<p>Methods have been developed to identify parameters in structural health monitoring in order to achieve a better description of a structure. These studies generally focus on a specific hysteretic model and rely on prior knowledge of the system’s behavior, such as the existence of pinching. However, in practice, neither the hysteretic model applicable nor behavioral characteristics, such as pinching, are known a priori. Therefore, a prior step of parameter identification, based on a recognition of the system from the measurements, is required. In this study, a convolutional neural network (CNN) is introduced to predict two essential pieces of information: The best-fitting hysteretic model and the degree of pinching, classified as severe, moderate, soft, and non-pinching, according to the reduction in seismic energy dissipation capacity. The prediction is based on the features of the special loops, which meet some requirements to be used as input in the CNN. Additionally, a second CNN model is proposed to estimate the yield displacement, a necessary value for extracting such special loops. The model is verified with structural and earthquakes different from those used in training, achieving an average accuracy of 97.5% for predicting the hysteretic model and 94.9% for predicting the degree of pinching. The model is also verified using experimental data from three different tests: A steel specimen, a reinforced concrete column, and a 20-story reinforced concrete building. The results indicate that the model is capable of extracting hysteresis features appropriately, making it a useful tool for recognizing systems prior to parameter identification.</p>