Wearable sensors achieved vast popularity in recent years due to the low cost achieved for their production, the miniaturization and improved reliability of their components, and the possibility to generate a large quantity of data containing valuable information. In industrial environments, such solutions are not widely adopted yet, due to the continuous, noisy, and unlabelled data streams that require extensive pre-processing. This paper focuses on task classification in a mixed-models assembly line with walking workers, in order to build clean and labelled datasets that can be used for several purposes. In particular, this paper evaluates the benefits of using data collected through a Radio Frequency Identification (RFID) glove that allows the detection of the tool that the operators are using during the assembling process. To this aim, a sequence classification approach has been chosen to train two Deep Learning models using different feature sets. The features are extracted from data collected via an Inertial Measurements Unit (IMU) placed on the wrist of the dominant arm for movement orientation and the RFID glove. A Long Short-Term Memory (LSTM) and a Temporal Convolutional Neural Network (TCN) are trained, validated and tested on four complete assembling processes. Results show that the LSTM architecture is able to achieve the best classification accuracy, and that the RFID data positively contribute to the assembly task classification.