The embodiment of the invention provides an image target classification method based on weighted column sampling XGBoost. The method comprises a step of using a convolutional neural network which is trained in advance in a large data set ILSVRC and is subjected to fine adjustment in a PASCALVOC 2012 data set to extract a target image characteristic, a step of connecting characteristics learned at multiple layers so as to obtain content information which determines an image classification, a step of using an XGBoost method based on weighted column sampling to classify the image characteristic, carrying out secondary sampling on an attribute before constructing a decision tree according to the important degree of the attribute, using attributes of more extracted information for the construction of the current decision tree, carrying out repeated iteration until convergence, and obtaining an image classification model with best performance. According to the technical scheme provided by the embodiment of the invention, when the attribute dimension of data is large and the redundancy is high, the method can be extended to other classification methods using column sampling, and the average accuracy of image target classification is improved.