Objective: To improve the classification accuracy of strawberries. Methods: A method of strawberry classification based on improved CNN was proposed by improving CNN through mixing pool method. Firstly, through the combination of maximum pooling and average pooling techniques, a hybrid pooling method was obtained. Then, the hybrid pool method was used to improve the generalization ability of CNN model. After that, image data acquisition, image preprocessing and image feature extraction were carried out. Finally, sensitivity, specificity, accuracy, recall rate and F₁ score were used to evaluate the effectiveness of the trained strawberry classification method. Results: The sensitivity, specificity, accuracy, recall rate and F₁ score of the proposed method for strawberry classification in 16 pixel×16 pixel images reached 0.993, 0.993, 0.994, 0.992 and 0.991, respectively. Compared with the other five classification methods, the sensitivity, specificity, accuracy, recall rate and F₁ score of the proposed method were improved by 3.44%, 3.96%, 4.26%, 3.92% and 4.08%, respectively. Conclusion: This method can achieve accurate classification of strawberries with different maturity, and is expected to provide technical support for the research and development of high-performance strawberry packaging robots and supermarket fruit automatic recognition machines.