thick ground, we have employed the artificial neural network into the research of X fluorescence archaeology and conducted three kinds of research works. As the first one, we have applied the linear OLAM network (Optimal Linear Association Memory Network) and the non-linear BP network(Back-Propagation Network) respectively to analyze the complex X fluorescence spectrum of archaeological samples, and taken both results of spectrum analysis to compare with each other. The second, the method of pattern recognition of BP network was tentatively used to perform intelligent identification of production places of these archaeological samples. The third, we have set up three pretreatment methods for data of spectrum before putting them into network learning to improve the reliability of network operation and to reduce the effect of substrate and the influence of electronic noise.During this research work, we have respectively studied the OLAM network and the BP network on their basic theories, arithmetic, learning process, learning samples and outcome of spectrum analysis. By improving BP network arithmetic, it is found that the structure of the network is optimized and that the convergent speed of the network is quickened. To improving the single pattern learning, a new method of whole pattern learning was brought forward in this thesis to get better results of spectrum analysis. For these tow networks, it has been taken to comparewith their learning precious, as well as their outcomes and abilities of spectrum analysis under the condition of different arithmetic and different spectrum data pretreatment. To identify the production places with BP network, we have set up a theory of identification of production places and the pattern characteristic of each production place. From the outcome, it has been proved that much better results for the identification of production places were obtained. For example, the rate of the identification of production places was reached 100 percent as production places were well classified, and the rate was also above 60 percent as production places were fuzzily classified. It was proved as well that as for the results of both spectrum analysis and identification of production places the pretreated one is much better than those without pretreatment.All our experiment works proved that in science and technology archaeology the satisfactory results are gained for the outcome of spectrum analysis and the identification of production places as well as pretreatment methods of spectra data by using the artificial neural network and X fluorescence analysis. All of them show that the artificial neural network has powerful value and widely applied future in science and technology archaeology.Based on the platform of Windows 98 operation system and Visual C++ 6. 0 programming language, we have developed a series of visual software, including X fluorescence spectrum analysis and identification of production places, which are powerful and have friendly and visual interfaces and functions easily operated.