Blood cell morphological analysis plays a vital role in clinical diagnosis, especially in the early detection of leukemia, anemia and other blood system diseases. Conventional image processing techniques are difficult to deal with complex situations such as cell overlap and uneven staining, and basic machine learning methods also have obvious limitations in extracting complex morphological features. Deep learning has shown excellent performance in the field of medical image classification and provides a new technical approach for automated analysis of blood cells. This study aims to develop an efficient and accurate blood cell classification model to assist in the early diagnosis of blood diseases and cancer. By adopting the InceptionV3 network structure and combining the 'Grid Search Enhanced with Coordinate Ascent' hyperparameter optimization method, the study provides a systematic automated classification model training method for blood cell multi-classification tasks. The experiment was based on a dataset containing five types of cells. The results showed that the final model achieved an accuracy of 99.20% on the test set, the AUC of all classes reached 1.00, and the average specificity was as high as 99.80%, providing a reliable technical reference for clinical blood pathology analysis and early cancer screening.