With the widespread utilization of resting-state functional magnetic resonance imaging(rs-fMRI) technology in exploring brain connectivity, an increasing number of studies have shifted their focus towards accurately constructing brain networks for the purpose of classifying brain diseases. Normalization is recognized as a pivotal step in the establishment of brain connections, serving to mitigate signal amplitude variations. However, current research on its application in brain network construction remains relatively limited. In order to solve this problem, a brain connectivity approach based on subject normalization was proposed, aiming to standardize signal amplitude differences between subjects while preserving amplitude distinctions between brain regions. Furthermore, entropy-based methods for network construction were introduced and evaluates their efficacy in brain disease classification tasks by comparing various cross-entropy methodologies. The objective was to offer a novel analytical perspective distinct from conventional correlation-based approaches in studying brain connectivity. Experimental findings demonstrate that the proposed subject-normalized cross joint entropy(CJE) method achieves notable classification accuracy in schizophrenia(SCZ), mild cognitive impairment(MCI), and autism spectrum disorder(ASD), exhibiting enhancements of approximately 4%, 6%, and 7% respectively compared to traditional normalized correlation-based methods.