The paper introduces an Entropy-based Text Watermarking Detection (EWD) method to improve the detection of machine-generated texts, particularly in low-entropy scenarios. Traditional text watermarking algorithms often struggle with low-entropy texts, where the watermark's influence is less detectable. EWD addresses this by customizing the weight of each token during detection based on its entropy, giving higher-entropy tokens more influence. This approach ensures that the detection process is more accurate and robust, especially in low-entropy scenarios. The method is training-free and fully automated, making it efficient and versatile. Experiments demonstrate that EWD outperforms existing methods in low-entropy scenarios and maintains similar performance in high-entropy scenarios. The paper also provides theoretical analysis and empirical evaluations to support the effectiveness of EWD.The paper introduces an Entropy-based Text Watermarking Detection (EWD) method to improve the detection of machine-generated texts, particularly in low-entropy scenarios. Traditional text watermarking algorithms often struggle with low-entropy texts, where the watermark's influence is less detectable. EWD addresses this by customizing the weight of each token during detection based on its entropy, giving higher-entropy tokens more influence. This approach ensures that the detection process is more accurate and robust, especially in low-entropy scenarios. The method is training-free and fully automated, making it efficient and versatile. Experiments demonstrate that EWD outperforms existing methods in low-entropy scenarios and maintains similar performance in high-entropy scenarios. The paper also provides theoretical analysis and empirical evaluations to support the effectiveness of EWD.