Automatic extraction of non-functional requirements plays a crucial role in improving efficiency of requirements elicitation, change management, and validation testing. In recent years, machine learning methods have been widely applied in the field of requirements engineering. Although these methods have achieved some promising results, many of them have been evaluated only on small-scale and relatively balanced datasets of non-functional requirements, which may not reflect their typical characteristics in real-world applications. Therefore, in this paper, we propose a novel deep neural network model called NFRNet-LT to address the challenges of extracting non-functional requirements from various types of documents, by considering the higher granularity of non-functional requirement categories and the imbalanced long-tail distribution of data.