Music is a product of human conscious activity, which, as a special form of artistic expression, can directly hit the psyche and trigger people’s strong emotional experiences. In this study, the continuity of pitch significance is first utilized to represent the musical melody and the feature extraction of polyphonic musical melody is carried out based on harmonic peak and harmonic sum functions. Based on this basis, the features of the extracted musical melody are recognized by combining convolutional neural networks. In addition, the study also constructs an emotion evocation model based on musical melody, and empirically demonstrates the relationship between musical melody and emotion evocation using statistical analysis. The p-values of different music melodies and different music preferences are all less than 0.05. According to the difference analysis, sad emotions have an arousal rate of 90%, which makes them the easiest to induce and arouse. The p-value for the three factors between musical melody, gender, and professional background was less than 0.05, and the interaction was significant. Music melody, gender, and professional background of emotion evocation have a considerable difference. The influence of emotion evocation has a significant effect. While the role of music preferences on emotion evocation is not substantial, the trend and characteristics of different music melody types evoked emotions provide an effective and realistic basis.