Abstract: Efficient financial analysis is essential for extracting insights from financial statements and supporting decision-making. To address the challenge of interpreting complex financial data, we propose an intelligent visual analysis method combining DuPont analysis and large language models. Our approach extracts key financial metrics, decomposes them via DuPont analysis, and leverages LLMs to generate insights and visualizations, helping users identify trends and anomalies. We implement a prototype system, FinDecipher, and evaluate it on real-world financial data. Results show that FinDecipher improves financial comprehension and decision accuracy, demonstrating its practical feasibility and potential in fintech applications.