Automatic Generation of Pantograph Image Caption Based on Deep Learning

As the essential component of the pantograph-catenary system, the pantograph operates in a complex environment, and its failure can result in safety accidents during train operations. Currently, there is insufficient research on the automatic generation of pantograph maintenance records. The conventional approach involves manual recording during the maintenance process and manual compilation of maintenance reports, which suffer from issues like human error and heavy reliance on experience. Consequently, this paper proposes the PantoCap model for pantograph image caption. The model uses DenseNet-121, replacing VGG-16, to enhance feature extraction capabilities. Additionally, a relational memory module is devised to preserve the association information among texts. At the same time, the layer normalization module of Transformer decoder is improved, and the calculation method of memory mask is added to improve the processing of memory information during model decoding. Finally, the effectiveness of the proposed method is verified on the pantograph image caption dataset, and caption of the pantograph image is generated.