This study presents an efficient technique for real time cognitive load monitoring utilizing multimodal peripheral biosignals. Present cognitive assessment methods lack instantaneous capability or require sizable computationally demanding data inputs. In order to address these issues, the researchers developed an innovative engineering pipeline that functions by converting short physiological signal segments into image-like representations using Graminan Angular Difference Fields and Motif Difference Fields. Causal interdependencies were captured through forward-backward copula Granger causality analysis. These features were then processed by a light capsule neural network equipped with self attention routing which enabled the accurate decoding of cognitive load from brief signal windows. The framework was tested on the benchmark datasets of WESAD and CLAS and achieved strong performance as well as computational efficiency,