Diatoms are microscopic algae that dynamically adapt their shape and texture in response to various environmental factors, making them one of the most precise bioindicators for water quality assessment. Traditionally, identifying diatom genera in microscopic images has relied on expert biologists with specialized knowledge of their morphometric characteristics. Building an automatic diatom genus recognition system using state-of-the-art Convolutional Neural Networks (CNNs) is challenging for the next reasons: the absence of datasets that allow the development and evolution of automatic diatom recognition systems, the inefficacy of transfer-learning alone from different domains, i.e., the traditional ImageNet, and the resulting unbalance when a new dataset is constructed. This work is three-fold: 1) it provides a new high-quality, publicly available database comprising 44 genus-level diatom classes, 2) designs a robust approach that addresses the aforementioned challenges, and 3) implement a user-friendly interface for seamless diatom identification. All these elements are publicly available through this link ( DiatomNet dataset, 2022).