Recognition of fish species is of great importance to marine biology and aquaculture. Ubiquitous devices represent an efficient solution for the preservation of species through the monitoring of fish in the risk of extinction. This approach is essential for endangered population assessment as well as for ecosystem preservation. Several methods have been assessed in these devices to solve the complex task of identifying at-risk of overfishing. As an alternative, the convolutional neural networks (CNNs) represent an accurate method for pattern recognition. Hence, this paper proposes the performance evaluation of a fish recognition model based on CNNs in ubiquitous devices, focusing on the preservation of these species principally during the closed period.