Abstract: To solve the problem of low recognition accuracy due to high noise and insufficient feature information of digital instrument images, this paper proposes an image recognition method of high noise digital instrument based on convolution recursive neural network combining projection threshold segmentation and number sequence correction. Firstly, the projection threshold segmentation binarization algorithm is proposed to preprocess the image. The vertical projection method is used to divide the image into different regions, and the binarization threshold is set adaptively according to the noise intensity of different regions to binarization the image and reduce the noise. Secondly, according to the changing characteristics of the number rules between images, the number sequence correction algorithm is used to transform a single number recognition into a number sequence recognition, and the recognition result is obtained by comparing the recognition probability of different number sequences, so as to solve the problem of low recognition accuracy caused by insufficient feature information of a single image. The experimental results show that, compared with the convolutional recursive neural network model, the accuracy of the high-noise digital instrument recognition model proposed in this paper is improved on the high-noise data set by about 61.95%, reaching 93.58%.