The invention belongs to the technical field of laser probe component analysis, and discloses a method for improving classification accuracy of laser probe by utilizing image characteristics. The method comprises the following steps: S1, collecting plasma spectrum of a sample by adopting a spectrum collecting device, wherein the plasma spectrum is a multi-dimensional spectrum; S2, performing imageprocessing on the plasma spectrum to obtain a 16-bit gray scale image; S3, extracting image features based on the gray image, wherein the image features are multi-dimensional spectral features; S4, training the image features as an input of a classification algorithm to obtain a classification model based on the image features; S5, inputting the image features of the sample to be classified to the classification model, and outputting the classification result by the classification model, thereby completing the classification. The invention improves the splitting precision and does not need manual or automatic route selection, simplifies the classification process and improves the classification efficiency.