To improve the accuracy of absolute grating ruler positioning and decoding, a positioning and decoding method based on end-to-end deep learning framework is proposed. Attention modules were integrated to improve the positioning of symbol edges of UNet++, and a channel edge information extraction network was designed to achieve regression prediction of channel images to position information. To reduce cumulative errors, a loss function was designed based on the characteristics of absolute grating ruler images, which integrates the symbol edge-positioning network and the code channel edge information extraction network into an end-to-end network framework, thereby constructing a code channel-positioning module. A pseudo-random code decoding method was designed based on the center pixel of the code path to achieve absolute grating size path decoding. The experimental results demonstrate that the proposed positioning decoding method can improve the measurement accuracy of absolute grating rulers, within a 95% confidence interval (-0.206, 0.243) μm, the root mean square error is 0.265 μm, superior to existing absolute grating ruler positioning and decoding methods.