The invention discloses a semen protein prediction method based on a convolutional neural network, and belongs to the technical field of big data and artificial intelligence. According to the method,a protein list verified by biological experiments in semen of existing literatures and databases is used as a positive sample for model training; protein family information corresponding to the positive sample is deleted from a Pfam protein family information database, protein families with the number of proteins exceeding 5 in the families are searched for in the remaining protein family information database, and five pieces of protein information are randomly selected from the protein families to serve as negative samples for model training. The method also comprises the steps of dividing the positive sample data and the negative sample data into a training set, a verification set and a test set; carrying out feature selection on protein features, building a model, training the model byusing the training set, carrying out parameter adjustment on the verification set, and carrying out performance evaluation on the test set. The input is a protein feature, and the output is a prediction result, so that the semen prediction accuracy is improved, and finally, semen protein prediction is realized.