An image classification method based on deep learning. The method relates to intelligent decision making, and comprises: constructing a first classification model, training the first classification model to obtain the adjusted first classification model, and inputting a first sample to the adjusted first classification model so as to obtain first feature data (S2); performing compression processing on the structure of the first classification model to obtain a second classification model, sending the first feature data and the second classification model to each second server (2), receiving a second parameter fed back by each second server (2), obtaining the updated first classification model on the basis of the second parameter, and training the updated first classification model to obtain a target classification model; and inputting an image to be classified to the target classification model so as to obtain an image classification result. Further provided are an image classification apparatus, a server and a medium. The model training efficiency is improved, and the image classification accuracy is improved.
展开▼