Fitting Multiple Alpha Peaks Using Neural Network Techniques

Despite the sophistication of today´s radiochemical separation techniques, it often occurs that the peaks in the spectra of ¿-emitting radioactive samples partially overlap. We here demonstrate the usefulness of a procedure based on a neural network, a multilayer perceptron with backpropagation training method, trained with isolated alpha peaks of environmental samples in resolving such partially overlapping alpha peaks and in predicting the activities of the ¿-emitters detected.