In this paper, we present a continuous volcano seismic analysis performed in two steps: 1) detection and 2) classification. Detection corresponds to the segmentation of a seismic event extracted from the raw signal and classification corresponds to the event labelling, according to the physical phenomena that generated it. Here the classification is performed using two methods: first the events are represented using their spectrograms and classified by a VGG neural network and second, the events are represented using the Qerlet representation and classified using the coefficient obtained by this representation. Detection is applied before classification, and it is adjusted using an empirical criterion that evaluates the number of false positive and false negative events, in a test dataset, as a function of the amplitude threshold. Higher values of the threshold improves the system specificity, while lower values give higher sensibility. This trade-off determines the performance of the systems when the automatic continuous classification is compared with the expert s classification. The seismic signals are obtained from the Nevados del Chillan Volcano located in the south of Chile. In this paper we optimally choose the amplitude threshold for continuous event detection that best fits with expert classification.