The STORM European research project aims in developing an integrated set of technological and organizational resources for monitoring and safeguarding cultural heritage sites. In this context, wireless acoustic sensor networks that detect and record sound samples are deployed over cultural sites. The features of the recorded samples are calculated and forwarded to a feed-forward neural network for classification in order to extract information about the potential hazardous effect of the activity or activities that generated the sounds. Implementation details about the classifier and the sound features are discussed in this paper. Sound features are grouped with respect to their scalar or time-frequency nature and results are presented regarding the classification accuracy of each subset. Results are also presented regarding the accuracy of the proposed approach when operated in noisy environments.